feat: checkpoint Go2 FL simulation performance v0.4.2

Include callable skill and live-GUI prerequisites, epoch sensor reads, packed snapshots, sampled bit-exact kinematics, and verified-tick rendering without changing the physical/control schedule.

Record bounded analytic S1 GUI/headless qualification, original failures and CPU evidence (1557 passed, 6 skipped). Preserve accepted_skill=false and sub-real-time limitations. Exclude native authorities, caches, large media and local delivery copies; retain sealed artifact digest.
This commit is contained in:
2026-10-16 15:46:01 +08:00
parent 9fe9d27865
commit c528dd268f
136 changed files with 15553 additions and 15 deletions
+7 -2
View File
@@ -4,10 +4,15 @@
## 当前版本进度
- **0.4.2性能与限定S1已实测**:[完整测量/限制](docs/contact_cycle_sim_performance_v042_result.md)。真实Kit窗口截图33.98FPS,全S1绘制提交35.30次/秒;GPU整卡均值39.74%/133.43W。相同profiler/单周期下应用循环吞吐1.498×,不删逐步检查或500Hz轨迹。新GUI/headless S1各27000tick,70张量及整个轨迹文件与0.4.1逐字节一致。**GUI物理推进仅0.07060×,headless0.18325×,并非实时或GPU100%**;原严格RootXY失败/accepted_skill=false不改。独立包`performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2/`,图/原速视频在其`demo/`;全部新原件与最终封装凭据在当前仓库`performance_v042/`,旧包原样保留;当前源码未重验完整矩阵/u256或新绘制节拍下的原生故障注入。
- **0.4.1真实Kit GUI已验证**:独立包`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.1/`中`bash demo_gui.sh`一条启动真实解析S1;GUI/headless各六周期通过,提前关窗2040tick按预期中止并清理,不算完整成功。右侧实时HUD、原3kg负载/足端/地面、最终原场景render-only保留;真实截图/240秒原录和35秒原速跳切见包demo。[实际结果/限制](docs/contact_cycle_sim_gui_v041_result.md)。旧0.4.0分版原样保留,21Kit/83040实际调用、0训练,GUI约2–3FPS、不保证墙钟1×实时。未重验新版本完整矩阵/u256,不扩控制域或优化RootXY。
- **先前已本地交付Go2 FL接触周期仿真功能版0.4.0**:可调用Python/CLI;解析20连续周期、三条六周期切换、u256同S1、安全停止/非法拒绝及独立目录真实复现全部通过,0训练。[实际通过表与产物](docs/contact_cycle_sim_skill_v4_result.md) / [API](docs/go2_fl_skill_v4_api.md)。包在`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.0/`,`bash demo.sh`启动、`bash verify_functional.sh`真实验证;带资产/冻结权重/原始证据/校验和与非Kit录像的实测重建GIF。旧严格验收/accepted_skill不改写。
- **先前v3限定仿真功能已完成**:新u256原命令连续3周期完整补验通过,随后3方向×2幅度六组、各解析/u256连续3周期全部通过物理/监督/退出清理,6对初态一致。实际175500物理调用/39周期,0训练/0重试;[逐组表、测量、命令与GIF](docs/contact_cycle_functional_followup_v3_result.md)。
- 双方功能通过率均6/6,解析本身足够,不预设学习必要;30mm根XY仅诊断,原保护不改。历史ff-01的u256三周期GPU监督失败仍保留、具体原因未知,本次通过不改写[旧结果](docs/contact_cycle_functional_flow_v2_result.md)。
- 原46D参考+残差PPO已完成256更新,冻结u256名义9秒开发验证128/128通过,P0回放通过。
- 严格P1连续周期双方均在首周期水平根漂移超限;P2虽完成20行,但10组命令初态全部失配,不能支持策略优越性。
- RootXY-v2实际完成81次PPO更新;u64功能128/128通过,XY RMS约改善2.8%,但126/128仍超30mm门槛,**没有合格新候选**。约44分钟因监控文件竞态中止,修复目前仅CPU验证。
- `accepted_skill=false`;无连续稳定、独立鲁棒性、硬件或sim2real结论。当前运行授权已消费,不自动重跑。
- `accepted_skill=false`;本轮解析名义20周期及有限切换通过,不等于全域/多seed独立鲁棒验收,无硬件或sim2real结论。本轮任务提前完成,旧grant/目录不复用,不自动重跑。
## 从这里继续
@@ -18,7 +23,7 @@
- [冻结身份与证据路径](docs/evidence_index.md)
- [统一软硬件描述v2](docs/software_hardware_spec.md):Schema及静态校验已实现,通用运行adapter尚未实现。
训练默认RTX5080。新Kit运行必须新预算、新配置绑定、新授权及唯一监督器;不得复用失败目录/旧grant、改写账本或终止桌面等外来进程。准备通过不是物理资格,源码提交也不是技能验收。
训练默认RTX5080,当前任务0训练。用户已明确后续预算无需逐次请批;每次仍自主固定并记录运行护栏、新配置绑定/独立登记及唯一监督器,见[授权边界](docs/execution_authority.md)。不得复用失败目录/旧grant、改写账本、自动重试失败作业或终止无关进程。准备通过不是物理资格,源码提交也不是技能验收。
资产配置在`configs/legs/`,接触周期配置在`configs/contact_cycle/`。依赖本机IsaacLab/IsaacSim、Unitree资源及未随本次提交交付的日志/权重/生成资产;跨机器需另带实际产物并验证身份。`.env`与本机授权不提交。
@@ -0,0 +1,66 @@
{
"format": "skillfactory.contact-cycle-functional-flow.v2",
"status": "prospective_functional_development_not_independent_acceptance",
"execution_authorized": false,
"optimizer_steps": 0,
"num_envs": 1,
"seed": 93501,
"physics_dt_s": 0.002,
"control_ticks": 5,
"phase_durations_s": [
3.0,
1.0,
0.6,
1.4,
0.8,
2.2
],
"cycles_allowed": [
1,
3
],
"swing_displacement_m": [
0.025,
0.02,
0.0
],
"anchor_sequence": [
"A",
"B",
"A",
"B"
],
"functional_gates": {
"minimum_complete_cycles": 1,
"support_final_window_s": 0.5,
"reload_final_window_s": 0.2,
"contact_force_on_N": 2.0,
"support_reload_contact_fraction_min": 0.8,
"support_reload_load_fraction_min": 0.5,
"unload_final_window_s": 0.1,
"unload_remaining_support_fraction_max": 0.8,
"swing_gap_m_min": 0.01,
"swing_air_fraction_min": 0.8,
"swing_contiguous_air_s_min": 0.1,
"swing_xy_progress_fraction_min": 0.5,
"landing_position_error_m_max": 0.05,
"root_height_error_m_max": 0.1,
"normal_force_negative_tolerance_N": 0.5,
"root_tilt_rad_max": 0.6,
"ground_penetration_m_max": 0.002,
"auxiliary_vertical_N_max": 140.0,
"auxiliary_lever_arm_m_max": 0.06,
"guide_vertical_N_max": 1e-06,
"normalized_effort_velocity_max": 1.00001,
"joint_limit_excess_rad_max": 0.01
},
"diagnostics_only": [
"root_xy_error",
"root_xy_drift",
"stance_slip",
"landing_accuracy_beyond_functional_gate"
],
"runtime": "EvaluationContactCycleEnv_no_autoreset_original_phase_safety",
"phase_rule_changes": [],
"fixture_changes": []
}
@@ -0,0 +1,41 @@
{
"format": "skillfactory.functional-followup.v3",
"status": "prospective_finite_functional_development",
"execution_authorized": false,
"accepted_skill": false,
"optimizer_steps": 0,
"base_contract": "configs/contact_cycle/functional_flow_v2.json",
"base_contract_sha256": "16c064228b30f36bbd420b7c2290602bacae6601cac2cec5deb16440f7600625",
"seed": 93501,
"cycles_per_run": 3,
"anchor_sequence": ["A", "B", "A", "B"],
"direction_xy_ratios": {"x_bias": [25, 10], "baseline_direction": [25, 20], "y_bias": [15, 20]},
"amplitudes_m": [0.016, 0.024],
"direction_order": ["x_bias", "baseline_direction", "y_bias"],
"controller_order": ["analytic_reference", "learned_policy"],
"nominal_requalification_first": "learned_policy_only_0.025_0.020",
"matrix_scope": "inside_previous_component_command_envelope_not_previously_validated_every_direction",
"pairing": "reference_initial_record_then_exact_u256_check_before_first_rollout_step",
"pairing_scope": "recorded_tensors_RNG_IK_commands_not_complete_native_solver_state",
"failure_policy": "first_any_failure_stop_no_retry_remaining_not_run",
"perturbations": [],
"noise": false,
"morphology_changes": [],
"functional_or_safety_changes": [],
"fixture_changes": [],
"limits": {
"max_prepare_launches": 1,
"max_physical_launches": 13,
"max_rollout_calls": 175500,
"max_physics_calls": 178750,
"max_initialization_calls_per_launch": 250,
"max_prepare_seconds": 300,
"max_physical_seconds": 480,
"cleanup_seconds": 120,
"max_campaign_seconds": 6540,
"max_owned_bytes": 2147483648,
"num_envs": 1,
"optimizer_steps": 0,
"retry_or_resume": false
}
}
@@ -0,0 +1,4 @@
{
"mode": "combined_display",
"profile": false
}
@@ -0,0 +1,9 @@
{
"/usr/lib/xorg/Xorg": "4aa5299cfccd1b1ab1b4837288c16c2dc3a325b1f6a9147f4bf9d5017d108ea6",
"/usr/bin/gnome-shell": "5f678999f051dd1e17f2da50cb55ddc152d9b1f5d45a6ea8ead8b4df603b2a17",
"/usr/libexec/xdg-desktop-portal-gnome": "cf6839a8003098a7a4d710d520b8fcd9afbba6bbd21931bc0e420ac724b79bc7",
"/usr/share/code/code": "f36c0ef83163bb3c7b99c5860ce98a5d4055dd6b69d3a4d393d7563e499d3456",
"/home/cen/App/Antigravity/AntigravityIDE/antigravity-ide": "56987be1a655ae5903bc47963a67e147d8ee3c36e41144b03210cc9c3e57336f",
"/opt/google/chrome/chrome": "4d2512ae84986bf987e6ea8ef14ca1af555ae574fbaff5b5a8c78a8dd73fa36f",
"/usr/lib/chatgpt/ChatGPT": "c81da8d4de29a2eb4a8ad9d8c32cf090b1a3ac96158eec288c6412eeba86072d"
}
Executable
+13
View File
@@ -0,0 +1,13 @@
#!/usr/bin/env bash
# Genuine live Kit window, not the offline FK animation. No training.
set -euo pipefail
ROOT=$(cd -- "$(dirname -- "$0")" && pwd)
: "${DISPLAY:?A local graphical DISPLAY is required for the real Kit window}"
OUT=${SKILL_DEMO_OUTPUT:-$(mktemp -d "${TMPDIR:-/tmp}/go2-fl-gui.XXXXXXXX")}
mkdir -p "$OUT"
CLI="$ROOT/scripts/go2_fl_skill.py"
echo "Live GUI S1 (analytic); new simulation evidence: $OUT"
echo "Close Kit while running = ABORT. After completion, final frame stays (max 10 min)."
/usr/bin/python3.12 "$CLI" authorize --registration "$OUT/session/registration.json"
/usr/bin/python3.12 "$CLI" run --registration "$OUT/session/registration.json" --output "$OUT/S1" --sequence S1 --controller analytic --gui
/usr/bin/python3.12 "$CLI" status --output "$OUT/S1"
+8 -4
View File
@@ -1,13 +1,17 @@
# 文档入口
主线:Go2 FL仿真参考+残差,当前RootXY-v2已完成81次更新但中断,**没有合格新候选,accepted_skill=false**。本次文档/源码提交不是新的运行授权。
最新:[0.4.2应用循环与GUI性能结果](contact_cycle_sim_performance_v042_result.md),真实窗口33.98FPS截图/全段35.30提交每秒,GUI/headless S1轨迹与0.4.1文件SHA相同;物理推进分别仅0.07060×/0.18325×。源码/原件/复核在仓库`performance_v042/`,独立包`performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2/`及最终封装/库存凭据见该工作根;不是GPU100%或1×实时,不继承完整矩阵/u256/原生故障注入资格,旧包及accepted_skill=false保留。
主线:**0.4.1真实Kit GUI已实际验证**,[结果/真实图像/限制](contact_cycle_sim_gui_v041_result.md)、[GUI固定计划](contact_cycle_sim_gui_v041_plan.md)及[API](go2_fl_skill_v4_api.md)。独立包`bash demo_gui.sh`真实解析S1;GUI/headless各六周期通过、提前关窗按预期中止并清理。0.4.0原件不变,0.4.1不自动继承完整矩阵/u256资格,不追加训练/优化/扩域。
先前**Go2 FL接触周期仿真功能版0.4.0已交付**。可调用Python/CLI、解析20连续周期、三条六周期切换、u256同S1、停止拒绝及独立包真实仿真复现均通过,零训练;[本轮实际结果/命令/产物](contact_cycle_sim_skill_v4_result.md)、[API](go2_fl_skill_v4_api.md)、[预先固定合同](contact_cycle_sim_skill_v4_plan.md)。有限单seed/有辅助条件不是独立鲁棒验收,历史accepted_skill=false/严格失败不变,不追加精度优化/整机实机;文档或源码提交不是运行grant。先前[v3结果](contact_cycle_functional_followup_v3_result.md)保留。
## 下一阶段必读
1. [当前进度](current_status.md):唯一当前状态摘要;物理结果、失配、部分训练与预算分开记录。
2. [下一阶段](next_stage.md):监控资格验证→新绑定/新授权→有界试训→后续能力验证。
2. [当前阶段结论](next_stage.md):限定功能目标完成,不自动试训/扩域;[持续执行授权](execution_authority.md)允许后续自主设预算护栏而不逐次请批。
3. [接口与CPU测试](contact_cycle_runtime.md):46D、动作/物理边界、实现入口和测试命令。
4. [RootXY固定配方](contact_cycle_rootxy_v2_plan.md) / [受监督启动合同](contact_cycle_rootxy_v2_launch.md):预算、baseline/开发门槛、GPU共享及最终候选发布。
4. 当前先读[0.4.1真实GUI结果](contact_cycle_sim_gui_v041_result.md)和[公开接口](go2_fl_skill_v4_api.md);[0.4.0完整矩阵结果](contact_cycle_sim_skill_v4_result.md)保留为历史;历史[功能补验预先计划](contact_cycle_functional_followup_v3.md) / [实际逐组结果](contact_cycle_functional_followup_v3_result.md):命令、原门槛、初态配对、独立预算/绑定和真实Fn/gap/位移;原RootXY配方不再是默认推进方向。
5. [评估/溯源/证据/监控](contact_cycle_native_monitor_v3.md):合并P1/P2、初态修复、v5绑定、XZ与native扫描合同。
6. [证据索引](evidence_index.md):冻结身份、外部报告、失败账本和不可选取检查点。
@@ -19,7 +23,7 @@
## 历史资料
过程性修复/第N次尝试及过期“下一步”已移出docs活动文档。归档仅用于溯源,其中的命令、预算、时钟与授权均不能直接执行。12份活动Markdown(含2份冻结输入)+2份Schema;不再按失败次数新增文档。
过程性修复/第N次尝试及过期“下一步”已移出docs活动文档。归档仅用于溯源,其中的命令、预算、时钟与授权均不能直接执行。历史文档数量记录不作为当前清单;新功能计划/实际结果按任务保留,不按失败次数追加“第N次修复”文档。
- 本次整理前全部docs及根README:[归档](../archives/documentation_before_rootxy_handoff.tar.gz)、[逐文件SHA256清单](../archives/documentation_rootxy_handoff.json)。
- 更早阶段:[原归档](../archives/documentation_before_simplification_20261014.tar.gz)、[原清单](../archives/documentation_cleanup_20261014.json)。
+131
View File
@@ -0,0 +1,131 @@
# Go2 FL 功能流程 v2:实验前固定判据与准备记录
## 最新结论(详细实际结果另列)
用户批准后已完成新物理运行:**解析单周期及连续3周期连同监督检查通过**;冻结u256单周期通过,三周期物理完成但GPU归属监督失败,原campaign失败不改写。实际36000共享物理调用、0优化,没有重试。实际命令、阶段结果、图/GIF和未解决的监督问题见[运行结果](contact_cycle_functional_flow_v2_result.md)。
## 运行前结论(以下准备/提案为当时记录,不是当前待授权状态)
当时尚未完成新物理验证,尤其没有连续3周期通过证据;准备阶段新增物理调用=0、优化更新=0。不使用RootXY u64、不训练、不开展整机/实机/sim2real。
历史u256单周期可以复核为功能通过;历史严格P1双方仍是首周期卸载阶段漂移失败,不能事后改为通过或延长其轨迹。最近RootXY训练仍是81更新后基础设施中断,和本轮功能进展分开。
## 1. 新实验前固定的简单判据
固定文件:`configs/contact_cycle/functional_flow_v2.json`,SHA256 `16c064228b30f36bbd420b7c2290602bacae6601cac2cec5deb16440f7600625`。功能阈值**逐值复用**旧`simulation_functional_v1.json`,不是看新实验结果调门槛。
| 阶段 | 本轮功能判据 |
|---|---|
| 承重 | 承重末0.5s,真实法向力>2N的采样比例≥80%,平均Fn≥0.5W |
| 卸载 | 卸载末0.1s平均Fn / 前述承重窗口平均Fn≤0.8,即至少下降20%;不要求归零 |
| 抬腿/摆动 | 摆动段gap≥10mm且Fn≤2N的比例≥80%,连续满足≥0.1s |
| 三维摆动 | 保持50mm抬高+10mm拱高,命令XY为25/20mm;实际有向XY进度每轴≥命令50%(约12.5/10mm) |
| 落地/重新承重 | 重新承重末0.2s,接触比例≥80%、平均Fn≥0.5W;该窗口平均3D落点误差≤50mm |
| 连续动作 | 先完整9s单周期,再独立完整27s三周期;每周期都通过,不用平均成绩抵消失败 |
| 安全 | 全程有限;gap≥−2mm;根部高度误差≤100mm、tilt≤0.6rad;关节超界≤0.01rad,归一化力矩/速度≤1.00001;有界辅助保持原值,竖直guide严格为0 |
W为仿真有效系统重量,当前约52.86N(质量约5.388kg),不是仅3kg载荷的重量。上述0.01rad是已有数值容差,不是扩大命令/执行器硬限制。
- 根部XY的30mm、旧P1接缝漂移20mm、额外落点/滑移精度只记录,不作为本轮功能门槛;依旧记录最大值和超限事实。保留碰撞、失控、高度、倾斜、关节、执行器、数值与辅助保护。
- 初次接触获取仍要求200ms内达到≥0.5W、gap≤0.5mm并持续10ms;不能以从未接触的飞行跳过承重。
- **初版不修改原相位终止规则**:仍保留提前离地保护、抬腿90%后gap≥20mm/Fn≤2N、摆动逐tick gap≥10mm、重新承重80%后Fn≥0.5W。它们可能比窗口功能判据严格;若成为首个实际阻碍,应另记具体失败、前瞻登记最小相位修订后再申请相应运行,不能忽略失败继续。
- 全程500Hz物理/保护,100Hz控制;每周期时长仍为3/1/0.6/1.4/0.8/2.2s。
## 2. 最小实现与诚实的控制归属
新增`legs/contact_cycle_sim/functional_flow_v2.py`:独立`FUNCTIONAL_V2`描述符,复用已验证P1整数时钟和固定世界锚点A→B→A→B解析命令,**不带P1精度门槛**。27s的13500个命令与旧P1逐tick一致;两次接缝位置/载荷命令连续。没有改变IK、Jacobian载荷补偿、46D观察、±0.05rad残差权限或装置。
拟复用原`EvaluationContactCycleEnv`无autoreset运行时及`RuntimeHooks`:接缝先检查端点,再仅推进命令元数据并刷新一次参考;禁止reset/pose写入、落地重锚、轨迹拼接,保留物理/IK warm-start/动作与延迟/接触状态/全局时钟。新描述符的CPU测试**不等于**新运行器已获Kit资格。
两种控制分别出结果:
1. **解析参考**:请求残差与有效残差为0,不加载actor做控制。
2. **解析参考+冻结u256确定性均值残差**:使用原候选`tanh(actor_mean)`,不是策略独立控制、不是随机采样、不是RootXY u64。记录请求残差、投影后残差、reference-used/next、最终关节目标。
原u256 SHA256仍是`c088bb3aecaca751867118858626fd508f7a3983303802d7524353f80802cf99`。本轮不要求学习优于解析,不从跨运行的历史数字推断学习增益。
**启动尚有明确未完成项**:原P1/P2 runner固定校验旧协议,不能直接载入新描述符;新功能受监督入口/新配置绑定尚未接入和Kit验证。不得把协议文件换个路径传给旧runner、改旧P1阈值,或借旧grant/通用监督器绕过其认证。下一步仅做这项必要适配,不追加RootXY监控建设。
## 3. 已有真实仿真证据的只读复核
产物根:`logs/contact_cycle/functional_flow_v2_prepare01/`(本机日志,不随源码自动分发)。所有新报告为离线审计,原轨迹与旧结果未写入。`trace_functional_passed`只表示保留张量满足功能窗口;`runtime_continuity_verified=false`始终明确保留,张量本身不能证明无状态写入。
| 来源 | 本次离线复核 | 局限 |
|---|---|---|
| 原P0 u256 `first_episodes.pt` | 1/1单周期功能通过,9s/4500采样,无记录失败,记录动作等于tanh(mean) | 原P0初始化额外tick计数不足,不能当新运行或连续证明 |
| rxy-04解析参考 `validation_analytic_zero_selected.pt` | 保留4/4单周期通过,记录有效残差为0 | 仅选定4条,不重审完整128环境;减字段轨迹缺原始failure_flags等,不能补写完整安全/退出证明 |
| rxy-04冻结u256 `validation_frozen_u256_selected.pt` | 保留4/4单周期通过 | 同上;没有新增受监督回放或有效成对比较 |
| 同一P0轨迹请求3周期审计 | 拒绝通过,退出码2,后两周期not_evaluated | 明确禁止把9s扩为27s |
### P0 u256的阶段测量
下表Fn为**整相位均值**,功能窗口结果另列,避免混淆重新承重的上升过程。
| 阶段 | Fn均值N | gap范围mm | 本阶段足端净位移XYZ mm |
|---|---:|---|---|
| 承重 | 52.86 | −0.0046~0.00006 | 1.86 / 0.17 / 0.002 |
| 卸载 | 40.41 | −0.0238~−0.0021 | −3.65 / −1.64 / −0.021 |
| 抬腿 | 5.19 | −0.0241~47.76 | 58.61 / −5.44 / 47.78 |
| 摆动 | 0 | 47.44~57.99 | 19.46 / 20.16 / 1.00 |
| 落地 | 0.016 | −0.00094~48.80 | −2.36 / −1.17 / −48.80 |
| 重新承重 | 25.90 | −0.0013~0.0022 | −1.34 / 0.11 / 0.002 |
- 承重末窗口1.000W,卸载末0.481W,重新承重末0.981W;摆动全1.4s满足离地,实际有向进度0.776/1.007,平均落点误差9.967mm。
- 最大根XY误差**68.981mm,仍如实记为超过30mm**;旧P1的严格失败不因此更改。
- 原P0终止为9s timeout,无physical_failure;最大执行器力矩/速度比0.19884,关节超界0,竖直guide0。无新视频;`p0_u256_trace_plot.png`/`analytic_selected_env0_trace_plot.png`是**历史测量曲线,不是渲染关键帧或新实验**。
- 解析参考保留env0:承重末1.000W、卸载末0.676W、重新承重末0.996W;摆动gap43.38~53.76mm、Fn0,阶段净XY21.55/18.34mm,落点误差13.703mm、最大根XY80.483mm。这些跨运行数字不构成策略优势对照。
- 历史P1首个失败仍分别为u256在3.658s、参考在3.540s,首周期卸载,原因`P1_root_horizontal`。其后动作没有采样,不能说明放宽此门槛后必然完成。
## 4. 可复现的当前命令(CPU,不启动Kit)
仓库根目录:
```sh
PY=/home/cen/miniconda3/envs/linker_sim_isaac/bin/python
CUDA_VISIBLE_DEVICES='' "$PY" scripts/audit_cycle_functional_flow.py \
--trace logs/contact_cycle/frozen_replay_p0_01/run/first_episodes.pt \
--controller-kind learned_policy --cycles 1 \
--output /tmp/sf-functional-p0-audit-NEW.json
# --output必须尚不存在;--cycles 3应返回2,不会拼接/重复数据。
CUDA_VISIBLE_DEVICES='' PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 "$PY" -m pytest -q \
tests/test_contact_cycle_functional_flow_v2.py tests/test_contact_cycle_functional.py
```
本次实际跑过全接触周期CPU回归:**1329 passed、6 skipped,108.37s**,日志/XML见证据根。不是全仓/PhysX回归;新增功能文件自身24项,加旧功能检查18项共42项。另做compileall和diff whitespace检查。
原P0真实启动命令记录在`logs/contact_cycle/frozen_replay_p0_01/supervisor.json`:
```sh
# 历史复现模板,不是本轮运行授权;只支持单周期,不支持连续3周期。
/home/cen/IsaacLab/isaaclab.sh -p scripts/replay_cycle_sim_residual.py \
--replay --controller learned --num-envs 1 --headless --device cuda:0 \
--output /home/cen/sf-runs/functional-p0-NEW/run
# reference将--controller learned换成reference;实际运行需新批准与监督。
```
新功能三周期启动命令**尚未交付为可执行/合格命令**,须在新预算批准后完成上述有限入口适配和配置准备,再将实际命令/绑定存证;不能提供一个旧runner实际拒绝的假命令。
冻结检查的差异也保留:旧`verify_delivery.py`本次在AGENT.md逐字节比较失败(已有用户指令修改,未被本次改动)。随后使用现有`replay.verify_bundle`允许的**唯一既有AGENT非运行时代码例外**,134个冻结快照、其余live输入、候选身份和CPU确定性加载通过,见`frozen_cpu_preflight/`。没有修改验证器或冻结manifest来凑通过。
## 5. 一次性新仿真预算请求(原提案保留,后经用户批准并已消费)
只为功能流程,不是训练预算,也不沿用P1/P2/RootXY旧授权:
| 项目 | 提议上限 |
|---|---|
| GPU/环境 | RTX5080,单环境,500Hz物理/100Hz控制;无并行作业 |
| 配置准备 | 最多1次Kit配置准备,600s含120s清理;禁止构造环境/物理步/优化 |
| 物理运行 | 最多4次独立启动,每次600s含120s清理 |
| 顺序 | 解析1周期 → u256 1周期 → 解析连续3周期 → u256连续3周期;两个单周期先全部通过再进入连续验证 |
| rollout | 4500+4500+13500+13500 = **36000共享物理调用**(本轮1环境) |
| 初始化/预热 | 每物理启动≤250次,共≤1000;必须在构造前安装计数,超限停止 |
| 总物理上限 | **37000共享调用**,不是预计实际用量;最大7200个控制步,优化步0 |
| 时限 | campaign同boot绝对总窗口≤3000s(50分钟,含上述准备/清理),不另加隐藏运行 |
| 存储 | 新隔离根,own产物≤1GiB;native正增长另监控并声明采样覆盖局限,不冒称硬配额 |
| 故障处置 | 首个功能/安全/运行失败即保留证据并停campaign;不重试、不恢复、不现场改阈值续跑;未执行预算不称作实测调用 |
先完成必要独立功能入口适配、绑定与CPU校验,再在这份**新授权**内配置准备并父级核对后串行执行。若已有安装不能满足准备/运行保护,停止并报告,不把本预算变为监控建设或训练许可。
新的物理运行须交付每tick原始轨迹、初态/命令/动作归属、初始化和部分控制计数、两接缝reset/pose写入计数零增量及carry-state核对、终止首因、clean exit/清理证据;尽量在同预算内采关键帧。视频失败不掩盖物理失败,也不能据此加启动。
**首要未解决问题**:放宽非安全精度门槛后,真实单腿能否不间断地完成第二、第三周期,当前证据不足。最小下一步是上述无训练对照验证,不是继续RootXY优化。
@@ -0,0 +1,140 @@
# Go2 FL 功能流程 v2:真实单周期与连续3周期结果
## 结论与必须保留的失败
**解析零残差控制已在现有辅助装置下完成单周期及无重置连续3周期,功能与监督检查均通过。** 本轮“完整流程能否连续完成”的主目标已有新的真实仿真证据,不需要追加训练。
冻结u256+解析参考也记录了完整单周期和连续3周期,全部物理功能窗口、保护与接缝检查通过;**但u256连续3周期这一行的监督器在GPU归属检查报错,故该行监督不通过,原`campaign-result.json.success=false`保持不变。** 不把worker完成、退出0或事后清理通过替代这一失败;没有重试。
| 控制方式 | 实际长度/物理采样 | 物理功能结果 | 监督结果 | 最大根XY漂移(仅记录) |
|---|---:|---|---|---:|
| 解析参考,残差严格0 | 9s / 4500 | 1周期通过 | 通过 | 80.49mm |
| 解析参考+冻结u256确定性残差 | 9s / 4500 | 1周期通过 | 通过 | 68.98mm |
| 解析参考,残差严格0 | 27s / 13500 | 连续3周期通过 | 通过 | 80.49mm |
| 解析参考+冻结u256确定性残差 | 27s / 13500 | 连续3周期物理完成 | **GPU归属检查失败** | 68.98mm |
对象仍为Go2 FL单腿、浮动根、仿真真值、3kg载荷/测力载体、有界XY/转动guide及主动竖直辅助,**竖直guide为0**。这不是整机、无辅助平衡、独立鲁棒验收或实机/sim2real能力;`accepted_skill=false`继续保留。所有旧严格P1失败和RootXY候选判定不改写。
## 1. 固定判据、控制与最小改动
判据在实验前固定于`configs/contact_cycle/functional_flow_v2.json`,SHA256 `16c064228b30f36bbd420b7c2290602bacae6601cac2cec5deb16440f7600625`;详情见[准备与判据](contact_cycle_functional_flow_v2.md)。本轮没有改变阈值、相位时长、原相位终止、IK/Jacobian载荷补偿、残差上限、装置或权重。只在独立功能协议中不施加原P1的额外精度门槛。
- 解析:实际请求/有效残差全程为0,actor不参与控制;不能称为学会了动作。
- u256:原冻结权重 SHA256 `c088bb3aecaca751867118858626fd508f7a3983303802d7524353f80802cf99`,46D观察,`tanh(actor_mean)`确定性残差;不是策略独立控制或RootXY u64。实际有效残差峰值0.047431rad,记录请求、投影后残差、reference-used/next、实际关节目标和执行力矩。
- 四行记录的初态张量逐字段相同;不据此宣称未观测的PhysX缓存/数值调度完整相等。记录动作等于确定性均值输出,且合成目标与有效残差逐tick核对通过。
- 解析本身已足以完成三周期。残差确实改变动作/轨迹,但本轮不要求、也不声称学习在独立验收中优于解析;尤其不能把监督失败的u256三周期当成完全合格对照。
新增实现仅为必要的独立入口/绑定/紧凑记录:
| 文件 | 用途 |
|---|---|
| `scripts/evaluate_cycle_functional_flow.py` | 固定顺序的一次性功能campaign;受监督FD交接,首失败即停,无训练/重试 |
| `scripts/functional_flow_launch_v2.py` | 新用户授权及预算;复用已有启动、owned-PID、GPU/native监控原语,不复用旧grant/ledger |
| `legs/contact_cycle_sim/functional_flow_binding_v2.py` | 实际cfg、冻结候选及有限源/安装依赖绑定;准备不构造环境 |
| `tasks/manager_based/contact_cycle_sim/functional_flow_recording_v2.py` | 同步500Hz完整记录、分阶段功能窗口、接缝carry-state核对;不每tick复制整份RNG |
| `scripts/visualize_cycle_functional_flow.py` | 只读测量曲线、FK关键状态图和GIF;不启动Kit或执行控制 |
`EvaluationContactCycleEnv`、原参考、原残差动作及原安全代码未原地修改。没有进行RootXY优化或任何训练。
## 2. 各周期实际测量
W为有效系统重量52.8597N。承重末窗口各周期约1.000W,接触比例均100%;下表“卸载剩余”是末0.1s载荷相对承重末0.5s窗口的比值,不是整相位均值。所有摆动阶段实际Fn=0,完整1.4s满足离地;XYZ净位移取摆动阶段首末采样。脚抬高阶段的额外水平运动不计入摆动进度冒充成功。
| 控制/周期 | 卸载剩余(要求≤0.8) | 摆动gap范围mm | 摆动净XY mm | 重新承重末/W | 平均落点误差mm |
|---|---:|---|---|---:|---:|
| 解析单周期 | 0.676 | 43.38~53.76 | +21.54 / +18.34 | 0.996 | 13.70 |
| u256单周期 | 0.481 | 47.44~57.99 | +19.46 / +20.16 | 0.981 | 9.97 |
| 解析连续第1周期 | 0.676 | 43.38~53.76 | +21.54 / +18.34 | 0.996 | 13.70 |
| 解析连续第2周期 | 0.210 | 43.46~53.88 | −24.31 / −19.82 | 0.988 | 13.79 |
| 解析连续第3周期 | 0.167 | 43.44~53.89 | +24.44 / +19.31 | 0.996 | 13.66 |
| u256连续第1周期† | 0.481 | 47.44~57.99 | +19.46 / +20.16 | 0.981 | 9.97 |
| u256连续第2周期† | 0.026 | 47.75~57.98 | −24.16 / −19.50 | 0.974 | 8.75 |
| u256连续第3周期† | 0.006 | 47.80~57.96 | +21.94 / +20.22 | 0.981 | 9.92 |
† 物理轨迹与worker检查完成,监督行仍失败,不作全面通过声明。
完整48行“每运行×每周期×每相位”的法向力min/mean/max、gap范围、XYZ位移、根XY最大值:`/home/cen/sf-runs/ff-authority-01/actual-phase-summary.csv`;结构化全部门槛与结果见`actual-summary.json`及各`*-offline-audit.json`。全程真实接触力来自物理contact sensor,不使用auxiliary/guide力冒充足端载荷。
## 3. 连续性与安全
两种控制的三周期原始轨迹均为**同一环境的全局tick1..13500、固定世界A→B→A→B**;不复用单周期产物、不拼接、不重锚。
- 初始reset/pose写入计数分别22/14;这包括允许的构造与唯一初始reset。起始后全程计数不增加,不能误写为“初始化也没有reset”。
- 在tick4500、9000端点先审计该周期,再推进元数据;各45项枚举状态/RNG指纹前后相同。物理姿态/速度、关节、动作/目标历史、传感/执行延迟、接触/离地/获取状态、调度器与全局时钟均承接。
- IK参考只正常刷新一次:900→901、1800→1801;没有controller reset或物理步夹入接缝。
- 全部四行终止均为`protocol_complete`,物理cause为空,无partial-control补齐。三周期各2700控制/13500采样,单周期各900/4500;没有timeout-autoreset后的轨迹。
- 原数值、碰撞、关节/执行器、初次接触获取、提前离地、抬腿/摆动/重新承重期限、低高度/倾斜与辅助保护均启用。初次承重接触在约0.010s获取。
四条完整实测轨迹的安全量:最小gap为−0.0298mm,最大关节超界0,最大归一化执行器力矩/速度比0.2153,竖直guide0,最大倾斜0.05892rad,最大高度误差67.78mm;没有记录到数值异常、明显穿透、失控倒塌或执行器越界。根水平漂移80.49/68.98mm继续记录,不是本轮失败条件;不改变旧30mm严格失败。
## 4. 监督失败:不隐藏、不反推原因
原失败位于`ff-01/campaign-result.json`的`u256-3.supervisor_error`:
```text
GPUOwnership.check -> RuntimeError:
GPU ownership lost or residual task/foreign GPU client
```
worker实际报告完成13500tick、所有物理/连续性检查通过、preclose同对象cfg身份通过,退出码0。监督行同时记录`success=false`;整个campaign返回失败,没有重跑。
父级事后只读核对:446个绑定文件重哈希通过,新运行源码与原u256未变;全部owned PID/birth身份消失,无计算或任务图形残留,原7个桌面图形客户端仍在。**这些事后事实不能反证失败瞬间没有外来GPU客户/残留,也不能证明一定只是退出时NVML滞后**。失败瞬间的NVML clients与owned PID集合没有落盘,具体瞬时原因当前证据不足。
动作阶段没有首个失败;首个未通过的是u256三周期运行末段的GPU归属门禁。若以后确需补齐该行监督资格,最小下一步是先为该失败保留当时clients/owned快照并CPU验证,不放宽归属检查,再另行批准一个全新有界回放。**当前解析目标已完成,不默认追加这一运行,更不训练。**
## 5. 预算、退出及证据
| 项目 | 实际/边界 |
|---|---|
| 授权 | 用户本轮明确“批准”;新`ff-authority-01/approval.json`,不沿用旧P1/P2/RootXY身份 |
| 启动 | 1次配置准备+4次串行物理启动,全部消费,无重试/恢复 |
| 物理 | 36000完成共享`SimulationContext.step`调用,初始化额外调用0;37000为保留上限,不是实测,不将剩余额度用于新启动 |
| 时长 | campaign约815.045s(13.58分钟),上限3000s;含准备、四运行及清理 |
| GPU/训练 | RTX5080、1环境、500/100Hz;优化0,不开展实机/整机/sim2real |
| 退出/清理 | 五个子进程均等待退出0且owned清理通过;最后一行的GPU监督失败仍单独保留 |
| 存储 | 父级检查时run根409655699字节+authority约数MB,合计远低于1GiB;最终逐文件inventory另存 |
| Native | 监控正增长340282589字节,含与run根重叠的redirect内容,不能与own bytes重复相加;不是硬配额、覆盖不完整 |
证据目录均为本机外部文件,不自动随源码提交:
- `/home/cen/sf-runs/ff-01/`:新授权副本、预算预留、零物理准备/独立审查、源快照、四行`trace.pt`/`initial-state.pt`/`events-and-seams.json`/worker及supervisor结果、原campaign失败和native报告。
- `/home/cen/sf-runs/ff-authority-01/`:新授权原件、实际命令、GPU初态、父级退出核对、独立离线审计、全量阶段CSV/JSON、图/GIF及最终inventory。
- 准备绑定 SHA256 `f6a6edc91ba7113ff1fa5a8a8c78e10d0dfe2fd1397881abc4e0a34add8bc7b7`;inventory SHA256 `770f2d923777fcf4ed3a993b86347ee27d1cf49f3beef0b61c58fcb8114d2bd0`。它们是这次配置身份,不是可重用运行授权。
测试:启动前相关CPU **341passed**;结束后接触周期CPU **1341passed、6skipped(108.65s)**;compileall和diff whitespace检查。CPU测试不替代本次真实物理,也不改写监督失败;不是全仓全依赖或RootXY训练资格验证。
## 6. 启动与可视化复现
实际已执行的命令(**存证,不可重用旧grant/目录再次执行**):
```sh
/usr/bin/python3.12 scripts/evaluate_cycle_functional_flow.py --campaign \
--output /home/cen/sf-runs/ff-01 \
--approval /home/cen/sf-runs/ff-authority-01/approval.json \
--trusted-approval-sha256 1d46b9cb220594967c93cf6be271879686e309ccff891ca9bfcd241d31ac6b61
```
启动固定控制顺序:零物理PREP → 解析单周期 → u256单周期 → 解析连续3周期 → u256连续3周期。新机器/新运行需要实际冻结资产、新预算、新root/approval、配置绑定与监督;脚本拒绝已有root,不支持自动重试。具体子进程命令/每槽时钟/初态/参考身份在各`reservation.json`留存。
无需Kit即可复核实际三周期或生成图:
```sh
PY=/home/cen/miniconda3/envs/linker_sim_isaac/bin/python
CUDA_VISIBLE_DEVICES='' "$PY" scripts/audit_cycle_functional_flow.py \
--trace /home/cen/sf-runs/ff-01/reference-3/trace.pt \
--controller-kind analytic_reference --cycles 3 --output /tmp/sf-actual-reference3-NEW.json
CUDA_VISIBLE_DEVICES='' "$PY" scripts/visualize_cycle_functional_flow.py \
--trace /home/cen/sf-runs/ff-01/reference-3/trace.pt \
--label 'analytic zero residual | three continuous cycles' \
--output /tmp/sf-actual-reference3-visuals-NEW --gif
```
`*-NEW`必须不存在。离线audit只检查张量,`runtime_continuity_verified=false`是其有意保留的作用域;运行期连续性另由原worker/hook/seam证据证明,不能靠离线输出把监督结果提升为通过。
可直接查看:
- [解析三周期状态图](file:///home/cen/sf-runs/ff-authority-01/reference-3-visuals/measured_state_keyframes.png) / [GIF](file:///home/cen/sf-runs/ff-authority-01/reference-3-visuals/measured_state_animation.gif) / [测量曲线](file:///home/cen/sf-runs/ff-authority-01/reference-3-visuals/timeline.png)
- [u256单周期状态图](file:///home/cen/sf-runs/ff-authority-01/u256-1-visuals/measured_state_keyframes.png)
- [u256三周期状态图(监督失败已标注)](file:///home/cen/sf-runs/ff-authority-01/u256-3-visuals/measured_state_keyframes.png) / [GIF](file:///home/cen/sf-runs/ff-authority-01/u256-3-visuals/measured_state_animation.gif)
图/GIF由**真实测得根姿态和关节角的FK+真实足位置/接触力**只读生成,已核对显示帧FK足端与测量一致;它们不是Kit摄像机截图或新物理运行,不把示意渲染冒称为原始仿真视频。完整500Hz原始轨迹未抽稀,仅显示动画采样10fps。
@@ -0,0 +1,98 @@
# 功能补验 v3:实验前监督诊断与固定有限命令矩阵(准备记录)
**后续实际执行已完成且全部通过**:新u256连续3周期补验及六组成对矩阵物理/运行完整性全通过,175500共享物理调用、0训练/0重试;见[实际逐组结果](contact_cycle_functional_followup_v3_result.md)。用户同时授予[后续无需逐次请批的预算权限](execution_authority.md)。以下完整保留运行前诊断、计划、当时未运行状态和预算提案,不作为当前状态或可重用命令。
## 运行前交付/未完成(历史时点)
- 旧`ff-01`原件298个文件重哈希全部一致;原campaign/u256三周期监督失败不变,旧严格漂移失败不变。
- 已只读诊断、补充最小GPU失败观测、固定一次u256补验及6组成对矩阵,完成CPU回归。**本轮Kit启动0、物理调用0、训练0;没有生成新grant或复用旧grant。**
- 新授权尚未获批,新的配置绑定/物理运行/矩阵都未执行,不能把CPU mock、计划或历史u256轨迹当作补验通过。
## 1. 监督失败诊断:知道什么,不知道什么
源:`/home/cen/sf-runs/ff-01/u256-3/{worker-result.json,supervisor-result.json,stdout.log}`、原源快照、GPU初态/退出核对及native终态日志;全部输入hash见`logs/contact_cycle/functional_followup_v3_prepare01/previous-supervision-diagnosis.json`。
- 同一真实27s轨迹已完成13500采样/2700控制/两个接缝,worker物理/连续性/preclose身份通过并退出0;监督行仍失败。
- 原栈位于**live** `GPUOwnership.check()`的客户集合归属分支:`compute-active`或`graphics-active-approved_desktop`非空(含禁止桌面compute)。不是显式NVML库漂移、NVML查询异常或桌面身份变化报错分支,也不是final-only残留分支。
- 记录的owned PID/birth为`729465/27594951`、`729470/27594951`、`729475/27594955`、`729526/27595002`。worker最终CLOCK_BOOTTIME为276208.208381944,监督行结束为276211.857108627,差3.648727s。native记录了quick framework shutdown/exit;**native timestampNs时钟与BOOTTIME没有本轮校准,不混用作精确因果时序**。
- 退出后owned身份消失、GPU无compute/任务graphics、原7个桌面graphics客户仍在。这不证明检查瞬间没有外来客户、PID身份变化或观测竞态。
- 本轮另行只读查询当前GPU和`/proc`:没有compute客户,四个旧owned PID当前都不存在;accounting当前Disabled/Accounted Processes None,未启用或修改任何设置。这是当前状态,不能恢复或证明旧失败瞬间的情况;见`current-gpu-readonly.log`及`current-process-readonly.json`。
- **未知**:失败瞬间两类NVML客户PID列表、active owned/PID-birth集合、检查准确时间。因此无法区分真实外来、owned身份重用/变化与NVML/proc退出观测竞态;不归因为某个“最可能原因”。
最小修改是`scripts/functional_gpu_watch_v3.py`:调用原`GPUOwnership.check`,**原决策、顺序、失败条件不变,异常原样重新抛出**。仅保留检查直接消费的NVML列表、owned discovery/known身份、时钟和最近16次检查;失败后的`/proc`单独标为after-check证据。每秒及所有失败/最终检查发布`gpu-observations.json`。没有容忍/重试失败检查、NVML宽限、删除归属门禁、整体关闭检查或终止无关进程。
CPU构造了“NVML仍返回已知PID而discover已无此PID”的条件来复现该拒绝分支;它**仍失败**。这只是可复现的逻辑机制,不证明旧现场就是该机制。外来compute/新graphics、桌面compute、PID重用、桌面身份变化、库漂移、NVML异常与诊断存储失败均继续fail closed。
## 2. 实验前固定命令与判据
新描述符:`configs/contact_cycle/functional_followup_v3.json`,SHA256 **`aaed45884eab9912db6c7bac1d82e2ab27784982c524a838b7886cdd763146f4`**。它不是运行许可。
基准功能合同仍为`functional_flow_v2.json`(SHA256 `16c064228b30f36bbd420b7c2290602bacae6601cac2cec5deb16440f7600625`)。承重/重新承重≥0.5W且接触≥80%,卸载下降≥20%,真实摆动离地gap≥10mm且Fn≤2N比例≥80%/连续≥0.1s,有向XY各≥命令50%,落点窗口误差≤50mm;原数值/关节/执行器/高度倾斜/相位期限/碰撞及辅助保护均不改,30mm根XY只记录。
阶段/装置/权重保持原样:单Go2 FL、浮动根、有界XY/转动guide与主动辅助、竖直guide0,仿真真值,500/100Hz、9s周期、50mm抬高+10mm拱高、固定世界A→B→A→B;无扰动、噪声、形态变化、训练或参数/阈值搜索。u256仍为`c088bb3aecaca751867118858626fd508f7a3983303802d7524353f80802cf99`,解析参考+冻结确定性均值残差,不是独立策略。
先仅执行一次原命令`(+25,+20)mm`的u256连续3周期补验。必须同时满足物理功能、运行连续性、完整产物/preclose身份、退出0、GPU归属/清理/监督结果,才允许进入下表。失败则停止,不重试、不在剩余预算改阈值再跑。
幅度定义为水平向量长度16/24mm;方向为21.80°、38.66°、53.13°,两轴均非零,以沿用原XY功能门槛。新端点均落在已验证命令的组件包络`0<dx≤25mm、0<dy≤20mm`内;**包络内插不等于这些新方向已经验证,更不是全工作域/鲁棒性证据**。每组第2周期自动反向,第3周期再次正向。
| 固定组/顺序 | 水平命令dx/dy mm | 解析连续3周期 | 冻结u256连续3周期 |
|---|---|---|---|
| x_bias / 16mm | +14.855627 / +5.942251 | 未运行 | 未运行 |
| x_bias / 24mm | +22.283441 / +8.913376 | 未运行 | 未运行 |
| baseline_direction / 16mm | +12.493901 / +9.995121 | 未运行 | 未运行 |
| baseline_direction / 24mm | +18.740851 / +14.992681 | 未运行 | 未运行 |
| y_bias / 16mm | +9.600000 / +12.800000 | 未运行 | 未运行 |
| y_bias / 24mm | +14.400000 / +19.200000 | 未运行 | 未运行 |
在此准备记录形成时,全表均`not_run_new_budget_pending`;后续实际通过另见结果报告,不将ff-01旧结果填入本预先登记表。首次任意功能、安全、身份、监督/存储异常停止campaign,保留首失败及所有剩余not_run。
## 3. 初态配对与比较门禁
每组先独立新进程运行解析,**其首个rollout物理调用之前**记录初态。对应u256也是新进程;在初始化后、首个rollout物理调用前,与该组解析记录精确比对:所有原初态张量指纹、枚举carry-state/RNG、IK/action/refresh、实际观察、固定世界anchors和起始命令。没有测量落点重锚或失配后修复。
比对不通过时u256不进入控制循环;不发布学习差值。只在两行物理/监督都通过、初态文件hash一致时,才标记比较eligible。解析先执行不等于提前发布成对优劣结论;最终比较必须等u256初态门禁及完整运行。此检查不宣称PhysX内部缓存完整相等。
本轮不预设u256优于解析;若两者都能完成而无必要增益,或残差使小命令更差,将如实报告,不自动训练。
## 4. 运行前唯一预算提案(后已批准并执行,不是可复用授权)
| 项目 | 拟申请上限 |
|---|---|
| 运行设备 | RTX5080、1环境、串行、0优化/训练 |
| 启动次数 | 1次零物理配置准备+13次物理:u256补验1次,成功后矩阵6×2次;无重试/恢复 |
| 共享物理调用 | rollout `13×13500=175500`;每物理启动初始化≤250,共≤3250;合计**178750** |
| 时长 | 准备≤300s,每物理启动≤480s,均含120s清理预留;campaign总≤**6540s/109分钟** |
| 存储 | own根≤**2GiB**;native采样增长≤2GiB(部分重叠,非硬配额、覆盖有缺口);空闲空间门禁4GiB |
| 顺序 | PREP → u256原命令3周期 → 六组(每组解析3周期、u2563周期) |
| 停止 | 首失败立即停止后续,不终止无关进程,不退还/重用旧预留 |
建议全新目录`/home/cen/sf-runs/ff-v3-01/`及`/home/cen/sf-runs/ff-v3-authority-01/`,目前均未创建。旧ff-01或P1/P2/RootXY grant、binding、PID/时钟不能授权当前修改后的源码;新配置准备是必要门禁。
## 5. CPU资格与可复现命令
本轮接触周期CPU **1380passed/6skipped(109.33s)**,另保留早期针对性138passed日志;compileall/diff检查。新用例包含GPU决策等价、失败精确记录/不吞异常、PID重用、诊断写入错误、13,500tick基准命令相等、所有新方向的接缝连续、配对失配拒绝、旧grant拒绝、mock固定顺序及首失败停止。**mock文件明确为CPU MOCK,不是物理证据;CPU不授予Kit资格。**
纯stdlib检查计划,不依赖Kit/Torch、不启动环境:
```sh
/usr/bin/python3.12 scripts/evaluate_cycle_functional_flow.py --check-followup-v3
```
只读旧现场复核(脚本拒绝覆盖已有诊断输出;原留档298文件必须全部一致):
```sh
/usr/bin/python3.12 logs/contact_cycle/functional_followup_v3_prepare01/diagnose_previous.py \
--output /tmp/sf-previous-gpu-diagnosis-NEW.json
```
以下是当时尚未授权的命令模板;实际命令/新SHA及终态已单独留在结果报告。入口必须绝对路径,与FD身份检查一致。当前目录/授权已消费,不得重跑;新任务在持续授权内自主预设护栏并独立登记:
```sh
REPO=/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory
/usr/bin/python3.12 "$REPO/scripts/evaluate_cycle_functional_flow.py" --campaign --followup-v3 \
--output /home/cen/sf-runs/ff-v3-01 \
--approval /home/cen/sf-runs/ff-v3-authority-01/approval.json \
--trusted-approval-sha256 '<NEW-APPROVED-SHA256>'
```
CPU证据根:`logs/contact_cycle/functional_followup_v3_prepare01/`,包含旧现场诊断/输入hash、冻结计划与新源hash、回归日志/XML、冻结原候选核对和最终inventory/source snapshot。本机logs不自动随源码分发。此CPU准备形成时新物理轨迹尚不存在、首失败物理阶段not_evaluated;后续真实执行已完整通过,结果另列,不把CPU准备本身改判为步骤2/3通过。
@@ -0,0 +1,116 @@
# 功能补验 v3 实际结果:u256补验及6组成对连续3周期全部通过
## 结论与逐组表
**本次新u256原命令补验、六组命令的解析/u256对照均完成连续3周期;物理功能与运行完整性全部通过,原campaign success=true,监督器等待退出0。** 原ff-01失败不改写;本次没有失败、重试或训练。
表中“物理”是实际承重→卸载→离地三维移动→落地重新承重,逐周期符合原功能窗口;“完整性”还要求无重置连续性、preclose绑定、全部产物、GPU归属、正常退出及owned清理。
| 固定组 | dx/dy mm | 解析:物理 / 完整性 | u256:物理 / 完整性 | 初态配对 |
|---|---|---|---|---|
| x_bias 16mm | 14.855627 / 5.942251 | 通过 / 通过 | 通过 / 通过 | 通过 |
| x_bias 24mm | 22.283441 / 8.913376 | 通过 / 通过 | 通过 / 通过 | 通过 |
| baseline_direction 16mm | 12.493901 / 9.995121 | 通过 / 通过 | 通过 / 通过 | 通过 |
| baseline_direction 24mm | 18.740851 / 14.992681 | 通过 / 通过 | 通过 / 通过 | 通过 |
| y_bias 16mm | 9.600000 / 12.800000 | 通过 / 通过 | 通过 / 通过 | 通过 |
| y_bias 24mm | 14.400000 / 19.200000 | 通过 / 通过 | 通过 / 通过 | 通过 |
进入矩阵前,原命令`(+25,+20)mm`的**新u256三周期也物理/监督完整通过**(13500采样/2700控制)。每行是同一环境内27s、A→B→A→B,不拼接单周期、不瞬移、不在落点重锚;不同组/控制在各自新进程中初始化。所有13条物理轨迹均`protocol_complete`,无partial-control补齐。
限定对象依旧是**有原辅助条件的浮动根Go2 FL单腿、仿真真值、单开发seed93501**,非整机、无辅助平衡、实机或sim2real。六组位于原组件命令包络的有限内插点,不等于全工作域、扰动鲁棒性或独立验收;`accepted_skill=false`保持。
## 1. 监督失败诊断
旧ff-01已确认拒绝来自live GPU客户集合不属于active owned/已批准桌面graphics的分支。原失败瞬间的客户/PID-birth集合未存,**真实外来、owned身份变化或退出观测竞态仍无法归因**;事后GPU干净不能推翻旧失败。
本次只增加原检查直接消费的NVML列表/owned discovery/known身份、时间和最近16次样本落盘,**原决策、顺序、异常均不变,不增加NVML宽限、不吞失败、不关闭检查**。本次14个Kit子进程全部监督通过,原GPU异常没有复现;这不等于证明或修复了旧故障的具体原因。旧298个留档文件在本次运行后重新核对全部不变。
原输入hash、确切分支与未知项:[运行前诊断/固定计划](contact_cycle_functional_followup_v3.md)。本次`gpu-observations.json`、每行GPU终态/进程身份及父级事后核对见证据根。
**首失败:本次无。** 原ff-01的首个未通过项仍为GPU归属监督门禁,不改判为物理失败或全面通过。
## 2. 功能测量与控制价值
合同`functional_flow_v2.json`原字节SHA256 `16c064228b30f36bbd420b7c2290602bacae6601cac2cec5deb16440f7600625`不变;矩阵计划SHA256 `aaed45884eab9912db6c7bac1d82e2ab27784982c524a838b7886cdd763146f4`在实验前固定,未事后筛选方向/幅度或提高精度。
以下仅统计六组矩阵,每种控制18个周期。W≈52.8597N;卸载为末0.1s相对承重末窗口的载荷比,重新承重为末窗口平均Fn/W。
| 真实测量/门槛 | 解析零残差 | 解析参考+冻结u256 |
|---|---:|---:|
| 完整运行 / 周期 | 6/6;18/18 | 6/6;18/18 |
| 承重/重新承重接触比例最低(≥80%) | 100% / 100% | 100% / 100% |
| 承重末平均载荷最低(≥0.5W) | 0.999988W | 1.000007W |
| 卸载剩余比例最高(≤0.8) | 0.67577 | 0.48135 |
| 摆动真实腾空比例/连续时长 | 100% / 1.4s | 100% / 1.4s |
| 摆动实际Fn | 全部0N | 全部0N |
| 摆动gap范围 | 43.46~54.26mm | 47.45~58.12mm |
| 有向XY每轴进度最低(≥50%) | 68.38% | 53.09% |
| 重新承重末平均载荷最低(≥0.5W) | 0.98820W | 0.97448W |
| 每周期平均落点误差范围(≤50mm) | 13.52~13.82mm | 8.75~9.68mm |
| 根XY最大值(仅诊断,非30mm门槛) | 80.49mm | 68.98mm |
- 解析模式真实请求/有效残差为0,actor不参与控制;不能说学习从零学会了整套流程。
- u256权重不变,SHA256 `c088bb3aecaca751867118858626fd508f7a3983303802d7524353f80802cf99`,确定性`tanh(actor_mean)`+原解析参考;实际有效残差峰值0.047431rad,非独立策略。
- **u256没有提高该矩阵的功能通过率,解析本身已足够。** u256落点/根XY测量较小,但小X命令的进度余量更低(最差53.09%对68.38%);不据这些单seed描述性结果宣称必要性或统计优越。当前功能目标不需要追加训练/RootXY精度优化。
实际世界目标与预先固定的协议逐13500tick独立核对,最大绝对差0;没有改命令或删掉不利样本。全部13条轨迹/39周期的完整**234个相位**Fn min/mean/max、gap、XYZ净位移及漂移见`analysis/actual-phase-summary.csv`;每周期窗口、控制归属及全部布尔门槛见`analysis/actual-summary.json`和13份`*-audit.json`。
## 3. 初态、连续性与保护
- 每组解析在首rollout前保存初态;u256在其首rollout前精确核对该记录。六对初态文件hash一致,包含原初态张量、RNG/枚举carry-state、IK/action/refresh、观察、固定anchors与命令。没有失配后恢复/修复或比较失配行;仍不宣称完整PhysX内部缓存相等。
- 全部13轨迹各为global tick1..13500,完整500Hz采样、100Hz控制;两次接缝各45项枚举状态指纹相同,IK各正常刷新一次。所有初始化后reset/pose写入无增量(初始化API计数22/14,不冒称初始化也没有reset)。
- 原相位截止、初次接触获取、提前分离、碰撞/穿透、数值、关节/力矩/速度、根高度/倾斜、辅助/guide保护均保留。辅助条件/装置/重力/质量/时长/权重不变,无外扰、噪声或形态变化。
- 所有13轨迹:归一化执行器比最高0.215740、关节超界0、最小gap−0.0298mm、最大高度误差67.806mm(门槛100mm)、最大倾斜0.057871rad(门槛0.6);竖直guide0,无数值异常/失控或执行器越界。根XY>30mm如实记录,原严格P1失败不被重评为通过。
## 4. 授权、预算与清理
用户本轮批准固定计划,并明确后续预算无需逐次请批;[持续授权边界](execution_authority.md)保留每次自主预设护栏、零训练、首失败停止和独立登记。本次没有取消任何已经固定的上限。
| 项目 | 实际 / 上限 |
|---|---|
| Kit启动 | 1配置准备+13串行物理;无重试/恢复 |
| 共享物理API调用 | attempted=completed=**175500**;初始化额外0;保留上限178750不混为实测 |
| 完整模拟 | 39周期,351s仿真时间,13×27s连续轨迹 |
| 总时长 | **3585.201s / 6540s**,约59.75分钟,含配置与清理 |
| 设备/优化 | RTX5080,1环境,优化/训练0 |
| own逻辑字节 | 父核对时run根1440699387+authority约数MB,合计低于2GiB;最终逐文件inventory另存 |
| native正增长 | 1105834619字节;与own含重叠,不可重复相加;非硬配额、覆盖不完整 |
| OS退出/进程 | 14子进程均等待退出0,监督器也由外层driver实际wait退出0;全部owned PID/birth消失 |
| GPU/无关进程 | 无compute/任务graphics残留,原7桌面graphics客户身份均保持不变;没有终止无关进程 |
| 身份 | 实际同对象cfg由每worker preclose核对;父退出后重哈希450个有限绑定文件、当前源及原u256通过,不以新建cfg冒称同对象检查 |
原u256/134原快照的只读核对继续通过(沿用既有AGENT非运行时例外,不修改旧逐字节验证器/manifest或抹去旧AGENT失败)。启动前CPU1380passed/6skipped;运行结束后接触周期CPU回归再次**1380passed/6skipped(109.86s)**,日志/XML在authority根。本次真实Kit/物理与CPU资格明确区分,未宣称全仓全依赖资格。
## 5. 证据与复现命令
证据是本机文件,不自动随源码分发:
- `/home/cen/sf-runs/ff-v3-01/`:新approval/预算保留、0物理准备与独立审查、原源快照、13条`trace.pt`/`initial-state.pt`/`initial-pair.json`/`events-and-seams.json`及每行worker/supervisor结果;`campaign-result.json.success=true`。
- `/home/cen/sf-runs/ff-v3-authority-01/`:当前用户原指令、实际绝对入口命令、GPU初态、driver实际wait退出、`parent-postexit.json`、旧失败298文件未变证明、原冻结核对、完整离线`analysis/`、状态图/GIF与最终inventory/source snapshot。
- 本次准备绑定SHA256 `043cc8776ecae4ef27449d98165e1c5e3d9b73dc00ee41cb387dc822fcfe3e6a`;numerical inventory SHA256 `770f2d923777fcf4ed3a993b86347ee27d1cf49f3beef0b61c58fcb8114d2bd0`。与旧inventory相同不代表复用旧配置绑定或授权。
实际命令(**存证,旧目录/grant不得再次执行**):
```sh
REPO=/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory
/usr/bin/python3.12 "$REPO/scripts/evaluate_cycle_functional_flow.py" --campaign --followup-v3 \
--output /home/cen/sf-runs/ff-v3-01 \
--approval /home/cen/sf-runs/ff-v3-authority-01/approval.json \
--trusted-approval-sha256 a8310a9706ac54be7e2c5b3b34fb29af46bf0a24e3481197807822f469f46d00
```
仅重算已有结果,无Kit/物理/训练(输出必须不存在):
```sh
CUDA_VISIBLE_DEVICES='' /home/cen/miniconda3/envs/linker_sim_isaac/bin/python \
/home/cen/sf-runs/ff-v3-authority-01/analyze_actual.py \
--root /home/cen/sf-runs/ff-v3-01 --output /tmp/sf-v3-analysis-NEW
```
离线`runtime_continuity_verified=false`是张量审计自身的有意限制,不能单独证明无状态写入;本次运行期证明来自原hooks、seam、绑定和监督终态,而非离线工具提升判定。新实际执行仍按持续授权自主设定新护栏/目录/登记,不复用本次grant,当前任务已完成,不自动扩大范围。
可视化为**实测根/关节的FK+实测足端/接触力**,不是Kit摄像机录像或新物理;完整500Hz轨迹未抽稀,只有显示采样10fps:
- [新u256原命令三周期GIF](file:///home/cen/sf-runs/ff-v3-authority-01/nominal3-visuals/measured_state_animation.gif)
- [y_bias16mm解析状态图](file:///home/cen/sf-runs/ff-v3-authority-01/y-bias16-reference-visuals/measured_state_keyframes.png) / [GIF](file:///home/cen/sf-runs/ff-v3-authority-01/y-bias16-reference-visuals/measured_state_animation.gif)
- [同组u256状态图](file:///home/cen/sf-runs/ff-v3-authority-01/y-bias16-u256-visuals/measured_state_keyframes.png) / [GIF](file:///home/cen/sf-runs/ff-v3-authority-01/y-bias16-u256-visuals/measured_state_animation.gif)
+31 -1
View File
@@ -1,6 +1,36 @@
# 接口与运行约束:Go2 FL参考+残差
## 当前契约
## 当前0.4.1:真实GUI与headless分开
[0.4.1实际结果](contact_cycle_sim_gui_v041_result.md):公开`start(..., gui=True)`/CLI `--gui`贯穿approval、监督和认证worker,默认仍headless。独立包`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.1/`中`bash demo_gui.sh`真实解析S1开窗;原`demo.sh`和`verify_functional.sh`保留。当前源码GUI/headless S1各六周期原物理/完整性通过;活跃GUI提前关闭2040tick按预期中止,exit2不算完成。
GUI仅新camera/light/UI,原负载及控制/物理/辅助不变;完整后只render保留原场景(最多10分钟),关闭后才完成身份/wait/清理。真实截图和原始录像/原速剪辑在包demo,不是FK动画,Kit约2–3FPS、不保证墙钟1×实时。原0.4.0全部字节/校验保留;0.4.1未重验其完整矩阵/u256或GUI安全停止/超时/极早启动关闭,不继承新资格。所有旧严格失败/accepted_skill=false保持。
---
## 历史可调用仿真功能版0.4.0
新公开接口/CLI已真实仿真验证,见[API及生命周期](go2_fl_skill_v4_api.md)与[本轮实际结果](contact_cycle_sim_skill_v4_result.md)。`scripts/go2_fl_skill.py`提供start/query/queue/request_stop/wait及authorize/validate/run/status/queue/stop;复用原解析IK/接触/保护,u256仍为可选冻结残差,不是通用adapter。
独立包`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.0/`:`bash demo.sh`真实启动S1;`bash verify_functional.sh`串行真实验证20周期/三序列/u256/停止拒绝。每次新有限登记/新grant/新目录;本轮当前登记已消费,不重跑旧例名。
原名义解析20周期、S1/S2/S3各6周期、u256 S1、边界停止/非法拒绝、独立目录公开CLI和外部实时Python队列均通过。目标为固定世界A去/回成对,命令仅A边界生效,不累积或测量重锚。保持原辅助/真值/单seed限制;30mm根XY只诊断,历史accepted_skill=false/严格失败不改写。下列v2/v3及冻结v1约束作为复用/历史继续保留,不取代当前API说明。
---
## 本轮独立功能版本(先于下列冻结v1合同阅读)
当前优先级已由RootXY精度优化改为完整功能流程。新判据与预算见[功能流程v2](contact_cycle_functional_flow_v2.md),配置`configs/contact_cycle/functional_flow_v2.json`;原v1/严格P1/P2/RootXY配置和失败结果不变。30mm根XY在新功能版本中只记录,旧严格验收仍按原门槛判失败。
新增`legs/contact_cycle_sim/functional_flow_v2.py`只准备独立协议,复用整数时钟/固定A-B-A-B命令;原相位终止和全部安全保护初版不动。本轮复用无autoreset的`EvaluationContactCycleEnv`,两次接缝禁止reset、pose写入、测量落点重锚及拼接。分别报告零残差解析控制与解析参考+原冻结u256确定性残差。
`scripts/audit_cycle_functional_flow.py`是CPU只读审计,输出`trace_functional_passed`,**不授予运行许可、不证明运行期连续性**;全局时钟重启、失败标记及不足3周期均不能通过。实际受监督入口现为`scripts/evaluate_cycle_functional_flow.py`,新绑定在`functional_flow_binding_v2.py`,同步完整记录/接缝核对在`functional_flow_recording_v2.py`。它们使用独立用户批准,不通过旧P1/P2/RootXY授权绕过门禁。
本轮新授权已执行且消费:解析单周期及连续3周期连同监督检查通过,u256单周期通过;u256三周期物理完成但监督器GPU归属检查失败,**原campaign失败保持**。实际36000物理调用/初始化额外0/优化0,无重试;参见[完整结果与启动命令](contact_cycle_functional_flow_v2_result.md)。该次运行后未重试,不追加RootXY试训。
用户新批准的[功能补验v3已实际全部通过](contact_cycle_functional_followup_v3_result.md):同入口`--followup-v3`、新协议/绑定及原检查决策不变的GPU观测;新0物理准备后,先u256原命令3周期完整通过,再六组解析/u256成对3周期全通过,6对初态一致。175500共享调用、39周期、3585.201s、0训练/0重试;GPU/退出清理正常,旧失败仍保留、未查明具体原因。源码/权重/功能阈值/辅助不变,双方功能通过率6/6,不声称学习必要或全域鲁棒。用户另授[持续预算权限](execution_authority.md),后续自主预设护栏并新绑定/独立登记,不再逐次请批、不复用旧grant。原计划与CPU准备为[历史记录](contact_cycle_functional_followup_v3.md)。
## 原冻结v1契约(历史与复用边界保持)
- 任务:`SkillFactory-Go2FL-ContactCycleSim-v1`。
- 特征:`contact-cycle-sim-truth-residual-46-v1`;46D、float32、有限值,不能载入旧32D策略。
+23
View File
@@ -0,0 +1,23 @@
# Go2 FL 0.4.1实时GUI交付固定计划
2026-10-16用户授权:仅补可视化,不训练/精度优化/扩展控制范围。0.4.0目录、压缩包、SHA及实际证据逐文件保留;新版本0.4.1,不覆盖旧版。原功能参数、500Hz/100Hz、S1成对固定世界目标、保护和全率轨迹不改。
实现:公开API/CLI显式gui参数,经精确spec/approval/FD传给worker。PREP保持无窗口零物理,RUN gui=False保持headless;gui=True打开Kit窗口,用原真实物理环境。相机/灯光/只读HUD显示解析/冻结模式、真实相位、周期/命令/接触载荷/间隙。使用资产原有蓝色3kg刚体载荷,不新增代理物理。GUI约25Hz显示,保留原render_interval20,不改物理参数。完结后禁止额外物理步,保留最终画面和响应(最多10分钟有界查看);关闭窗口/退出按钮正常结束。提前关闭记录failure/gui_window_closed_before_completion,保存部分证据,不提升完整通过。
固定实测:完整GUI解析S1六周期(从独立0.4.1目录公开入口真实运行);完整无窗口解析S1六周期回归;另一次GUI在首周期提前关窗,验证中止/证据/非零状态/清理。开发必要GUI检查可保留为独立新目录,不覆盖失败,最多两次有证据修复同一问题。GPU串行,复用原有限新登记/24Kit/650000调用/24GiB/12h监督护栏,预计6~10Kit、约60000~108000调用;零训练,不消耗余量追求精度。
桌面实测:记录实际Kit窗口X11可见信息及截图(包含真实渲染和HUD),录制短mp4的真实窗口连续屏幕帧(非离线重建/预录),至少覆盖承重、卸载、腾空移动、重新落地及命令切换;最终画面/无物理增量查看和关闭窗口单独验证。观察结果与全率物理审计分开报告,未实测项目明确标注。使用真实桌面DISPLAY,不启动第二个无关Isaac实例;不终止非owned进程。
## 2026-10-16实测后必要的监督收尾修正(不改GPU判定)
前两次PREP暴露无窗口配置没有加载viewport/UI;保存原源码和失败后,将展示模块改为无native依赖的CPU安全导入,仅GUI PREP/RUN显式加载同一UI依赖。第三、第四次PREP真实完成且退出0,但在退出窗口里,NVML见已登记owned CUDA PID,约30ms后的/proc发现该PID已消失,原严格GPU判定使campaign失败。两次均零环境/零物理,无RUN,原失败保留;不据此解释任何旧失败。
采用显式LIVE→CLEANUP交接,而非容忍NVML未知/消失PID:v4 worker完成env.close、hook退出、preclose核对并写最终worker receipt后,继续保留owned GPU进程,最多30秒等候其精确receipt SHA的release。父观察ready后先执行**原样严格live GPU检查**,通过才发布release;随后进入受原deadline/存储/native/stdout约束的有界关闭等待,必须实际wait并通过**原样严格final GPU无残留检查**。handoff缺失/摘要改变/超时或任何原GPU检查失败都不能通过。不修改GPUOwnership/ObservedGPUOwnership、allowlist门槛、物理门槛或控制数学,不把已消失PID冒充active,不重试任何失败GPU查询。该修正需CPU负例及独立复核后才继续GUI实际验证。
GUI实际构造继续遵守原来源门禁:无窗口PREP须预加载GUI经验默认存在的`omni.replicator.core`(仍无配置采样,不改变可选包判定),以及构造相机时才导入的`isaaclab_physx.renderers.kit_viewport_utils`及其父包。第8次诊断的完整绑定差异只有这两个imports条目,effective_config无差异;把确切依赖纳入PREP,而不是删去比较或放行失配。诊断失败原件保留,所有实际物理调用仍为零。第5次另有未知graphics PID在admission之后出现,原监督正确中止,不把它豁免或归入已批准桌面进程;后续新admission仍仅原7个桌面graphics PID,不在运行中扩白名单。
本机AppLauncher没有写旧`/isaaclab/has_gui`标志,但实际已打开窗口并构造KitVisualizer;该遗留位不是窗口真值。GUI attach应检查真实`SimulationApp.config.headless=False`及具体运行中的KitVisualizer,不更改该遗留位(PhysX会据此修改scene-query配置)。保持物理cfg无差异,实际桌面可见性继续用外部X11截图/录像验证,不凭布尔位认定可见。
实际完整GUI第10次:27000调用/6周期,真实OS关闭,最终保留状态及计数完全相等;但退出前原绑定检查发现首个render才加载`isaaclab_physx.renderers.isaac_rtx_renderer_utils`,整次仍失败,不作正式合格。将这个确切依赖也纳入PREP(保留比对),不改cfg/物理参数。截图工具原缓存启动时960高,后来Kit窗口纵向展开,原短录像裁掉腿,保留但不作合格证据;外部捕获改为刷新几何,录制时仅缩放自有窗口,HUD默认移到右侧避免缩小时覆盖原蓝色负载。必须新目录重新完整验证。
交付demo_gui.sh一条启动GUI;demo.sh与verify_functional.sh保持headless含义;README区分实时GUI、无窗口验证及历史离线动画,给准确命令、依赖/范围/结果、截图mp4路径与校验和。独立目录、原始监督/中止证据、CPU回归、文件SHA、压缩副本及当前状态/索引更新后结束。
+45
View File
@@ -0,0 +1,45 @@
# 0.4.1真实GUI交付结果(2026-10-16)
**物理资格与GUI资格分别通过,范围仅本轮当前源码解析S1。** 原0.4.0不变;不是严格accepted_skill、精度优化、训练或整机/实机资格。
| 当前源码真实测试 | 物理/完整性 | GUI/退出 | 原判定 |
|---|---|---|---|
| `gui-complete11`,解析S1 | 6周期/27000tick,5接缝,逐周期原功能门槛及完整绑定通过 | 可见Kit、原载荷/足端/地面、实时HUD;最终保留约5秒后真实WM_DELETE;carry/计数不变、wait0、GPU清理 | success=true |
| `headless-S1`,解析S1 | 6周期/27000tick、原功能门槛及完整绑定通过;70个轨迹张量与GUI逐位相同(仅这两次同seed实测) | 无窗口;wait0及清理通过 | success=true |
| `gui-abort01`,运行中提前关窗 | 2040tick部分轨迹、0完整周期;无物理失败标志,无额外初始化步 | 原因`gui_window_closed_before_completion`;wait2、窗口/owned/GPU消失 | success=false,符合中止测试预期;不提升为完成 |
## 实证与原件
交付根`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.1`;全部原始运行根`/home/cen/sf-runs/skill-gui-v041-*`,审计/范围根`/home/cen/sf-runs/simskill-gui-v041-authority-01`。实际由独立包公开CLI/demo_gui.sh启动,记录binding.root来自该包,项目源不从开发仓库加载。GUI11/无窗口/中止父重哈希分别464/457/464个有限绑定文件,无失配;全部2143原保护文件无变化。
- 真实截图和35秒原速跳切剪辑见包`demo/`;原始240秒桌面视频及剪辑时间段保留。截图/录像不是FK重建,运行GUI不读取这些影像。录制前仅外部缩放自有X11窗口,物理/机器人位姿未改动。
- GUI实际控制、载荷、接触/保护均原实现:500Hz物理、100Hz控制、render_interval20。实际墙钟GUI约2–3FPS,不保证1×实时,录像12FPS只是采样。当前两次完整GUI/headless轨迹70张量全部逐位相同;PREP有效cfg删除唯一视觉`sim.visualizer_cfgs`字段后完全一致。
- 原解析S1指标:根XY最大80.4934mm(仅诊断)、高度误差51.2482mm、倾角0.0552400rad、最小gap -0.0297930mm、执行器比0.214642、关节超界0、竖直guide 0。所有6周期原功能窗口及固定世界目标逐tick核对通过。**不以这些结果改写旧30mm严格失败。**
- 初始化API reset/pose为22/14,rollout无增量、初始化物理调用0。完整运行27000调用;保留最终场景时只render,枚举carry及hook计数不变。不是隐藏PhysX完整状态快照。
- GUI11外部观察器在已WM_DELETE、已取得两张最终保留图后,对刚销毁的同一X11窗口执行xprop而报错;原异常未抹去。原worker/监督success=true、调用方实际wait0,补充独立检查证实窗口/PID消失、严格GPU清理通过。该补件不覆盖失败worker判定,见`gui11-observer-supplement.json`。
## 全部尝试,不删除或翻案
| 尝试 | 原状态/原因 | 实际物理调用 |
|---|---|---:|
| 01、02 | PREP缺viewport/UI模块;改为GUI PREP/RUN有限显式启用/导入 | 0 |
| 03、04 | PREP退出间GPU观测/进程发现竞态;原判失败,确切最终收据SHA handoff修复,不放松GPU决策 | 0 |
| 05 | 未知graphics PID导致原严格检查拒绝,未豁免,原7桌面名单未扩展 | 0 |
| 06 | GUI经验已有Replicator而PREP未显式载入,可选模块可用性失配 | 0 |
| 07、08 | 构造相机晚加载两个renderer helper;第8次仅加诊断,cfg无差异,确切imports纳入PREP | 0 |
| 09 | 已有真实非headless App/KitVisualizer,但SDK未写遗留has_gui位;改查真实对象,不设置会改scene-query cfg的遗留位 | 0 |
| 10 | 6周期物理完成,退出前发现首个render延迟导入RTX utility,原完整性失败仍保留;确切依赖纳入PREP | 27000 |
| 11 | 当前源码真实GUI完整成功 | 27000 |
| headless-S1 | 当前源码无窗口完整成功 | 27000 |
| gui-abort01 | 真实提前关窗,按预期中止,原success=false | 2040 |
所有旧失败的批准源码快照逐项与各自approval核对,无失配。显式PREP依赖不调用渲染/物理函数,不删来源检查;原GPUOwnership及observed wrapper决策源码与0.4.0相同。handoff后只有有界关闭,实际wait、owned清理及原strict final GPU检查仍强制。无NVML重试/宽限、未知客户端豁免或静默降级。
## 护栏与范围
本轮新范围开始2026-10-16 09:51:52 +08,12小时截止21:51:52;旧0.4.0授权此前已结束,未复用。每次新有限会话登记,父级额外跨会话汇总并串行执行,不把每会话上限当无限总预算。**21/24 Kit,354250/650000预留调用,83040实际调用(含失败10和预期中止),18个实际完成时钟周期中只有12个为合格完整周期;优化0。** 不退失败扣款、不使用剩余预算追加仿真。最终存储/耗时与压缩包SHA见外部authority的task-complete.json(压缩后生成,不自包含于包内)。
当前CPU回归1514passed/6skipped;独立源码复核无问题,CPU不授予GUI/物理资格。只做解析S1 GUI、headlessS1和活跃窗口中止;0.4.0已验完整矩阵不自动升级为0.4.1资格,旧权重不变但u256仍需当前源四条解析前置。
未实测:AppLauncher返回前极早关窗、GUI公开安全停止、10分钟自动超时、0.4.1 u256/GUI、其余GUI命令/20周期、多seed/全域/无辅助/整机/实机/sim2real。native采样监控覆盖有缺口,不是硬配额。目录/压缩包/校验分版;0.4.0 archive SHA256仍`c2ce36fee1109c785aa7edc5aca48c96f56d82b0102a15def1f2af4e08c4226c`。
@@ -0,0 +1,57 @@
# 0.4.2 application-loop / live-GUI performance result
Simulation only, **accepted_skill=false**, zero training/search. Qualification scope is current-source **analytic S1**, not a new complete matrix/u256 or hardware qualification. Original0.4.0/0.4.1 deliveries and historical evidence remain sealed.
## Measured result
| Case | Protocol | cProfile | Active loop wall s | Physics ticks / wall s | Render submissions / wall s | Sim seconds / wall s |
|---|---|---:|---:|---:|---:|---:|
| baseline-profile02 | x_bias16,1cycle/4500ticks | on |113.5415|39.633|1.982|0.07927|
| sensor-profile03 | same | on |92.1030|48.858|2.443|0.09772|
| combined-profile04 | same | on |75.7835|59.379|2.969|0.11876|
| display-S1-05 | S1,6cycles/27000ticks | **off** |764.8630|35.300|35.300|**0.07060**|
| headless-S1-06 | same S1 | **off** |294.6793|91.625|0|**0.18325**|
Startup, reset and final viewing are excluded from these active-loop times. The like-profiled same-seed/same-protocol baseline→combined comparison is **1.4982× throughput /33.25% lower wall time**, not a claim that every workload is1.5× faster. All three4500tick trace files are identical SHA256 `2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94` (authoritative digest in performance-result.json).
The genuine moving Kit-window capture at tick2519 shows **33.98 FPS**. The full S1 loop submitted27000 current-state renders, including25650 added renders and the unchanged1350 interval20 renders. Submissions35.30/s are **not independently measured display-present timestamps**. The120s original-speed owned-X11-window video samples at30Hz; that recording rate is not the renderer FPS. GUI measurement includes recording and early concurrent CPU regression-test overhead.
Whole-device active-loop sampling: GPUutil mean**39.74%**, power**133.43W**, SMclock**2887.82MHz**, unchanged powerlimit360W. These include baseline desktop clients; not process-exclusive metrics. No artificial workload, extra environments, GPU power/clock cap or frame interpolation was used.
**Important:** fluid GUI is not1× real-time physics. With a render per verified new state, the GUI run advances only0.07060sim-seconds/wallsecond (54sim-seconds take764.86s). Extra rendering deliberately trades physical throughput for responsiveness. Headless is faster at0.18325×, still below1×. Single-environment Python/CUDA small-operation and mandatory synchronous checking overhead remains; many-environment training amortizes that overhead and its aggregate throughput is not comparable. No GPU100%/1× promise is made.
## What changed
- Bound cProfile/perf-counter/runtime metadata observation and device-wide read-only nvidia-smi sampling.
- Epoch sensor proxies avoid repeated getters within one scene-update/reset epoch; update/reset delegates to the original guarded SDK instance and invalidates even on zero-dt/errors.
- Exact device/dtype-grouped synchronous snapshot copies; all original expressions/full500Hz samples remain, with independently owned CPU fields.
- Read-only accessor reuse is confined to ONE original safety or snapshot invocation, restoring original methods/presence in finally; never caches physics/control/latch/history/RNG updates.
- The same six finite-difference FK evaluations are batched; epsilon, IK iteration/solve/limits unchanged.33sample strict float32-representation qualification passes on this native CUDA stack and is exactly bound between PREP/RUN. This is not an all-input proof; inexact samples explicitly retain the original Jacobian.
- `combined_display` draws each new verified physical state after safety and synchronous evidence. Original configured render_interval20 remains; those existing renders are not duplicated. No physical step, time scaling, teleport, reset or target-domain change is added.
- A specific expected Kit instance must remain running/notclosed/notpaused before/after added rendering, before subsequent physics, after controls/original interval20 renders and during/following final viewing. SDK-swallowed removal failures fail closed, not as successful paints.
## Correctness evidence
- GUI and headless original campaigns/workers succeed:6cycles27000actualticks each,5seams each, preclose identity verification, unchanged original functional gates, actual callerwait0 and strict shutdown handoff/cleanup.
- Both70tensor complete native S1 traces are byte-identical to each other **and prior0.4.1 GUI11**. All three whole-file SHA256: `d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba`.
- GUI/headless effective-config differences are confined to `/fields/mapping/sim/fields/mapping/visualizer_cfgs/list`;41authority inputs and30base binding-source inputs match between the two runs.
- Both independent zero-step CPU physical audits pass all6original functional windows and exact fixed-world targets. This audit is separate from original runtime continuity/integrity verification.
- Original physical/controller/protocol files and thresholds unchanged;500Hz simulated physics/100Hz controls retained;27000rollout calls,0initialization physics,22constructor reset calls/14pose writes, no rollout reset/reanchor. Genuine retained final scene: carried-state/counters equal and hold_no_physics=true; actual WM_DELETE followed by wait0/window absent.
- CPU full suite **1557passed/6skipped**. Targeted77passes include signed-zero representation gate, epoch update/reset/error boundaries, snapshot expressions/ownership, per-tick failure/evidence ordering, swallowed visualizer removal, original interval20 failure before next physics/final completion, closed/paused/replaced visualizer, headless no-painter and GUI-close semantics.
- Physical diagnostics unchanged: RootXYmax0.080493420m, height error0.051248193m, tilt0.055240028rad, gapmin−0.000029793009m, effort/velocity ratio0.214641616, joint excess0, guideZ0. **Historical strict30mm RootXY failure is not fixed or reclassified.**
## Failures / evidence limits retained
- baseline-profile01 fails PREP before physics because SimulationApp lacks get_app_version; corrected to omni.kit.app.get_app. Its approved source and original failure remain.
- Review findings (numeric-vs-bit equality; SDK-swallowed renderer removal) were fixed before their relevant native runs; reports and negative tests retained. First presentation CPU collection failed on absent pxr; only the CPU harness import was corrected, not a runtime fallback.
- First independent headless06 postexit observer raises GPU admission mismatch without reporting PID sets. **Exact transient cause is unknown.** Original campaign remains its original success. After actual callerwait0 and a read-only clean nvidia-smi observation, the unchanged strict verifier passes a second observation. No new Kit/physics, whitelist expansion, foreign termination or retroactive explanation. Both observations remain evidence.
- New-cadence native early-abort, injected visualizer-failure, startup-close, GUI-safe-stop and600s automatic viewing-timeout are not exercised; related CPU negatives are not native acceptance. No current-source nominal20/S2/S3/u256 qualification, global determinism, whole hidden PhysX-state completeness, hardware/sim2real or generic adapter claim.
- Shared/native cache monitoring remains sampled and incomplete, not a hard filesystem quota. SO/AV standalone packages unavailable; no unrequested USD operation/install performed. This is application optimization, not USD geometry tuning.
## Artifacts / accounting
Root `performance_v042/`: original six attempts,41-input approved source snapshots, original receipts/traces, profiles/GPUCSV, genuine GUI PNG/video, audits, review findings/resolutions, CPUlogs/XML and `performance-result.json`. No large raw logs are presented as substitutes for compact evidence.
Task totals before packaging: **11/12Kit,73500/120000reserved calls,67500actual calls,0optimizersteps**,57owned process identities checked/none alive; source41 qualifies12cycles, prior three successful one-cycle benchmarks add3cycles. Protected4427historical files rehashed unchanged in postexit checks. Final inventory/protected/process checks and delivery sealing remain separate receipts; elapsed/storage final accounting must be read from task-complete.json. Delivery is `performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2/` with adjacent tar.gz/sha256; actual source41bytes equal the recorded staging qualification, without claiming another GPU run at the relocated delivery path.
The first relocated package CPU suite found39failures (36missing historical storage document fixture,3missing saved u256 validation trace),1517passes/7skips. After adding those immutable original fixtures,1556passed/7skipped; the additional skip was missing recorded P0replay inputs. Original P0initial/episode and P1failuretrace are now embedded, with ONLY the bundled test's P1fixture locator made relative; original assertions/hash/tolerances and all41runtime authority inputs unchanged. Final relocated package suite **1557passed/6skipped**; source/frozen validation passes. All three CPU receipts and repair inventories remain; these historical test fixtures are not a live authority, training run or promotion of old P1failure.
+39
View File
@@ -0,0 +1,39 @@
# Go2 FL 仿真功能版 v4:运行前固定合同
开始:2026-10-15 17:38:07 +08:00;总截止:2026-10-16 05:38:07 +08:00(最多12小时)。最后至少2小时保留修复复验/打包。零训练、单RTX5080、单环境、GPU串行。旧严格验收、accepted_skill、历史release及失败证据不变。
## 固定功能判据
逐周期沿用 `functional_flow_v2.json` 与旧运行期物理保护,不能放宽:承重/重新承重末窗口接触≥80%、平均Fn≥0.5W;卸载末0.1s剩余≤承重的80%;真实摆动gap≥10mm、Fn≤2N占比≥80%且连续≥0.1s;XY每轴有向进度≥命令50%;落点3D误差≤50mm;根高度误差≤100mm、倾斜≤0.6rad、穿透≤2mm、关节/速度/力矩/数值/辅助/guide原保护全部保留。根XY及落点误差完整记录,30mm根XY不是硬门槛。每个实验首次异常立即停止并保留原件。
连续性:全局500Hz tick连续;100Hz控制;仅初始化reset;周期间无reset/pose写入/重锚/拼接;每个接缝枚举承接状态不变及一次参考刷新。正常停止只在当前周期已落地重新承重的终点,不强求立即停止。非法命令验证先于任何队列/控制状态修改。
## 固定命令/世界坐标规则
复用已通过三方向 `x_bias=(25,10)`、`baseline_direction=(25,20)`、`y_bias=(15,20)`,幅度16/24mm;另保留原名义(+25,+20)mm用于20周期。单位m,方向世界XY;无随机化/外扰。
A为初始化时已有命令的固定世界地面目标,不取测量落点;根目标保持原初始命令挂载点。每对周期 A→B(c)→A,B(c)=A+归一化方向×幅度。返回周期沿用本对的B,禁止切换时跳变或累积目标。新命令仅在A处的完整周期边界生效(0、2、4…周期),非该边界请求可排队,当前对不修改。所有未来B始终相对同一A,不依赖实测足端。停止可在任一完整周期终点,不改变目标/物理状态。
预先固定三条各6周期序列(每项执行去/回2周期):
1. S1:x_bias16 → baseline_direction24 → y_bias16。
2. S2:y_bias24 → x_bias24 → baseline_direction16。
3. S3:baseline_direction24 → y_bias24 → x_bias16。
解析必做:原名义20连续周期、S1/S2/S3各6周期;全部通过后u256执行相同S1。另做安全停止(申请≥3周期、首个周期中请求,边界1周期结束)及非法命令拒绝真实运行证据。独立交付目录通过公开接口复现S1。前置CPU/配置准备不等于仿真通过。
## 自主资源与失败护栏
每个物理实验使用全新目录/授权/当前源码与配置绑定,不沿用旧grant。复用现有归属、超时、物理调用、存储、preclose、wait/cleanup监督,不建设通用adapter或监督重构。
- 单次20周期上限90250物理调用(90000 rollout+250初始化)、2400s含120s清理。
- 单次6周期上限27250调用、1200s含120s清理。
- 单次停止实验上限13750调用、900s含120s清理(实际应1周期)。
- 配置准备每次最多1个Kit进程、0物理、360s。
- 每个实验owned产物≤4GiB,全部新任务owned≤24GiB;磁盘自由空间保持≥8GiB。native监控仍非硬配额、覆盖缺口按原协议报告。
- 本轮最多24次准备/物理Kit启动,最多650000物理调用(上限不是实测);训练/优化0。总截止优先于单次上限。
- 同一问题最多2轮有依据修复复验,每轮新版本/新目录;不能盲重试或删失败。扩展项仍失败则保留通过能力交付并明确未完成。
## 分工与交付
父负责固定判据、整合、所有GPU运行/验收/打包;子agent仅并行只读审查或CPU测试。独立版本本地仿真包带公开Python/CLI、源码/配置/资产或hash依赖、冻结权重、说明/示例、功能证据与校验;必须从该目录实际仿真复现。切换演示以实测状态重建动画交付并标识非Kit录像,若另录Kit须分开标注。
+85
View File
@@ -0,0 +1,85 @@
# Go2 FL 接触周期仿真功能版0.4.0:已交付并实际复现
**可调用Python/CLI、本地独立技能包、解析20连续周期、三条切换序列及独立目录真实仿真均通过。** 冻结u256同S1补验也通过。不是旧严格验收或accepted_skill发布;`sim_functional_accepted=true`、历史`accepted_skill=false`不变。零训练、零物理失败/重试,未修改旧发布或失败证据。
## 1. 实际通过表
全部轨迹为单次环境内完整连续运行;物理=逐周期承重→卸载→真实离地移动→落地重新承重;完整性=原保护、无reset/pose增量、承接接缝、绑定、GPU归属、实际wait退出0及owned清理。
| 实验 | 控制 / 完成周期 | 物理 / 完整性 | 实际物理调用 | 终止 |
|---|---|---|---:|---|
| nominal20 | 解析 / 20连续 | 通过 / 通过 | 90000 | protocol_complete |
| S1 | 解析 / 6连续 | 通过 / 通过 | 27000 | protocol_complete |
| S2 | 解析 / 6连续 | 通过 / 通过 | 27000 | protocol_complete |
| S3 | 解析 / 6连续 | 通过 / 通过 | 27000 | protocol_complete |
| u256-S1 | 冻结残差+解析 / 6连续 | 通过 / 通过 | 27000 | protocol_complete |
| stop | 解析 / 请求3、完成1 | 通过 / 通过 | 4500 | requested_stop,非3周期完成 |
| invalid | 解析 / 1 | 通过 / 通过 | 4500 | protocol_complete |
| independent-S1 | 独立包公开CLI / 6连续 | 通过 / 通过 | 27000 | protocol_complete |
| independent-live | 独立包Python接口+外部实时命令 / 6连续 | 通过 / 通过 | 27000 | requested_stop |
矩阵在首次执行前固定于[本轮计划](contact_cycle_sim_skill_v4_plan.md),原功能门槛来自未改的`functional_flow_v2.json`。S1/S2/S3每项占去/回2周期:
- S1:x_bias16 → baseline_direction24 → y_bias16。
- S2:y_bias24 → x_bias24 → baseline_direction16。
- S3:baseline_direction24 → y_bias24 → x_bias16。
三方向世界XY比为(25,10)/(25,20)/(15,20),幅度16/24mm;nominal20为原(+25,+20)mm。A只取已有初始化命令的世界地面目标,根命令固定;B(c)=A+方向单位向量×幅度,每对A→B(c)→A。下一命令只在返回A后的完整周期边界生效;返回腿不换B,目标不累积、不使用实测落点重锚。独立CPU逐261000tick核对固定目标/相位,最大目标差0。
## 2. 真实物理测量
W约52.8597N。解析20周期:承重/重新承重接触均100%;承重末平均≥0.99998W、卸载剩余最高0.67577、重新承重末≥0.98829W;每周期摆动真实腾空100%/1.4s;有向XY每轴进度最低86.03%(门槛50%);平均落点3D误差13.66~13.79mm(门槛50mm)。
| 运行 | 最差XY每轴进度 | 周期平均落点误差mm | 最大根XY误差mm(仅诊断) |
|---|---:|---:|---:|
| 解析20 | 86.03% | 13.66~13.79 | 80.49 |
| 解析S1 | 77.39% | 13.62~13.79 | 80.49 |
| 解析S2 | 77.90% | 13.52~13.82 | 80.49 |
| 解析S3 | 82.41% | 13.58~13.79 | 80.49 |
| u256-S1 | 67.80% | 8.75~9.63 | 68.98 |
三条解析切换及u256 S1承重/重新承重接触均100%、摆动腾空均100%/1.4s;重新承重最低解析0.98817W/u2560.97449W。全9运行:最小gap−0.0298mm、最大高度误差67.78mm(≤100mm)、最大倾斜0.05897rad(≤0.6)、执行器速度/力矩归一化比最高0.215740、关节超界0、竖直guide0;无数值/失控保护触发。根XY超过30mm如实保留,不提升为旧严格P1通过,不宣称u256必要或优于解析。
完整58周期/348相位Fn min/mean/max、gap、实际XYZ净位移和根漂移,位于每个`*-audit/phase-summary.csv`及对应`actual-audit.json`;不以轨迹图代替物理检查。
## 3. 连续性、接口及退出
- Python公开入口:`scripts/go2_fl_skill.py`的`start`、`SkillRun.query/queue/request_stop/wait`;CLI含`authorize/validate/run/status/queue/stop`。输入SI单位,有限7个命令ID、周期整数1..20;模式解析或冻结u256。详细单位、范围、辅助条件、仿真真值及生命周期见[接口说明](go2_fl_skill_v4_api.md)。非法输入在队列/动作修改之前拒绝,不静默降级。
- 全局500Hz样本连续、100Hz控制,所有49个接缝45项枚举carry/RNG指纹相等且参考仅刷新一次。初始reset/pose API计数22/14后均无增量,初始化额外物理调用0;不冒称初始化也没有reset。原相位/碰撞/接触获取/数值/关节/力矩速度/辅助/guide保护未关闭。
- stop在tick5接受、tick4500完成本周期落地重新承重后正常退出,无额外接缝/参考刷新;非法命令在tick5被拒,carry/RNG前后SHA256一致,轨迹仍完整通过。
- **独立包外部实时接口另有实证**:从独立目录的公开Python接口启动无预装队列的6周期,另一CPU调用方依次`queue(base24)`、`queue(y16)`并收到真实消费receipt,查询真实状态;非法ID未发布任何request文件;末周期请求stop,最终6周期完整安全退出。不是仅封装固定演示命令。证据`independent-live-client.json`。
- 每个进程保留同对象配置preclose核对;18个Kit子进程均实际wait退出0,9个监督器由公开调用方实际wait退出0,owned身份消失。父各核对456个有限绑定文件,无compute/任务graphics残留,原7桌面graphics身份不变;未终止无关进程。离线张量审计仍不单独证明运行期连续性,PhysX内部完整缓存未观察。
## 4. 本地交付及复现
独立版本目录:
`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.0/`
包含控制/接口源码、配置、示例、生成USD资产及依赖、原冻结u256/manifest/134文件snapshot、实际原始轨迹/接缝/授权/监督/退出/离线审计、依赖校验、适用范围与限制、SHA256SUMS及实测状态演示。压缩副本为相邻`.tar.gz`及`.tar.gz.sha256`;不在旧`releases/`内发布。
一条真实启动演示命令(每次创建新有限登记/新运行目录,不复用旧grant):
```sh
cd /home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.0 && bash demo.sh
```
一条完整功能验证命令(GPU串行,首次失败即停,保留原件):
```sh
cd /home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.0 && bash verify_functional.sh
```
本轮实际独立复现从该目录、通过公开`go2_fl_skill.py run --sequence S1`启动,另一次通过公开Python接口实时排队/停止;不是从开发脚本替代执行。确切cwd、绝对命令、来源hash与实际wait退出见authority目录的`independent-S1-invocation.json`、`independent-S1-driver-exit.json`、`independent-live-client.json`及独立绑定。当前源码与独立包30个有限运行源hash一致。
演示:`demo/measured_state_animation.gif`、`timeline.png`及`measured_state_keyframes.png`,来自独立S1真实轨迹。**是实测root/joint的FK+实测足端/接触力重建动画,不是Kit摄像机录像**;原始500Hz数据完整,仅显示10fps,FK对实测足端显示帧误差<1mm。未提供Kit摄像机录像,不混称。
## 5. 资源、审查及限制
2026-10-15 17:38:07开始,约2小时余完成实际运行与交付(最终完成时间见task-complete.json),早于2026-10-16 05:38:07截止。新任务单独登记,不沿用v3 grant。实际18/24次Kit、261000完成共享物理调用/272250预留(整体护栏650000)、58周期/522秒仿真、零训练/零物理失败重试;运行+授权目录1269787712字节,连同包/分析/压缩副本低于24GiB。预留不是实际,未用余量不自动再跑。native观察非硬配额且覆盖不完整,不把它与own字节重复相加。
独立审查曾指出u256前置证据不能把安全早停/改队列冒充完整固定矩阵;已在物理执行前增加完整周期/tick/pair history严格校验及负例,复审通过。最终CPU接触周期1486passed/6skipped,仅辅助资格,不替代上表真实仿真。9次物理/监督首遍通过,旧失败/发布/冻结原件1217文件重哈希无变化。
**已完成全部必需交付。未做/不声明**:u25620周期或其他切换序列、全部任意转换/全域鲁棒、无辅助平衡、多seed统计、整机、实机、sim2real、RootXY精度优化、训练及通用v2 adapter;不是这些项目失败,而是未执行/超本轮范围。仍依赖本机固定IsaacLab/IsaacSim/RTX5080及仿真真值,未建设跨机安装兼容层。
完整本机证据:`/home/cen/sf-runs/skill-v4-01-{nominal20,S1,S2,S3,u256-S1,stop,invalid,independent-S1,independent-live}/`与`/home/cen/sf-runs/simskill-v4-authority-01/`;独立包内`evidence/acceptance.json`及各运行原始文件可随包本地复核。
+86 -2
View File
@@ -1,6 +1,90 @@
# 当前进度:RootXY-v2完成81次PPO更新,尚无合格新候选
# 当前进度:0.4.2性能与限定S1结果
当前对象是Go2 FL仿真单腿,浮动根、有界XY/转动辅助、竖直辅助为0;使用仿真真值。**accepted_skill=false**。单周期功能通过不等于连续稳定、扰动鲁棒、整机控制或实机能力。
[0.4.2结果与限制](contact_cycle_sim_performance_v042_result.md):当前仓库`performance_v042/`保存全部原件。GUI窗口截图33.98FPS,全段绘制提交35.30次/秒;原物理/控制/门槛不变,新GUI/headless各6周期27000tick、70张量及完整文件SHA与0.4.1相同。CPU1557passed/6skipped;11Kit/67500实际调用/73500预留、0训练,结束时57owned身份均不存活。独立只读复核无封装阻塞。
**不是1×实时**:GUI54模拟秒耗时764.863墙钟秒(0.07060×),headless294.679秒(0.18325×);GPU整卡活跃均值39.74%/133.43W,未人为加负载或设省电上限。新高频渲染下提前关闭/visualizer故障等仅CPU负例,原生注入未测;完整矩阵/u256未重验。headless首次独立postexit观察器GPU admission异常原因未知,原件保留;实际wait0后原规则第二次严格核验通过,不作回溯归因。独立包在`performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2/`,最终传输校验/库存/完成状态以`performance_v042/{package-verification,final-inventory,task-complete}.json`为准。包内完整CPU也1557passed/6skipped;首次打包漏历史测试fixture导致39项失败及补齐记录保留,不改任何运行源/断言/门槛。accepted_skill=false与旧严格失败不变。
---
# 历史:0.4.1真实Kit GUI与无窗口S1验证完成
[0.4.1完整结果](contact_cycle_sim_gui_v041_result.md) / [接口](go2_fl_skill_v4_api.md)。独立包`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.1/`,一条命令`bash demo_gui.sh`打开真实解析S1窗口;0.4.0目录、archive、校验与证据原样保留。
- **物理资格**:当前源码解析GUI S1与headless S1均6连续周期/27000tick,原功能/安全/连续性/完整绑定通过;两条实际轨迹70张量逐位相同,删除唯一视觉visualizer_cfgs后有效cfg相同。初始化reset/pose22/14后无增量,无重置/重锚/物理参数改动。
- **GUI资格**:真实桌面Kit窗口、原蓝色3kg负载/足端/地面、动态相位/命令/周期HUD;原场景最终保留约5秒且carry/计数相等,真实WM_DELETE后wait0、窗口/PID消失、strict GPU清理通过。真实截图、原始240秒录屏及35秒原速跳切剪辑在包`demo/`,不是FK重建/录像替代。GUI约2–3FPS,不保证墙钟1×实时。
- **提前关闭**:真实运行中在2040tick中止,0完整周期,原因gui_window_closed_before_completion、exit2、原success=false;部分轨迹无物理失败标志,preclose/wait/清理通过,不提升为完成。
- 全部13次尝试共21/24Kit、83040实际调用(含完整性失败10的27000)、354250预留,0训练;01..09零物理失败原件及10的来源绑定失败不改判。11外部观察器关窗后查询已销毁窗口的异常也保留,独立补件不覆盖原worker判定。464/457/464有限绑定文件、2143原保护文件核对无变化;CPU1514passed/6skipped。
- 未重新验证0.4.1完整20周期/三序列/u256矩阵或u256 GUI;极早启动关窗、GUI安全停止及10分钟自动超时未实测。范围仍单seed/原辅助/真值/单腿,旧严格accepted_skill=false/RootXY失败不变。本轮停止仿真,不耗用余量扩训/优化/扩域。
---
## 历史:仿真功能版0.4.0已交付、真实独立复现通过
本轮必需目标全部完成:[完整结果](contact_cycle_sim_skill_v4_result.md) / [Python与CLI接口](go2_fl_skill_v4_api.md)。独立本地包:`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.0/`(相邻tar.gz及SHA256)。不是旧严格accepted_skill发布,历史`accepted_skill=false`不变。
| 本轮功能 | 实际结果 |
|---|---|
| 可调用接口/本地包 | Python start/query/queue/request_stop/wait+CLI,带源码/配置/资产/u256/依赖hash/原始证据 |
| 解析长运行 | 原名义20连续周期通过,90000tick/19接缝 |
| 同运行切换 | 固定S1/S2/S3各6连续周期全部通过;固定世界A,去/回成对,无落点重锚 |
| 可选u256 | 解析全部通过后,同S1六周期物理/监督通过 |
| 停止/拒绝 | tick5请求、周期1安全停止;非法ID拒绝,carry/RNG不变 |
| 独立目录复现 | 公开CLI真实S1通过;公开Python外部实时queue/query/stop另一次6周期通过 |
| 演示/退出清理 | 实测状态重建GIF(非Kit录像);所有preclose、wait退出0、owned/GPU清理通过 |
新9运行/58周期/261000实际共享物理调用,18Kit,0训练/0物理失败重试。逐周期检查原物理门槛,49接缝carry相等且一次刷新;初始化后reset/pose计数无增量。解析最大根XY80.49mm仅诊断,旧30mm严格失败不改写。父核对原保护1217文件无变化,当前与独立包运行源hash一致。CPU1486passed/6skipped只作辅助资格。
范围仍为有原辅助条件的浮动根Go2 FL单腿/仿真真值/seed93501,不声称无辅助平衡、全域、多seed、整机、实机或sim2real。u256仅本轮S1六周期验证,不声称优于解析。当前任务提前完成,不追加训练/RootXY/扩域;所有实际命令和原始证据见结果与索引。
---
## 历史状态:u256补验与6组有限命令成对连续3周期全部通过(零训练)
当前对象是Go2 FL仿真单腿,浮动根、有界XY/转动辅助、竖直guide为0;使用仿真真值和主动辅助。**既有解析基线已完成完整功能流程,accepted_skill仍为false**(不表示独立鲁棒验收、无辅助平衡或整机/实机能力)。不继续RootXY精度优化,不训练,不开展整机/实机/sim2real。
## 最新实际结果:ff-v3-01全部通过,限定仿真功能目标已完成
- 用户明确“批准,以后你的预算是无限的,不需要我批准”;后续自主预设运行护栏,不再逐次请批,详见[持续执行授权](execution_authority.md)。本次仍严格执行原固定上限,0训练/0重试,未扩大任务范围。
- **新u256原命令连续3周期完整通过后,6组成对矩阵全部通过**:每组解析/u256各27s连续3周期,物理、无重置连续性、完整产物、preclose身份、GPU监督、退出0和清理均通过;6对初态在u256首rollout前精确相等。原campaign success=true,13条完整轨迹/39周期/234个相位,初始reset/pose计数22/14后无增量。
- 实际attempted=completed共享物理调用**175500**,初始化额外0;原保留178750不当实测或下次余量。总3585.201s(59.75分钟)<6540s;own约1.45GB<2GiB,native观察增长1105834619字节(与own重叠、非硬配额/完整覆盖)。14个Kit子进程及监督器都实际wait退出0,owned身份全部消失,无compute/任务graphics残留,7个原桌面身份不变。
- **解析本身已足够,u256未提高有限矩阵功能通过率(双方6/6)**。u256落点误差较小,但最差每轴进度53.09%对解析68.38%,不据单seed结果称必要或统计优越。保持权重/辅助/原门槛;根XY80.49/68.98mm仅诊断,不重写旧严格失败。
- 独立CPU重读13条原轨迹,原功能窗口与固定新命令逐tick一致(目标差0);离线工具自身不证明运行期连续性。父复核450个绑定文件、原候选/134快照及旧ff-01的298文件均通过;事后接触周期CPU回归1380passed/6skipped(109.86s)。旧GPU失败仍因缺瞬时列表而无法具体归因,本次未复现;没有放松归属检查。
[逐组物理/完整性表、真实力/gap/位移、预算/命令及图GIF](contact_cycle_functional_followup_v3_result.md)。原始根`/home/cen/sf-runs/ff-v3-01/`与`ff-v3-authority-01/`;不是全工作域/鲁棒性或独立技能验收。当前任务完成,不自动追加RootXY、训练、扰动或整机/实机。
## 运行前准备:功能补验v3只读诊断与CPU资格(当时未获新运行预算)
- 已读最新状态/索引/ff-01结果,原298个留档文件重哈希一致。旧失败可定位到live GPU客户集合不属active owned/桌面graphics的门禁;没有失败瞬间NVML/PID-birth集合,仍不能判定外来、身份变化或观测竞态。原失败不改写,不终止无关进程。
- 只增加原GPU检查的精确观测/近期检查落盘,保留原决策、顺序和异常,不增加NVML宽限或放松归属。针对性CPU等价/负例及首失败停止测试通过;接触周期全套**1380passed、6skipped(109.33s)**,不代表Kit或新物理资格。
- 已固定`functional_followup_v3.json`:先一次原命令u256连续3周期监督补验;成功后才做21.80°/38.66°/53.13°三个方向×16/24mm两个幅度,六组各解析/u256连续3周期;组件位于原命令包络内,但新方向尚未物理验证。u256首个rollout之前精确核对该组解析初态/命令/RNG/IK,不匹配则停,不发布优劣差值。
- 新模式/新绑定和CPU检查已接入;原功能阈值/辅助/权重不变,0训练,30mm根XY仍只记录。拟申请1准备+13物理启动、178750共享物理调用、6540s/109分钟、own2GiB/RTX5080单环境;**尚待用户对这一新预算明确批准,未生成grant、未创建新运行目录、未启动Kit/物理**。首次任意失败停止,剩余not_run,不重试/扩训。
[诊断、固定矩阵逐组未运行表及复现命令](contact_cycle_functional_followup_v3.md);CPU证据`logs/contact_cycle/functional_followup_v3_prepare01/`。本轮新增物理0,新的物理结果/运行完整性均not_evaluated,不拿ff-01物理完成冒充补验。
## 最近真实仿真:ff-01(此前新授权已消费,不重试)
- 先固定功能判据,再经用户明确“批准”,完成1次零物理配置准备+4次串行物理启动。解析和冻结u256各完成新的9s单周期;**解析零残差27s/13500tick连续3周期连同监督检查通过**。两次接缝45项枚举承接状态相同、参考各刷新一次,初始reset之后没有reset/pose写入增量,没有瞬移/重锚/拼接。
- 解析3周期:承重末各约1W,卸载剩余0.676/0.210/0.167,摆动Fn0/gap约43.4~53.9mm、净XY分别+21.54/+18.34、−24.31/−19.82、+24.44/+19.31mm,重新承重末0.996/0.988/0.996W;终止`protocol_complete`。最大根XY80.49mm只记录。
- 冻结u256+解析参考的27s也记录全部物理功能/安全/连续性检查通过,worker退出0;**监督器末段GPU归属检查失败,该行监督不通过,原campaign success=false不改写**。故障瞬间clients/PID集合未存,不能断言只是NVML退出滞后或没有外来客户。未重试;不把事后清理通过换成通过判定。解析本身已足以达成本轮目标,不声称学习必要或优于解析。
- 实际36000完成共享物理调用,初始化额外调用0,优化0;37000为预留上限,不能作为新启动余量。campaign约815.045s(13.58分钟),own产物约数百MB,未超50分钟/1GiB。五个子进程均等待退出0且清理通过;父级确认全部owned身份消失、无计算/任务图形残留、原7个桌面图形客户端存活,446个绑定文件及原u256重哈希通过。native监控非硬配额、覆盖不完整。
- 启动前相关CPU341passed;结束后接触周期**1341passed、6skipped(108.65s)**。全程物理记录无数值/明显穿透/失控/关节执行器越界;最大归一化执行器比0.2153,关节超界0、最小gap−0.0298mm、竖直guide0。不是全仓全依赖资格。
实际命令、每阶段力/gap/位移、终止与监督失败详见[本轮结果](contact_cycle_functional_flow_v2_result.md)。证据根`/home/cen/sf-runs/ff-01/`及`ff-authority-01/`;有实测状态FK关键图/GIF(非Kit摄像机视频)。原严格P1失败、旧权重/发布包和全部失败账本保持不变。当前不再启动新仿真。
## 运行前准备记录(历史,当时尚未批准/启动)
- 先核对实际代码、P0原轨迹、P1严格失败及rxy-04最新保留结果,再固定独立`functional_flow_v2.json`:复用原功能窗口,承重/重新承重≥0.5W且接触比例≥80%、卸载至少下降20%、摆动gap≥10mm且Fn≤2N的比例≥80%/连续≥0.1s、有向XY各≥命令50%。原碰撞、数值、关节/执行器、辅助和失控保护保留;30mm水平漂移只记录。
- 新`FunctionalFlowProtocol`复用原整数时钟和解析A→B→A→B命令,不带P1额外精度门槛。13500tick命令逐一与旧P1相同,未修改原参考/残差/权重/历史配置。原相位终止初版不动。
- **只读复核历史真实仿真**:P0 u256单周期1/1通过,卸载末0.481W、摆动gap47.44~57.99mm/净XY19.46、20.16mm、重新承重末0.981W;最大根XY68.98mm如实记录。最近解析/冻结u256各保留4条单周期也满足功能窗口,减字段轨迹局限见详细报告;不宣称重审了完整128环境或有效成对学习增益。
- 明确拒绝以一条9s轨迹报告3周期通过。**本轮新物理调用0、优化0;连续3周期仍未验证**,旧P1双方首周期卸载阶段严格漂移失败不改写。
- 接触周期CPU回归**1329passed、6skipped(108.37s)**;不是全仓/Kit/新物理资格。冻结CPU preflight通过134个原快照及候选核验,沿用已有AGENT非运行时例外;旧`verify_delivery.py`对AGENT逐字节失败也保留,不修改验证器。
- 新功能受监督入口/配置绑定尚未接入与Kit验证,不能把新描述符直接塞入旧P1/P2 runner。已准备一次性新预算提案:配置准备1次+物理启动最多4次,RTX5080/单环境、36000 rollout+1000初始化=37000共享调用、总50分钟、own产物1GiB、0训练,无重试/恢复;**尚待批准,不创建grant、不借用旧授权**。
判据、历史阶段数据、离线复现命令、未完成的启动适配及预算详见[功能流程v2](contact_cycle_functional_flow_v2.md)。以新的完整动作证据判断进展,不以此前81次更新或精度收益替代。
## 既有RootXY与严格验收历史(保留,不作为本轮推进方向)
单周期功能通过不等于连续稳定、扰动鲁棒、整机控制或实机能力。RootXY-v2实际完成81次PPO更新,尚无合格新候选。
## 已确认的里程碑
+128
View File
@@ -1,7 +1,135 @@
# 下一阶段证据索引
## 最新:0.4.2应用循环与高频GUI(限定解析S1)
[结果/测量边界](contact_cycle_sim_performance_v042_result.md)。仓库`performance_v042/`内:`performance-result.json`、`runs/display-S1-05/`(真实PNG/120s原速录屏、原worker/campaign/27000tick轨迹)、`runs/headless-S1-06/`、两份`*-physical-audit/actual-audit.json`、`display05-v041-GUI11-comparison.json`、`display-S1-05--headless-S1-06-comparison.json`、`final-source-evidence-review.md`、CPU1557passed/6skipped日志/XML及41输入来源快照。GUI截图33.98FPS/全段35.30提交每秒;GUI/headless RTF仅0.07060×/0.18325×。
首个baseline PREP失败、CPU收集失败、两次P1审查及修复、headless首个独立postexit观察失败(原因未知)和后续严格通过原件全部保留。GUI/无界面原campaign均成功,70张量/文件SHA与0.4.1一致,不继承完整矩阵或原生故障注入资格。旧0.4.0/0.4.1原件/严格失败不改写。独立包`performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2/`及相邻tar.gz/sha256;最终`package-verification.json`、`final-inventory.json`、`task-complete.json`在该工作根。包内CPU完整1557passed/6skipped;首次漏测试fixture的39fail、补齐两次记录与最终成功原件均保留,历史fixture不是运行grant。
## 最新:0.4.1真实Kit GUI(物理与GUI分别验证,零训练)
[实际结果与限制](contact_cycle_sim_gui_v041_result.md) / [固定GUI计划](contact_cycle_sim_gui_v041_plan.md) / [API](go2_fl_skill_v4_api.md)。GUI S1/headless S1各6周期成功;提前关窗2040tick/exit2为预期中止,原success=false不提升。21Kit/83040实际调用;2143原件无变化,0.4.0分版保留。
| 路径 | 原始证据/用途 |
|---|---|
| `/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.1/`及相邻archive/sha256 | README、demo_gui.sh、原headless入口、源码/资产/冻结权重、校验;不捆绑已装Isaac运行时 |
| 包`demo/{support,lift,airborne,touchdown,reload,command-switch,final-retained}.png` | 真实X11 Kit窗口像素,非FK重建 |
| 包`demo/live-kit-window.mp4`、`live-kit-edited-original-speed.mp4`及`excerpt-provenance.json` | 原始240秒连续录像与35秒原速跳切;12FPS是采样,不是Kit渲染帧率 |
| `/home/cen/sf-runs/skill-gui-v041-{gui-complete11,headless-S1,gui-abort01}/S1/` | 原始approval/PREP绑定、500Hz trace、events/seams、worker/supervisor、strict GPU监督和wait退出 |
| `/home/cen/sf-runs/skill-gui-v041-gui-complete01`至`11`(第8为gui-diagnostic08) | 所有失败/修复现场原样保留,01..09零物理,10完整27000调用但来源绑定失败,不翻案 |
| `/home/cen/sf-runs/simskill-gui-v041-authority-01/acceptance.json` | 跨13次独立会话的整体预算/原判定/源对照汇总,不冒充单共享会话ledger |
| authority各`*-postexit.json`、`*-physical-audit/`、`abort-partial-audit.json` | 当前源码464/457/464有限绑定文件、owned/GPU/2143保护文件及独立原轨迹功能核对 |
| authority`gui-headless-trace-comparison.json` | 两次同seed当前源码的70轨迹张量实测逐位相等,不宣称其他seed必相等 |
| authority`gui11-observer-supplement.json`、原`gui-complete11/trial-result.json` | 真实WM_DELETE后xprop查已销毁自有窗口的原异常与补件;不覆写原worker/监督判定 |
| authority`cpu-acceptance.log/.xml`、`final-{source,frozen}-validation.json` | 1514passed/6skipped与原u256/134快照校验,仅CPU资格 |
| `/home/cen/sf-runs/simskill-gui-v041-*-review.md` | 独立源码/实际证据只读复核,不能代替真实运行 |
| 包`evidence/v041/` | 上述主要原件与失败源码快照;native缓存和无关桌面原图仅原路径本机留档 |
---
## 历史:仿真功能版0.4.0完成并独立复现(零训练)
[实际通过表/命令/限制](contact_cycle_sim_skill_v4_result.md) / [接口](go2_fl_skill_v4_api.md) / [运行前固定计划](contact_cycle_sim_skill_v4_plan.md)。解析20周期、S1/S2/S3各6周期、u256 S1、停止/拒绝、独立目录公开CLI及实时Python接口均物理/完整性通过;58周期/261000实际物理调用、18Kit、49接缝,无物理失败重试。`sim_functional_accepted=true`,历史`accepted_skill=false`不变。
| 路径 | 作用 |
|---|---|
| `/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.0/` | 独立可调用本地包,源码/配置/资产/u256/依赖及原始证据;README、demo.sh、verify_functional.sh、SHA256SUMS |
| 相邻`.tar.gz`及`.tar.gz.sha256` | 同版本本地压缩副本及文件校验,不在旧releases发布 |
| 包内`evidence/acceptance.json` | 9运行逐项物理/连续性/监督/父退出核对及真实指标;不升级旧严格验收 |
| 包内`demo/measured_state_animation.gif`及PNG | 独立S1实测root/joint FK+实测足端/接触力,**非Kit摄像机录像** |
| `/home/cen/sf-runs/skill-v4-01-{nominal20,S1,S2,S3,u256-S1,stop,invalid,independent-S1,independent-live}/` | 全新当次grant/准备/绑定、逐500Hz原trace/initial/events/seams、worker/supervisor结果及GPU/native监督原件 |
| `/home/cen/sf-runs/simskill-v4-authority-01/registration.json`、`skill-v4-ledger.json` | 本轮用户12小时独立登记、deadline、24Kit/650000调用/24GiB护栏;实际9次run均共享serial锁/ledger,不复用旧grant |
| authority根各`*-invocation.json`、`*-driver-exit.json`、`*-postexit.json` | 真实命令/调用方wait退出、456有限绑定文件及owned/GPU清理;独立live另见client JSON |
| authority根各`*-audit/{actual-audit.json,phase-summary.csv}` | 9条完整原轨迹/58周期/348相位离线物理核对及固定世界目标逐tick差0;离线自身不证明运行期连续性 |
| authority根`independent-live-client.json`、`live_public_demo.py` | 独立包公开Python实际启动/外部queue两命令/query/非法拒绝/stop/wait结果;不是预装固定演示替代 |
| authority根`independent-S1-invocation.json`及独立`prepare/binding.json` | cwd与所有运行源/资产来自独立目录的真实Kit复现,非开发目录冒充 |
| authority根`final-source-validation.json`、`independent-frozen-validation.json`、`acceptance.json` | 当前与独立包有限运行源一致、原134snapshot/u256核对,保护原件1217文件无变化 |
| authority根`final-cpu-tests.log/.xml` | CPU1486passed/6skipped,非Kit/物理资格替代 |
| `/home/cen/sf-runs/simskill-v4-{review,acceptance-review}.md` | 独立代码/实际证据审查;前置完整矩阵门禁问题已修复复审,物理结果以原件为准 |
| authority根`task-complete.json`、`final-inventory.json` | 最终任务收尾/实际护栏使用、产物hash;未用预算不触发后续运行 |
本轮源码接口支持有限7命令/1..20周期;只声明真实表内验证范围。无u25620周期/任意切换、无辅助平衡、多seed、整机实机/sim2real结论。旧目录/原件保持原路径原字节。下面是历史索引,旧grant与当时“未运行”状态不是当前授权或新版本结果。
---
相对路径以仓库为根;`/home/cen/sf-runs/`是本机外部证据,**不随此源码提交**。原始轨迹、权重、授权、账本不移动/覆写;历史grant不是新运行许可。跨机器使用须另行携带并校验所需资产、快照、权重和报告。
## 最新实际结果:ff-v3-01全部通过(13条连续轨迹/39周期,0训练)
[逐组物理/完整性、测量与命令](contact_cycle_functional_followup_v3_result.md)。**先新u256补验完整通过,再完成6组各解析/u256连续3周期;6对初态一致,物理/监督/退出/清理全部通过,原campaign success=true**。双方6/6,没有功能通过率学习增益;accepted_skill=false,不称全域/鲁棒/无辅助平衡。旧GPU失败未知原因仍保留。
| 本机证据(不随源码自动分发) | 用途/范围 |
|---|---|
| `/home/cen/sf-runs/ff-v3-01/campaign-result.json` | 原始最终通过判定、14行退出/清理/身份与实际175500共享物理调用;无失败/重试 |
| 同根`prepare/`、`preparation-reviewed.json` | 新0物理配置准备及父450有限文件复核,非复用旧绑定 |
| 同根`u256-nominal-3/` | 原命令新13500采样/2700控制/两个接缝,完整补验通过 |
| 同根六组`*-reference/`与`*-u256/` | 原trace/initial-state/initial-pair/events-and-seams及worker/supervisor结果,连续3周期不拼接 |
| 同根`gpu-observations.json` | 原检查直接消费的NVML/active/known集合及最近16次,原决策不放宽 |
| `/home/cen/sf-runs/ff-v3-authority-01/{approval.json,invocation.json,gpu-baseline.json}` | 用户新批准与持续预算指令、真实命令/硬件/桌面初态 |
| 同根`supervisor-exit.json`、`parent-postexit.json` | 监督器实际wait退出0,owned消失、GPU无残留、7桌面身份不变、450文件重哈希;同对象cfg是原worker preclose检查 |
| 同根`analysis/actual-summary.json`、13份`*-audit.json` | 原张量独立CPU功能复核及固定新命令逐tick目标差0,离线审计自身不证明运行期连续性 |
| 同根`analysis/actual-phase-summary.csv` | 真实234相位力/gap/XYZ净位移及漂移,不是计划轨迹 |
| 同根`final-contact-cycle-tests.log`/`.xml` | 事后接触周期CPU1380passed/6skipped(109.86s),不是全仓/依赖资格 |
| 同根`previous-failure-still-unchanged.json`、`final-frozen-verification.json` | 旧298文件不变;原候选/134快照通过,AGENT非运行时例外不抹去旧逐字节失败 |
| 同根`nominal3-visuals/`、`y-bias16-{reference,u256}-visuals/` | 实测状态FK关键图/GIF,非Kit摄像机录像;原500Hz数据不抽稀 |
| 同根`final-inventory.json`、`final-source-snapshot/` | 最终证据逐文件hash/大小及当前交付源码;inventory不递归包含自身 |
用户后续不再逐次请批预算,见[持续授权边界](execution_authority.md);每次仍须自行固定护栏/新身份/独立登记,不复用已消费grant,不默认扩大任务或自动训练。
## 功能补验v3运行前只读诊断/CPU准备(历史阶段,新物理0)
[监督诊断、固定小矩阵与预算/命令](contact_cycle_functional_followup_v3.md)。旧ff-01的298文件校验一致;只能确认live客户集合归属拒绝,失败瞬间PID集合缺失,真实外来/身份变化/观测竞态仍未知。旧失败不改判。
| 证据/源码 | 作用域 |
|---|---|
| `configs/contact_cycle/functional_followup_v3.json` | 先u256原命令3周期补验,再3方向×2幅度×2控制各3周期;固定新预算,不是grant;SHA256 `aaed45884eab9912db6c7bac1d82e2ab27784982c524a838b7886cdd763146f4` |
| `logs/contact_cycle/functional_followup_v3_prepare01/previous-supervision-diagnosis.json` | 只读旧终态/源快照/GPU初态复核、全部输入hash、已知分支及未知项;不倒推具体失败PID |
| 同根`frozen-plan-check.json` | stdlib检查的14槽顺序、具体命令/预算/当前源hash,execution_authorized=false |
| 同根`contact-cycle-tests.log`/`.xml` | CPU1380passed/6skipped;含GPU检查决策等价、退出/身份变化负例、初态门禁、首失败停止的mock,不是物理通过 |
| 同根`targeted-tests.log`/`.xml` | 早期专项138passed,保留中间资格记录 |
| 同根`frozen-verification.json`、`final-inventory.json`、`source_snapshot/` | 原冻结候选/134快照核对、CPU准备身份;不冒充新的Kit同对象绑定 |
| `scripts/functional_gpu_watch_v3.py` | 原GPU决策不变,仅精确记录已有查询/discover及近期样本,失败仍抛出;没有NVML宽限/重试 |
本CPU准备形成时拟用`/home/cen/sf-runs/ff-v3-01/`和`ff-v3-authority-01/`尚不存在,步骤2/3均未运行、尚无新grant;后续实际完整通过见上方最新证据。不得复用旧授权/绑定;解析和学习物理功能与监督完整性继续分别报告。
## 最近真实仿真:功能流程v2单周期与连续3周期
[实际结果、阶段数据与启动命令](contact_cycle_functional_flow_v2_result.md)。**解析单周期及连续3周期监督通过;u256三周期物理完成但GPU监督失败,原campaign success=false保留**。实际36000共享物理调用、0优化,无重试;不是独立技能或实机验收。
| 证据根/产物 | 用途与限制 |
|---|---|
| `/home/cen/sf-runs/ff-01/` | 本轮新批准的零物理配置准备+4行真实物理;原始轨迹/动作/接缝/worker与supervisor结果、授权副本及源快照 |
| `ff-01/campaign-result.json` | **整体监督失败原件**;前三物理行监督通过,最后u256三周期GPU归属检查失败,不能用worker退出0改为通过 |
| `ff-01/reference-3/trace.pt`、`events-and-seams.json`、`supervisor-result.json` | 解析27s/13500tick、三个功能周期/两个无重置接缝及监督通过 |
| `ff-01/u256-3/` | 完整27s物理通过的原始记录,worker/preclose/退出0与GPU监督失败分别保留 |
| `/home/cen/sf-runs/ff-authority-01/approval.json`、`invocation.json` | 用户本轮“批准”对应的新授权和实际命令;已消费,不得复用 |
| `ff-authority-01/actual-summary.json`、`actual-phase-summary.csv` | 4运行/8周期/48相位的Fn、gap、XYZ、漂移、安全及终止汇总;不隐藏最后监督失败 |
| `ff-authority-01/parent-postexit.json` | 446绑定文件/原u256核对、owned身份消失、无计算/任务图形残留、原7桌面客户端存活;不能反证故障瞬间归属 |
| `ff-authority-01/*-offline-audit.json` | 四条实际全轨迹离线功能复核;自身不证明运行连续性或提升监督结果 |
| `ff-authority-01/reference-3-visuals/`、`u256-3-visuals/`等 | 实测状态FK关键图/GIF及曲线,非Kit摄像机视频;u256三周期图已标注监督失败 |
| `ff-authority-01/cpu-qualification.log`、`final-contact-cycle-tests.log` | 启动前相关341passed;结束后接触周期1341passed/6skipped,非全仓全依赖资格 |
| `ff-authority-01/final-inventory.json` | 本轮产物与源码hash/字节留档;不修改原campaign失败或旧权重/旧账本 |
表内`ff-01/`及`ff-authority-01/`简写均以`/home/cen/sf-runs/`为前缀,是本机外部文件,不随源码自动分发。失败瞬间NVML clients/owned PID集合未存,具体原因证据不足;没有重试或追加训练。解析主功能目标已完成,当前不再启动仿真。
## 功能流程v2运行前准备证据(历史阶段,无新物理)
当时准备结论/固定判据见[功能流程v2](contact_cycle_functional_flow_v2.md)。证据根`logs/contact_cycle/functional_flow_v2_prepare01/`为本机日志,不自动随源码提交;该准备阶段物理调用0、优化0,当时尚无连续3周期验证,不将历史严格失败改判。
| 新产物 | 含义/局限 |
|---|---|
| `configs/contact_cycle/functional_flow_v2.json` | 实验前固定的独立功能判据;SHA256 `16c064228b30f36bbd420b7c2290602bacae6601cac2cec5deb16440f7600625`;不是授权 |
| `historical_p0_audit.json` | 原P0真实u256单周期1/1满足功能窗口;新仿真0,最大根XY68.98mm保留 |
| `historical_analytic_selected_audit.json`、`historical_frozen_selected_audit.json` | 最近rxy-04双方各4条保留单周期满足窗口;减字段/选择子集,不是完整128环境安全重审或有效配对增益 |
| `historical_p0_not_three_cycles.json` | 同一9s轨迹要求3周期被拒绝,后两周期not_evaluated |
| `frozen_cpu_preflight/` | 134原快照+u256身份/CPU加载通过,沿用既有AGENT非运行时例外;无Kit/物理 |
| `strict_verify_delivery_failure.log` | 旧逐字节验证器本次在用户已改的AGENT.md失败,未改验证器或旧manifest |
| `contact_cycle_tests.log` / `.xml` | 1329passed、6skipped;仅CPU接触周期回归,不是全仓/新运行资格 |
| `p0_u256_trace_plot.png`、`analytic_selected_env0_trace_plot.png` | 历史实测力/间隙/位移/漂移曲线,不是新仿真视频或渲染关键帧 |
| `inventory.json`、`source_snapshot/` | 本次CPU/只读证据与新代码的hash留档,不是Kit运行绑定或grant |
原P0/原P1/rxy-04轨迹、权重、grant、失败/账本保持原处原字节。新功能受监督入口仍需有限适配与新配置准备,不能把这些准备证据说成已获启动资格。
## 必须保护的身份
| 对象 | 路径 / SHA256 |
+31
View File
@@ -0,0 +1,31 @@
# 当前执行与预算授权
## 最新本轮:0.4.1真实GUI(独立新范围,停止仿真)
用户要求保留0.4.0并加入实时GUI,依既有自主执行指令固定[GUI计划](contact_cycle_sim_gui_v041_plan.md)。新scope从2026-10-16 09:51:52 +08开始,12小时截止21:51:52;原0.4.0授权已结束,绝不复用/延长。证据根`/home/cen/sf-runs/simskill-gui-v041-authority-01/`,scope.json及protected-registration.json固定24Kit/650000调用/24GiB、零训练/零优化,保留所有失败、GPU/身份/安全/退出保护。
实际13个新试验分别创建独立有限会话(不是共用一个ledger),父级另做跨会话预算/串行核对:21Kit、354250预留调用、83040实际调用,含失败10的27000与预期中止2040;所有owned清理完成。GUI/headless各六周期原监督通过,提前关窗原判失败且按预期保留部分轨迹;详见[结果](contact_cycle_sim_gui_v041_result.md)。当前停止物理,不自动耗用剩余预算。未来用户显式运行demo/authorize是新的有界调用,不是后台持续任务。
---
## 历史:12小时仿真功能包自主任务(已提前完成)
用户明确授权本轮新代码/测试/串行RTX5080仿真/本地打包,要求最多12小时、零训练、不做RootXY/整机/实机/sim2real/通用adapter,同一问题最多两轮有依据修复复验,不改门槛/保护、不删失败证据。开始2026-10-15 17:38:07 +08,截止2026-10-16 05:38:07 +08;[固定计划](contact_cycle_sim_skill_v4_plan.md)。
本轮独立登记`/home/cen/sf-runs/simskill-v4-authority-01/registration.json`,运行共享serial锁与ledger;24Kit/650000物理调用/24GiB整体护栏,单run4GiB、原GPU归属/物理/截止/退出/清理保护全部保留,绝不复用旧grant。实际18Kit/261000调用/58周期,0训练/0物理失败重试;[已完成结果](contact_cycle_sim_skill_v4_result.md)。剩余预算不自动使用。
包的`authorize`是**未来调用方显式操作**,只生成全新≤12小时的有界本地会话登记,不启动仿真、不复用/延长本任务;真实运行仍各新grant/新配置绑定/新输出。不是无限后台执行授权,当前自主任务在交付后停止。
---
## 先前持续预算授权及v3已消费运行(历史保留)
用户最新明确指令:**“批准,以后你的预算是无限的,不需要我批准”**。
- 后续不再逐次询问预算批准;在当前用户目标范围内,自主固定并记录每个campaign的时间、物理调用、启动次数、存储及退出护栏,再执行。
- “不限总预算”不等于取消运行保护、放宽功能标准、无界后台任务或自动重复失败作业。仍需独立输出、当次源码/配置/设备身份和独立授权登记,不沿用消费过的grant。
- 当前任务仍零训练,不扩大到整机、硬件、sim2real、外扰、噪声或形态变化。精度门槛不提高,30mm根XY仅诊断;所有原数值/关节/执行器/失控和GPU归属保护继续启用。
- 先u256原命令连续3周期完整补验,再固定六组成对矩阵。首个功能、安全、配对或运行异常停止当次campaign,保留现场及剩余not_run;不因“预算无限”自动重试。
- 当前`ff-v3-01`仍严格沿用用户已经批准的预先固定上限:1次零物理准备+13次物理启动,178750共享物理调用,6540s,own2GiB,单环境RTX5080,0优化。
本轮原指令、实际身份/预算及新授权保存在`/home/cen/sf-runs/ff-v3-authority-01/approval.json`。此文件表达持续授权范围,不是可重用运行grant,也不是物理验证结果。
+116
View File
@@ -0,0 +1,116 @@
# Go2 FL仿真功能接口:0.4.1实时GUI扩展(非严格accepted_skill)
## 0.4.1:实时窗口与headless明确分开
公开Python增加严格布尔`gui=False`参数;CLI增加`--gui`。默认仍无窗口,控制/物理范围与0.4.0完全相同。示例:`start(NEW_OUTPUT, NEW_REGISTRATION, sequence='S1', controller='analytic', gui=True)`。`gui`进入精确spec、approval及认证FD,worker仅RUN使用`AppLauncher(visualizer='kit', headless=False)`;PREP始终headless/零环境/零物理。GUI依赖在PREP/RUN预导入并绑定,仅演示配置增加KitVisualizerCfg,不改物理参数、原render_interval20或控制步长。
独立新包的`bash demo_gui.sh`是一条实时GUI启动命令:真实Kit窗口,默认解析S1。`bash demo.sh`仍真实仿真但没有窗口;`bash verify_functional.sh`为原完整headless功能验证。`demo/measured_state_animation.gif`若查看0.4.0中的该文件,则只是**历史离线重建**,不是GUI也不是Kit摄像机录像;0.4.1真实截图/短录像路径见该包README。
界面显示真实模式、命令、6相位及进度、周期与tick、接触Fn/gap/足端位置。相机/灯光只修改新视觉prim;蓝色3kg载荷是原资产刚体,不额外创建假载荷。原500Hz安全/记录先于25Hz渲染更新,KitVisualizer在app.update期间禁用player物理,保持窗口响应而不额外步进。
完整序列或显式安全停止后,`query().state`在查看期为`viewing_final`;这不是最终监督通过。保留原场景/最终画面(最多10分钟),无额外物理步;关闭主窗口或Close按钮后才完成preclose身份核对、实际wait和清理。若先到600秒查看上限,则有界退出并报告`view_timeout`,不把它标成用户关窗;原总截止/存储等guard优先,若它们更早触发仍按原规则失败退出,不保证自动查看超时总会先到。提前关窗会中止并保存部分轨迹、非零退出,原因`gui_window_closed_before_completion`(不等待下一周期、不冒充protocol_complete)。显式公开`request_stop()`仍是原完整reload边界的安全停止,不等于关窗或完整矩阵通过。
GUI receipt单独记录真实render更新、完成标志、关闭原因、最终查看carry/counter不变;截图/桌面可见性仍须外部实测,receipt本身不证明窗口真的可见。启动早期AppLauncher尚未返回时的关窗尚未实测;任何缺失worker/完整证据均不能通过监督。0.4.0全部原包/证据保留;新版本只对本轮实际列出的结果重新验证,不自动继承为全部GUI命令/u256资格。
---
## 原有功能API与约束(继续保留)
This is the existing aided, floating-root Go2 FL single-leg simulation with the original IK, contact logic, residual bound, fixtures and functional/safety gates. Analytic is zero residual; u256 is the unchanged deterministic frozen actor **plus** that analytic reference. No training. CPU qualification is not physical validation.
## Public entrypoints (no Kit in caller)
From the repository or a same-layout independent copy:
```python
from scripts.go2_fl_skill import start, SkillRun
run = start('/absolute/NEW/run', '/absolute/current-task/registration.json',
command='x_bias-16mm', cycles=6, controller='analytic')
status = run.query() # detached JSON snapshot, no tensor aliases
request_id = run.queue('y_bias-24mm') # one outstanding request; may raise FileExistsError
stop_id = run.request_stop() # request, not immediate termination
exit_code = run.wait() # waits for supervisor, which waits/cleans Kit
# Existing process: SkillRun('/absolute/NEW/run').query()/queue()/request_stop()
```
`cycles` is an exact integer 1..20 (not bool). IDs are `validated_nominal` (+25,+20 mm) and `{x_bias,baseline_direction,y_bias}-{16,24}mm`. Directions are normalized world-XY ratios (25,10), (25,20), (15,20). No arbitrary vectors, durations, controllers, resets or randomization. Controllers are `analytic` or `u256`. Physical u256 execution requires complete current-source nominal20/S1/S2/S3 analytic successes supplied using `--analytics-evidence`. Prerequisites must have full requested counts and exact fixed pair-command history; an early safe stop or changed schedule never qualifies. Both controllers accept the same finite command/cycle interface. Historical 0.4.0 physically qualified u256 only on S1, not every combination or 20 cycles; 0.4.1 has not requalified u256/headless or u256/GUI. Its current-source analytic prerequisites remain mandatory.
`sequence=` supplies the entire initial command/cycles/queue; do not combine it with custom command/cycles/queue overrides. Frozen schedules:
- `nominal20`: nominal, 20 continuous cycles.
- `S1`: x_bias16 → baseline_direction24 → y_bias16, six cycles.
- `S2`: y_bias24 → x_bias24 → baseline_direction16, six cycles.
- `S3`: baseline_direction24 → y_bias24 → x_bias16, six cycles.
A is the initial **commanded** ground target, not measured landing; root target remains the initial commanded port. Each command owns an A→B(c)→A pair. Queued commands append FIFO (ten pending maximum) and apply only after an admitted return-to-A endpoint (completed cycles 2,4,...). Initial cycle zero is set by launch arguments; requests arriving during preparation queue for the next A endpoint, not an extra initial reference refresh. A pair's B never changes mid-pair, targets never accumulate, and history remains available for every recorded sample. Queueing never extends the requested cycle limit; stopping or reaching that limit can leave queued commands unexecuted. Stop is honored at **any** complete, audited, reloaded cycle endpoint, before seam admission/reference refresh; immediate safety failure always wins. Early completion is `requested_stop`, not `protocol_complete` for the requested maximum.
Invalid commands are rejected before any queue/control mutation. Runtime also validates exact mailbox fields. One atomic outstanding mailbox request, max 1024 input bytes, 64 receipts/run; malformed/overflow input cannot become control. Client rejection does not create mailbox files. Runtime rejection receipts may update audit metadata, never queue/control state. The mailbox is a bounded same-user local seam, not a hostile-user security sandbox.
`query()` reports state, authoritative global/cycle/local ticks, actual phase/progress, latched contact/normal force/gap, measured foot, commanded target/fixed anchors, pending/active IDs, stop request and failure. Physical status is copied every 500 physics ticks (one simulated second), at initialization, at endpoints and on failure; its tick identifies freshness. `receipt.json` records the latest consumed request immediately; `last_receipt` in status is sampled. Do not mistake `starting` for initialized physical state, or `stopped` for supervisor success. Final success needs `supervisor_success=true` and the successful waited campaign result; early stop retains requested/completed counts separately. If the API-owned supervisor exits before mailbox creation, its handle reports failure.
## CPU validation only
```sh
/usr/bin/python3.12 scripts/go2_fl_skill.py validate
CUDA_VISIBLE_DEVICES='' /home/cen/miniconda3/envs/linker_sim_isaac/bin/python \
scripts/go2_fl_skill.py validate --verify-frozen
```
The second command calls unchanged `replay.verify_bundle`, including all 134 immutable snapshots/live dependencies and the exact u256 hash; only the preexisting AGENT.md exception remains. Neither command queries GPU, creates authority, launches Kit or advances physics.
## Parent's real run commands (fresh authority generated automatically)
For a future explicit local invocation, create a fresh finite session first (CPU only, no GPU/Kit):
```sh
/usr/bin/python3.12 scripts/go2_fl_skill.py authorize --registration /absolute/NEW-session/registration.json
```
Python equivalent: `scripts.skill_launch_v4.new_registration(Path('/absolute/NEW-session/registration.json'))`. This is explicit caller authorization, not automatic reuse/extension of the development task. The new session is capped at12 hours and retains the same launch/call/storage limits. A new explicit invocation uses its own fresh registration and deadline. It does not revive the expired 0.4.0 task or the completed 0.4.1 development task.
Use system Python3.12 for supervision. Each `run` creates **NEW** `<output>-authority/approval.json`, binding exact current source hashes, spec, fixed limits, current boot, current allowed desktop identities, admission window, registration hash and deadline. It then runs a new zero-physics config-preparation Kit and one physical Kit, serially. No old approvals, binding preparations, failure directories, resumes or retries are reused. Preparation is not physical qualification.
Choose never-used output names; these are examples, **not an instruction to reuse an existing root**. Shared registration is the parent's current task registration, not an old physical grant. Keep it identical across repository and independent-copy runs so they share `skill-v4.lock`/`skill-v4-ledger.json` in its parent directory.
```sh
set -eu
REPO=/path/to/current/repository-or-independent-copy
REG=/absolute/NEW-session/registration.json
P=/home/cen/sf-runs/skill-v4-NEW-PREFIX
CLI="$REPO/scripts/go2_fl_skill.py"
/usr/bin/python3.12 "$CLI" run --registration "$REG" --output "$P-nominal20" --sequence nominal20
/usr/bin/python3.12 "$CLI" run --registration "$REG" --output "$P-S1" --sequence S1
/usr/bin/python3.12 "$CLI" run --registration "$REG" --output "$P-S2" --sequence S2
/usr/bin/python3.12 "$CLI" run --registration "$REG" --output "$P-S3" --sequence S3
# Parent reviews all analytics before executing this gated run:
/usr/bin/python3.12 "$CLI" run --registration "$REG" --output "$P-u256-S1" --sequence S1 --controller u256 \
--analytics-evidence "$P-nominal20" --analytics-evidence "$P-S1" \
--analytics-evidence "$P-S2" --analytics-evidence "$P-S3"
# Public transport safety-stop probe: request after first 5 ticks, reserve 3 cycles; expect endpoint1.
/usr/bin/python3.12 "$CLI" run --registration "$REG" --output "$P-stop" --cycles 3 --stop-at-tick 5
# Public invalid-send probe: records unchanged metadata and enumerated carry/RNG digests; expect cycle1.
/usr/bin/python3.12 "$CLI" run --registration "$REG" --output "$P-invalid" --cycles 1 --reject-at-tick 5
```
For external live control from another CPU process:
```sh
/usr/bin/python3.12 "$CLI" status --output "$P-S1"
/usr/bin/python3.12 "$CLI" queue --output "$P-S1" x_bias-24mm
/usr/bin/python3.12 "$CLI" stop --output "$P-S1"
```
Do not inject external commands into frozen matrix evidence runs; use a separately reserved run for custom live-command evidence. For independent-copy S1 reproduction, change REPO to that copy and use another never-used output, same registration. Source hashes do not embed the development repository path; actual runtime binding captures the current copy's absolute paths.
## Limits, evidence and dependencies
- Whole task shared ledger pre-reserves 2 Kit launches/run and `4500*requested_cycles+250` physics calls, never refunds failed/unrun slots; maximum 24 launches/650000 calls. It holds the existing-task lock through cleanup to prevent parallel public runs. A fresh finite authority registration is required; its boottime deadline and the per-run deadline apply, never an unbounded grant. The historical 0.4.0 task registration expired2026-10-16 05:38:07 +08:00 and is not extended. The separate 0.4.1 GUI task records its own scope beginning2026-10-16 09:51:52 +08:00, with the same finite whole-task caps; its independent demo trials use fresh bounded sessions, additionally accounted together by the parent. Future user-invoked `authorize` creates a different bounded session without launching simulation.
- Preparation: 360s total, zero physics. Physical run: ≤6 cycles1200s, >6 up to20 cycles2400s, stop probe900s; each includes120s cleanup. Calls: nominal20 ≤90250, six-cycle ≤27250, three-cycle stop ≤13750. One env/RTX5080, optimizer0.
- Existing process identity, fail-closed NVML, observed GPU diagnostics, stdout cap, native monitor, storage checks, authenticated worker FD, preclose same-object config binding, wait and owned cleanup are reused. Per-run owned cap4GiB and free headroom8GiB; shared ledger checks actual run/authority directory bytes against24GiB. Parent must additionally account for its package/analysis outputs outside those run directories. Native growth monitoring remains sampled, not a hard quota; existing coverage gaps remain reported.
- Automatic desktop admission uses only exact parent-approved executable PATH+SHA256 pairs in `configs/contact_cycle/skill_v4_desktop_allowlist.json`; fresh PID/birth/argv/inode/hash captured each run, any compute or unapproved graphics rejects. Optional expert `--gpu-identity` supplies an explicit current desktop identity JSON, never an old physical grant; existing GPUOwnership still rechecks every identity/client.
- Results: `prepare/{binding,inventory}.json`, `preparation-reviewed.json`, source snapshot, per-slot reservation/stdout/worker/supervisor results; `run/{trace.pt,initial-state.pt,events-and-seams.json}`, cycle audits, actual/requested counts and live receipts/pair history; `mailbox/{status,receipt}.json`; native/GPU observations; final waited `campaign-result.json`. Full500Hz samples and100Hz controls, integer local-cycle windows, no thinning/splicing/reset/reanchor. Evidence distinguishes physical functional gates and runtime/supervisor integrity.
- Required same-layout repository files include all current `validate` source entries, original functional v2/v3 configs, frozen candidate `logs/contact_cycle/sim_residual_training_v1/prepared01/candidate_u0256.pt`, its exact `manifest.json` and complete134-file `source_snapshot/`, all corresponding live frozen files/assets/configs, and the replay source hashes. Parent assembles this snapshot; do not regenerate its manifest. Candidate SHA256 `c088bb3aecaca751867118858626fd508f7a3983303802d7524353f80802cf99`.
- Installed runtime remains the pinned `/home/cen/IsaacLab`, `/home/cen/isaacsim-6.0.1`, native numerical dependencies and RTX5080. Repository ROOT is relocatable; those installed dependencies are intentionally not made portable or re-pinned. The v4 runner explicitly sets `SKILLFACTORY_GO2_ASSET_DIR=<this ROOT>/assets/generated/go2_fl` and `SKILLFACTORY_LEG_BUNDLE=<this ROOT>/assets/generated/legs/go2_fl` before frozen import-time defaults run; retain the complete generated base Go2 FL, support and leg assets (including conversion.json/manifest.json/USD dependencies) and installed source pins. See `contact_cycle_runtime.md` for CPU namespace mapping and runtime constraints.
Historical 0.4.0 primary runs passed analytic20 continuous cycles, S1/S2/S3 six-cycle switches, u256 S1, requested stop at cycle1 and live invalid-command rejection, including preclose/GPU/exit/cleanup. Those are not a new-source/full-GUI matrix qualification for 0.4.1. See the separate v4 result/evidence index for independent-directory reproduction and precise acceptance; this interface document does not itself confer acceptance or general robustness/full-robot/real-world capability.
+38 -2
View File
@@ -1,6 +1,42 @@
# 下一阶段:完成监控资格验证,再决定新的有界RootXY试训
# 当前阶段:0.4.2性能与限定S1实测通过
这是待办,不是运行授权。当前没有合格RootXY候选,旧授权不可复用;不恢复u64、不追加旧账本、不扩大残差或放宽30mm门槛。
[0.4.2结果](contact_cycle_sim_performance_v042_result.md):高频真实GUI约34FPS、全段35.30绘制提交/墙钟秒;完整解析S1 GUI/headless各27000tick与0.4.1轨迹逐字节相同。0训练/不改原保护,剩余原生槽不用于扩训或扩域。独立包在`performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2/`;完成状态以工作根`task-complete.json`及包校验/库存凭据为准,不再启动仿真。
仍非1×实时(GUI0.07060×、headless0.18325×),GPU整卡约40%,不通过额外环境/虚假负载追求100%。进一步吞吐优化需另固定范围,不能牺牲逐步安全/证据或冒充已有资格。新绘制节拍的原生故障注入及完整矩阵/u256未验证;历史strict RootXY失败/accepted_skill=false不变。
---
# 历史阶段完成:0.4.1真实GUI演示与实际验证
[0.4.1结果](contact_cycle_sim_gui_v041_result.md):独立包`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.1/`,`bash demo_gui.sh`真实解析S1开窗,`bash demo.sh`仍无窗口,`bash verify_functional.sh`仍为原完整headless验证入口(0.4.1未重跑完整矩阵)。GUI/headless各六周期通过,活跃窗口提前关闭按预期中止;真实截图/原始录屏和原速剪辑见包demo。0.4.0原件/校验完全不变。
当前停止仿真,不自动使用剩余3个Kit槽。只建议直接使用已有仿真功能接口;新目标另固定有界范围。不追加训练、RootXY精度优化、控制扩域、整机/实机/sim2real或通用adapter。未实测项(极早关闭、GUI安全停止/超时、u256 GUI等)见结果,不把0.4.0历史矩阵升级成新版本资格。
---
## 历史:仿真功能版0.4.0技能包已本地交付
[本轮真实结果](contact_cycle_sim_skill_v4_result.md):Python/CLI、解析20连续周期、三条六周期切换、可选u256 S1、停止/非法拒绝、独立包公开接口真实复现、演示与监督退出清理全部完成。包在`/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.0/`;启动`bash demo.sh`,完整验证`bash verify_functional.sh`。每次显式调用创建新有限登记,不复用本轮已消费grant/输出。
下一阶段先使用现有仿真功能包集成上层调用;新需求再固定有限验证范围。当前不追加物理、训练、RootXY精度优化、整机/实机/sim2real或通用adapter。尚未验证u25620周期/其余切换、多seed/无辅助/全域,不把可调用接口等同于这些能力。历史严格验收与accepted_skill保持原值。
本轮12小时授权提前完成,未用预算不自动耗用;将来明确调用`authorize`仅新建有上限的本地会话,不自动启动任务或延长旧任务。
---
## 历史阶段:限定仿真功能与有限成对矩阵通过;不自动扩训
[补验v3实际结果](contact_cycle_functional_followup_v3_result.md):新u256原命令连续3周期完整通过,随后固定3方向×16/24mm的六组,解析/u256各3周期,全部物理/运行完整性通过,6对初态一致。实际175500共享物理调用、39周期、3585.201s,0训练/0重试;进程/GPU清理和原件核对通过。
解析零残差已经足以完成本阶段功能,u256没有提高有限矩阵功能通过率,不必为当前目标追加训练或RootXY精度优化;冻结u256保留为已验证对照,不更换历史权重/发布包。根XY超过30mm仅诊断,旧严格失败不改写。旧GPU失败仍缺瞬时列表,本次未复现不等于查明旧原因。
用户明确后续预算无需逐次请批,见[持续执行授权](execution_authority.md)。如有后续任务,在该目标范围内自主预设时间/调用/存储/退出护栏及独立登记,不复用已消费grant/目录;不再以“等待预算批准”阻塞。当前限定任务已完成,不自动拓展扰动、噪声、形态、整机/实机/sim2real。有限单seed矩阵不是全域/鲁棒性或独立技能验收,accepted_skill=false。
---
## 原RootXY下一阶段计划(历史保留,本轮不执行)
以下原待办不构成本轮授权或默认技术方向。当前没有合格RootXY候选,旧授权不可复用;不恢复u64、不追加旧账本、不扩大残差。原30mm门槛继续约束旧RootXY/严格P1,不移用于本轮独立功能版本。
## 1. 优先解决基础设施可靠性
+32
View File
@@ -0,0 +1,32 @@
# Version source helpers and compact audit evidence, not local native authority,
# frozen-runtime copies, private issuer/session data, video, caches or archives.
/*
!/.gitignore
!/README.md
!/*.py
!/plan.md
!/presentation_amendment.md
!/*review*.md
!/scope.json
!/staging.json
!/assembly.json
!/performance-result.json
!/performance-summary-pre-headless.json
!/*comparison.json
!/*review-extract.json
!/*postexit*.json
!/package-verification.json
!/task-complete.json
!/delivery-*.json
!/cpu-*.xml
!/delivery-cpu*.xml
!/cpu-*.log
!/delivery-cpu*.log
!/display05-physical-audit/
!/headless06-physical-audit/
/display05-physical-audit/*
/headless06-physical-audit/*
!/display05-physical-audit/actual-audit.json
!/display05-physical-audit/phase-summary.csv
!/headless06-physical-audit/actual-audit.json
!/headless06-physical-audit/phase-summary.csv
+31
View File
@@ -0,0 +1,31 @@
# 0.4.2 Git checkpoint / local delivery evidence
This directory versions performance helper source, the bound plan/presentation
amendment, independent reviews/resolutions, CPU results including original
failures, compact original receipt extracts, trace comparisons, physical audits,
and final package/whole-task verification receipts.
The measured result and limitations are in
[`../docs/contact_cycle_sim_performance_v042_result.md`](../docs/contact_cycle_sim_performance_v042_result.md).
`accepted_skill=false`; only current-source analytic S1 GUI/headless is qualified.
The native runs used the recorded staging path, not this Git commit as a new
native trial. Committing does not authorize another run or revive a closed grant.
## Local artifacts deliberately not committed
- `deliveries/go2_fl_contact_cycle_sim-0.4.2/` and adjacent `.tar.gz`/`.sha256`:
sealed source-identical runtime, original selected evidence and genuine PNG/
original-speed MP4. Archive SHA256:
`16c855f6086a90aedc3f50eb510d44f64f802254505146a61c83402df7d99151`.
- `runs/`, `staging/`, approved-source snapshots, large inventories and generated
native caches. Original raw evidence remains locally; compact records retain
original paths and hashes. No private issuer/session/authority files enter Git.
- Historical frozen candidate/snapshot/validation inputs and generated local
robot assets remain under the existing repository output policy. A fresh Git
clone alone is not the complete qualified-machine runtime; use the sealed
local delivery and documented Isaac Sim/IsaacLab dependencies.
`package-verification.json` and `task-complete.json` refer to the already sealed
local artifact; they do not attest that every file in the later Git checkpoint
was inside that archive. This README and output exclusion policy were added
only for the requested Git commit, without altering sealed delivery bytes.
+10
View File
@@ -0,0 +1,10 @@
#!/usr/bin/python3.12
"""Preserve first packaging CPU failure; add missing immutable test data only."""
import json,pathlib,shutil,hashlib
A=pathlib.Path(__file__).resolve().parent;D=A.parent;P=pathlib.Path(json.loads((A/'assembly.json').read_text())['package']);sha=lambda p:hashlib.sha256(p.read_bytes()).hexdigest();I=json.loads((P/'evidence-inventory.json').read_text());entries=[]
for n in ['logs/contact_cycle/evaluation_runner_recovery_v1/USER_STORAGE_SCOPE_AMENDMENT.md','logs/contact_cycle/sim_residual_training_v1/prepared01/validation_u0256_trace.pt']:
src=D/n;dst=P/n;assert not dst.exists();dst.parent.mkdir(parents=True,exist_ok=True);shutil.copy2(src,dst);entry={'original':str(src),'package':n,'sha256':sha(src),'bytes':src.stat().st_size};I.append(entry);entries.append(entry)
(P/'evidence-inventory.json').write_text(json.dumps(I,indent=2)+'\n')
auth=json.loads((A/'runs/display-S1-05/campaign/approval.json').read_text())['sources'];assert all(sha(P/n)==h for n,h in auth.items())
R={'first_package_CPU_result':'39failed1517passed7skipped, delivery-cpu.log/xml untouched','diagnosis':'36tests missing original historical storage-scope document;3tests missing original saved u0256 validation trace. Full bundled suite needs extra historical test data beyond minimal native runtime. No runtime source, safety rule, test assertion, skip or frozen weight changed.','added_original_fixtures':entries,'bound41_inputs_still_equal':True,'meaning':'These historical fixtures are data, not permission to reuse an authority or run training. New full CPU verification required before sealing.'}
(A/'delivery-fixture-repair.json').write_text(json.dumps(R,indent=2)+'\n');print(json.dumps(R))
@@ -0,0 +1,8 @@
#!/usr/bin/python3.12
import pathlib,json,hashlib,shutil
A=pathlib.Path(__file__).resolve().parent;D=A.parent;P=pathlib.Path(json.loads((A/'assembly.json').read_text())['package']);I=json.loads((P/'evidence-inventory.json').read_text());entries=[]
for src,n in [(D/'logs/contact_cycle/frozen_replay_p0_01/run/initial_state.pt','logs/contact_cycle/frozen_replay_p0_01/run/initial_state.pt'),(D/'logs/contact_cycle/frozen_replay_p0_01/run/first_episodes.pt','logs/contact_cycle/frozen_replay_p0_01/run/first_episodes.pt'),(pathlib.Path('/home/cen/sf-runs/p1-01/launch-0/trial-00/trace.v1.jsonl.gz'),'tests/fixtures/p1-trace.v1.jsonl.gz')]:
dst=P/n;assert not dst.exists();dst.parent.mkdir(parents=True,exist_ok=True);shutil.copy2(src,dst);entry={'original':str(src),'package':n,'sha256':hashlib.sha256(src.read_bytes()).hexdigest(),'bytes':src.stat().st_size};entries.append(entry);I.append(entry)
(P/'evidence-inventory.json').write_text(json.dumps(I,indent=2)+'\n')
(A/'delivery-replay-fixtures.json').write_text(json.dumps({'second_CPU':'1556passed7skipped, zero failures; one additional skip from missing P0replay fixture','action':'Include exact original P0initial/episodes and P1failuretrace as CPUfixtures. Portable bundled test locator points to embedded P1 trace; assertions/hash/tolerances unchanged. Not a native P1 rerun or promotion; no runtime source changed.','fixtures':entries},indent=2)+'\n')
print('copied3 historical replay fixtures')
+162
View File
@@ -0,0 +1,162 @@
#!/usr/bin/python3.12
"""Copy bounded qualified source/evidence; never mutate the staging run source."""
import pathlib,shutil,json,hashlib
A=pathlib.Path(__file__).resolve().parent;D=A.parent;S=pathlib.Path(json.loads((A/'staging.json').read_text())['root']);P=A/'deliveries/go2_fl_contact_cycle_sim-0.4.2'
assert not P.exists();P.parent.mkdir(parents=True,exist_ok=True)
shutil.copytree(S,P,ignore=shutil.ignore_patterns('__pycache__','.pytest_cache'))
# Documentation/tests not in the authority are refreshed; all bound inputs below
# must remain byte-for-byte those actually qualified, including the two plans.
for name in ['README.md','current_status.md','next_stage.md','evidence_index.md','contact_cycle_sim_performance_v042_result.md']:
shutil.copy2(D/'docs'/name,P/'docs'/name)
for src in (D/'tests').glob('test_contact_cycle*.py'):shutil.copy2(src,P/'tests'/src.name)
shutil.copy2(D/'docs/contact_cycle_sim_performance_v042_result.md',P/'RESULTS.md')
shutil.copy2(D/'setup-preflight.json',P/'setup-preflight.json')
E=P/'evidence/v042';E.mkdir(parents=True)
inventory=[]
def copy(src,dst):
assert src.is_file() and not src.is_symlink(),src
dst.parent.mkdir(parents=True,exist_ok=True);shutil.copy2(src,dst)
inventory.append({'original':str(src),'package':str(dst.relative_to(P)),'sha256':hashlib.sha256(src.read_bytes()).hexdigest(),'bytes':src.stat().st_size})
# Small parent receipts, reviews, tests and reproducibility helpers; no recursive
# capture of this delivery, caches, external private session directories or TUI.
for src in sorted(A.iterdir()):
if src.is_file() and src.suffix in {'.json','.md','.log','.xml','.py','.txt'}:copy(src,E/'parent'/src.name)
for src in sorted((A/'approved-sources').rglob('*')):
if src.is_file():copy(src,E/'approved-sources'/src.relative_to(A/'approved-sources'))
for src in sorted((A/'runs').rglob('*')):
if not src.is_file():continue
rel=src.relative_to(A/'runs');parts=rel.parts
if 'native' in parts or 'authority' in parts or 'tmp' in parts:continue
if len(parts)>1 and parts[1]=='session':
if src.name!='registration.json':continue # no private issuer/authority files
if src.name=='reservation.json' and any(x in parts for x in ['prepare','run']):continue
copy(src,E/'runs'/rel)
for name in ['display05-physical-audit','headless06-physical-audit']:
for src in sorted((A/name).rglob('*')):
if src.is_file():copy(src,E/name/src.relative_to(A/name))
# Original immutable CPU fixtures required by the complete bundled test suite,
# not runtime authority or a new training/physical run.
for n in ['logs/contact_cycle/evaluation_runner_recovery_v1/USER_STORAGE_SCOPE_AMENDMENT.md',
'logs/contact_cycle/sim_residual_training_v1/prepared01/validation_u0256_trace.pt',
'logs/contact_cycle/frozen_replay_p0_01/run/initial_state.pt',
'logs/contact_cycle/frozen_replay_p0_01/run/first_episodes.pt']:
copy(D/n,P/n)
copy(pathlib.Path('/home/cen/sf-runs/p1-01/launch-0/trial-00/trace.v1.jsonl.gz'),P/'tests/fixtures/p1-trace.v1.jsonl.gz')
# This is generated portable test output: locator only, original hash/assertions
# remain unchanged; the native authority does not include test files.
test=P/'tests/test_contact_cycle_sim_evaluation_runtime.py'
test.write_text(test.read_text().replace("trace = Path('/home/cen/sf-runs/p1-01/launch-0/trial-00/trace.v1.jsonl.gz')", "trace = ROOT / 'tests/fixtures/p1-trace.v1.jsonl.gz' # identical original hash, portable fixture"))
# Actual demo bytes, not an offline re-render or speed-adjusted derivative.
M=P/'demo';M.mkdir()
for name in ['actual-moving-window.png','actual-final-window.png','live-kit-window.mp4']:
copy(A/'runs/display-S1-05/observer'/name,M/name)
(P/'verify_cpu.py').write_text('''"""CPU namespace harness; no Kit registration or GPU. Requires pytest+CPU Torch."""
import sys,types,pathlib,pytest
root=pathlib.Path(__file__).resolve().parent
p=types.ModuleType('SkillFactory');p.__path__=[str(root/'source/SkillFactory/SkillFactory')];p.__file__=str(root/'source/SkillFactory/SkillFactory/__init__.py');sys.modules['SkillFactory']=p
sys.path[:0]=[str(root),str(root/'scripts'),str(root/'tests')]
raise SystemExit(pytest.main(sys.argv[1:]))
''')
(P/'README.md').write_text('''# Go2 FL contact-cycle simulation / performance0.4.2
Simulation-only, single floating-root aided Go2 FL leg + original3kg payload.
**accepted_skill=false.** This is a bounded analytic-S1 performance result, not
full-matrix/u256 qualification, autonomous balance, whole robot, hardware or sim2real.
Old0.4.0/0.4.1 and their failures remain sealed separately.
## Launch genuine live Kit GUI
From a local graphical terminal on the qualified machine:
```sh
cd /home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2
SKILL_DEMO_OUTPUT="$PWD/manual-$(date +%s)" bash demo_gui.sh
```
Original headless launcher: `bash demo.sh`. Python/CLI:
`python3 scripts/go2_fl_skill.py --help`, `docs/go2_fl_skill_v4_api.md`.
Public start/query/queue/request_stop/wait and analytic/reference mathematics,
original physical parameters/functional gates and full500Hz evidence remain.
Every future call creates NEW registration/guarded PREP+RUN; old closed grants
and this task's unused budget cannot be reused. Close while running aborts;
after full completion the original scene stays render-only until close/deadline.
The reused UI library title says0.4.1; its message identifies0.4.2candidate.
No video playback/interpolation is used by the launcher.
## Actual speed, not a real-time claim
- Genuine window screenshot `demo/actual-moving-window.png`: native overlay33.98FPS.
- FullS1 GUI:27000current-state render submissions in764.863wallseconds,
**35.30submissions/s**, including25650added +1350originalinterval20renders.
This is NOT independently measured display-present average FPS.
- **Physics time advances only0.07060× in GUI,0.18325× headless.**
Smooth display does NOT mean1× real-time simulation. More drawing trades
physics throughput for responsiveness; single-env small-operation/serial
synchronous checking overhead remains. Many-env training amortizes that
overhead and its aggregate throughput is not directly comparable.
- Whole-device mean active GPUutil39.74%,133.43W,2887.82MHz; includes desktop.
No artificial load, extra envs, power cap, lowered physics fidelity or fake FPS.
- Like-profiled identical4500tick baseline/combined loop113.5415→75.7835s:
1.4982×throughput,33.25%less walltime. Do not compare that profile directly
with the longer profiler-disabled fullS1 GUI run, which includes recording
and early CPU regression-test observer overhead.
- `demo/live-kit-window.mp4` is120seconds ORIGINAL-SPEED X11 capture,30Hz
sampling, not the renderer rate and not an input to the running demo.
## Evidence / qualification limits
GUI and headless each pass six continuous analyticS1 cycles/27000ticks,
original gates/5seams/preclose identity/shutdown/wait0/ownedcleanup.
All70tensor traces and whole-fileSHA256 equal prior0.4.1GUI11:
`d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba`.
Both zero-step offline physical audits pass. CPU1557passed/6skipped.
Actual native runs used the recorded staging root, not another GPU run of this
relocated delivery; all41authority source/config/plan bytes are identical,
source equivalence checked separately. Future roots need fresh binding.
See `RESULTS.md`, `evidence/v042/parent/performance-result.json`, original
campaign/worker/trace files under `evidence/v042/runs/`, source snapshots,
CPUlogs/XML, reviews/resolutions and independently verified physical audits.
Original native caches and private authority/session files are excluded from
this portable package, but retained under the original repository run paths;
`evidence-inventory.json` maps every copied evidence file to its origin/hash.
The failed first PREP, both fixed reviewP1findings, initial CPUcollection
failure and FIRST headless postexit observer GPU-admission exception remain
preserved. That observer's precise cause is UNKNOWN; original runtime success
is unchanged. A later unchanged strict verifier passes AFTER callerwait0 and
clean GPU observation; no whitelist/source waiver, new physical run or foreign
termination. Do not reinterpret the later check as the earlier failure's cause.
No current-source nominal20/S2/S3/u256 test. New-cadence native early-abort,
injected visualizer failure, startup-close, safe-stop and600s viewing timeout
are untested (CPUnegative tests are not native acceptance). No all-input
bitwise/global determinism or complete opaque PhysX-state claim. Historical
strict30mmRootXY failure remains: observedmax80.49mm; not precision optimization.
Shared native/cache monitoring is sampled/incomplete, not a hard disk quota.
`verify_functional.sh` can initiate a NEW fullmatrix; its presence is not a
claim that this version already passed it. Do not auto-run it as part of this task.
## Dependencies / integrity
Local IsaacLab `/home/cen/IsaacLab`, IsaacSim `/home/cen/isaacsim-6.0.1`,
RTX5080CUDA0 and graphical DISPLAY/XAUTHORITY. See `dependencies.json` and
`setup-preflight.json`; SO/AV unavailable/unneeded for application profiling,
no USD optimization/install is claimed. Existing hardware/control domain unchanged.
CPU-only verification (pytest+CPU Torch required):
`PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 CUDA_VISIBLE_DEVICES='' python verify_cpu.py tests/test_contact_cycle*.py -q`.
Public source/frozen check uses `scripts/go2_fl_skill.py validate --verify-frozen`.
`SHA256SUMS` and adjacent archive `.sha256` provide transport integrity, not new
runtime authority or broader skill acceptance. Final sealing/whole-task evidence
is recorded separately by the parent in the repository.
''')
ap=json.loads((A/'runs/display-S1-05/campaign/approval.json').read_text())['sources']
sha=lambda p:hashlib.sha256(p.read_bytes()).hexdigest()
assert all(sha(P/n)==h for n,h in ap.items()),'Relocated source differs from qualified authority'
prep=json.loads((A/'runs/display-S1-05/campaign/prepare/worker-result.json').read_text())
(P/'dependencies.json').write_text(json.dumps({'isaaclab_root':'/home/cen/IsaacLab','isaacsim_root':'/home/cen/isaacsim-6.0.1','kit_runtime_metadata':prep['kit_runtime_metadata'],'GPU':'RTX5080 CUDA0','driver_observed':'595.99.02','physical_caller':'/usr/bin/python3.12 publicCLI -> originalIsaacLab guarded native worker','portable_runtime_bundled':False,'standalone_SO_AV_required':False},indent=2)+'\n')
(P/'skill-package.json').write_text(json.dumps({'version':'0.4.2','accepted_skill':False,'scope':'bounded analyticS1simulation functional/performance result','qualified_authority_inputs':ap,'native_qualification_root':str(S),'delivery_root':str(P),'qualification':'GUI/headless6cycles27000ticks each;fulltraceidentical to0.4.1GUI11','limitations':'README.md / RESULTS.md','training_steps':0},indent=2)+'\n')
(P/'evidence-inventory.json').write_text(json.dumps(inventory,indent=2)+'\n')
(A/'assembly.json').write_text(json.dumps({'package':str(P),'copied_evidence_files':len(inventory),'authority_inputs':len(ap),'source_bytes_match_native_qualified_inputs':True,'sealed':False},indent=2)+'\n')
print(json.dumps({'package':str(P),'evidence_files':len(inventory),'source_inputs_equal':len(ap)}))
+7
View File
@@ -0,0 +1,7 @@
{
"package": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2",
"copied_evidence_files": 697,
"authority_inputs": 41,
"source_bytes_match_native_qualified_inputs": true,
"sealed": false
}
@@ -0,0 +1,14 @@
{
"left": "baseline-profile02",
"right": "combined-profile04",
"trace_sha256": [
"2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94",
"2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94"
],
"tensor_keys": 70,
"keysets_equal": true,
"all_bitwise_equal": true,
"unequal_keys": [],
"diagnostics": {},
"scope": "Measured native same-seed same-protocol traces only; not a promise of determinism elsewhere. No new physical steps."
}
@@ -0,0 +1,14 @@
{
"left": "baseline-profile02",
"right": "sensor-profile03",
"trace_sha256": [
"2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94",
"2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94"
],
"tensor_keys": 70,
"keysets_equal": true,
"all_bitwise_equal": true,
"unequal_keys": [],
"diagnostics": {},
"scope": "Measured native same-seed same-protocol traces only; not a promise of determinism elsewhere. No new physical steps."
}
+39
View File
@@ -0,0 +1,39 @@
{
"campaign_original_success": true,
"owned_identities_checked": 11,
"owned_alive": [],
"finite_bound_files_checked": 465,
"binding_file_mismatches": [],
"protected_file_mismatches": [],
"gpu_clients": {
"compute": [],
"graphics": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
]
},
"gpu_review": {
"uuid": "GPU-7d882fc3-f771-6d29-4e58-f0fa2a18a3a5",
"compute_pids": [],
"approved_desktop_graphics_pids": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
],
"task_graphics_pids": [],
"clean": true,
"scope": "owned-compute-and-graphics.plus-pinned-desktop-graphics-only"
},
"static_check_not_same_cfg_proof": true,
"boottime": 357443.021005604,
"passed": true
}
+15
View File
@@ -0,0 +1,15 @@
#!/usr/bin/python3.12
import pathlib,shutil,json,hashlib,subprocess
A=pathlib.Path(__file__).resolve().parent;D=A.parent;OLD=pathlib.Path('/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.1');P=A/'staging/go2_fl_contact_cycle_sim-0.4.2-work'
assert not P.exists();P.mkdir(parents=True)
skip={'evidence','demo','demos','README.md','RESULTS.md','skill-package.json','SHA256SUMS','dependencies.json','build-inventory.json'}
for p in OLD.iterdir():
if p.name in skip:continue
if p.is_dir():shutil.copytree(p,P/p.name,ignore=shutil.ignore_patterns('__pycache__','.pytest_cache'))
elif p.is_file():shutil.copy2(p,P/p.name)
old=json.loads(pathlib.Path('/home/cen/sf-runs/skill-gui-v041-gui-complete11/S1/approval.json').read_text())['sources']
extra=['source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/performance_v042.py','configs/contact_cycle/performance_v042.json','performance_v042/plan.md']
for n in list(old)+extra:
(P/n).parent.mkdir(parents=True,exist_ok=True);shutil.copy2(D/n,P/n)
(P/'README.md').write_text('0.4.2 profiling worktree only. Not a qualified release. See parent performance_v042/plan.md. Original0.4.0/1 preserved.\n')
(A/'staging.json').write_text(json.dumps({'root':str(P),'base':str(OLD),'source_files':list(old)+extra},indent=2)+'\n');print(P)
+29
View File
@@ -0,0 +1,29 @@
## Review
Paths below use `T = source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/`.
- **Correct — invocation-local reads/restoration:** `T/readonly_scope_v042.py:24–53` memoizes only listed no-argument reads, preserves instance overrides versus inherited methods, and restores them in `finally`. Clock/anchor identities and completed-physics/reset/pose counters are checked before normal return. Installation scopes only original safety evaluation; packed snapshot has its own separate scope (`:56–62`; `T/packed_recording_v042.py:94–98`).
- **Correct — physical sequencing/reset guards:** mount-reference routing requires source identity (`T/readonly_scope_v042.py:65–71`). Proxy reset still delegates to the guarded original sensor. Installation occurs before initial reset (`scripts/evaluate_cycle_functional_flow.py:255–258`); per-tick safety and synchronous evidence remain before another step (`T/evaluation_env.py:457–471`).
- **Correct — batching/layout:** `T/batched_kinematics_v042.py:12–17` orders probes as three positive then three negative perturbations per input row, subtracts corresponding pairs, and transposes into the original Jacobian column layout. Epsilon, IK iterations, solve formula and limits are not changed.
- **Correct — capture/provenance:** packing groups by exact device/dtype and uses blocking CPU transfer plus independently owned field storage (`T/packed_recording_v042.py:17–42`). Both new modules are preloaded, source-bound and authority-bound; qualification precedes environment construction and its record participates in exact PREP/RUN verification (`scripts/evaluate_cycle_functional_flow.py:68–69,118–124`; `source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_binding_v4.py:15–16,46–53`; `scripts/skill_launch_v4.py:30–31`).
- **Fixed:** None; read-only review.
### Finding — P1: sampled “bitwise” qualification uses numeric equality
**Location:** `T/batched_kinematics_v042.py:32–36`.
`torch.equal(old, new)` determines both `all_bitwise_equal` and whether the batched Jacobian is installed. It tests numeric equality, not representation equality: float32 `+0.0` and `-0.0` compare equal despite differing bits. Their absolute difference also records zero.
This contradicts the requested sampled-bitwise gate. It does **not** establish that actual FK samples currently differ; it establishes that the qualification cannot enforce its stated criterion.
**Smallest fix:** compare contiguous float32 representations using integer views, and add a signed-zero negative test plus qualification-fallback coverage. Retain the existing numeric maximum-difference report and explicit original-Jacobian selection.
**Merge verdict: BLOCK the combined qualification run pending this correction.** Other inspected seams are suitable for bounded runtime qualification, not final acceptance.
### Validation and residual risks
- Inspected supplied `performance_v042/cpu-combined.log` and XML: **55 passed**, zero failures/errors/skips. These are supplied results, not tests executed by this reviewer.
- Current `configs/contact_cycle/performance_v042.json:2` selects `sensor_epoch`, not `combined`. Freeze and authenticate the intended combined configuration before running; the current selection will not exercise the combined path.
- No edits, tests, shell commands, Kit/GPU operations, installs or subdelegation were performed.
- Native combined trace equivalence, reset behavior, gates and performance remain unqualified. No all-input bitwise or new performance claim is approved.
- Protected-artifact hashes and historical receipt integrity were not independently recomputed. `watchdog_diff` does not establish committed-range equivalence; review focused on the named candidate source seams and supplied artifacts.
@@ -0,0 +1,9 @@
# Combined review P1 resolution, before actual qualification
The review correctly identified numeric torch.equal as insufficient for the sampled bitwise Jacobian gate (signed zero). Fixed with contiguous float32 int32 representation comparison; dtype/shape/device must also match. Nonfinite math qualification now fails before environment construction. The numeric max-absolute-difference remains separately reported.
Added negative test using +0 and -0 Jacobian outputs: numeric equality holds, byte equality fails, sampled qualifier.applied=false/max_abs=0/reason=not_exact_keep_original, and the original method is retained. Current performance/math/original runtime targeted CPU suite:56passed, cpu-combined-fixed.log/.xml. No extra Kit or physical work for this fix. Native all-input equivalence remains unproved; actual complete traces must be compared.
Also changed the future trace comparator from numeric equality to uint8 representation equality (all dtypes), including signed zero. The existing baseline02/sensor03 trace files already have identical whole-file SHA256, independent of the old numeric comparator.
The first actual combined case uses a fresh approval/session and mode=combined/profile=true, not the prior sensor_epoch configuration. All39source/config/plan inputs are frozen in its own approval and preserved at next staging sync. No old package, native SDK source, physical/controller parameter, safety threshold, source verification, GPU admission or shutdown gate changed.
+39
View File
@@ -0,0 +1,39 @@
{
"campaign_original_success": true,
"owned_identities_checked": 11,
"owned_alive": [],
"finite_bound_files_checked": 469,
"binding_file_mismatches": [],
"protected_file_mismatches": [],
"gpu_clients": {
"compute": [],
"graphics": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
]
},
"gpu_review": {
"uuid": "GPU-7d882fc3-f771-6d29-4e58-f0fa2a18a3a5",
"compute_pids": [],
"approved_desktop_graphics_pids": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
],
"task_graphics_pids": [],
"clean": true,
"scope": "owned-compute-and-graphics.plus-pinned-desktop-graphics-only"
},
"static_check_not_same_cfg_proof": true,
"boottime": 359474.643529006,
"passed": true
}
+11
View File
@@ -0,0 +1,11 @@
#!/usr/bin/env python3
import argparse,json,pathlib,hashlib,torch
A=pathlib.Path(__file__).resolve().parent
p=argparse.ArgumentParser();p.add_argument('--left',required=True);p.add_argument('--right',required=True);args=p.parse_args();torch.set_num_threads(2)
paths=[A/'runs'/n/'campaign/run/trace.pt' for n in (args.left,args.right)];traces=[torch.load(p,map_location='cpu',weights_only=True) for p in paths];a,b=traces
unequal=[];diagnostics={}
for k in a:
if k not in b or a[k].dtype!=b[k].dtype or a[k].shape!=b[k].shape or not torch.equal(a[k].contiguous().reshape(-1).view(torch.uint8),b[k].contiguous().reshape(-1).view(torch.uint8)):
unequal.append(k)
if k in b and a[k].shape==b[k].shape:diagnostics[k]={'max_abs':float((a[k].double()-b[k].double()).abs().max()),'unequal_values':int((a[k]!=b[k]).sum())}
r={'left':args.left,'right':args.right,'trace_sha256':[hashlib.sha256(p.read_bytes()).hexdigest() for p in paths],'tensor_keys':len(a),'keysets_equal':a.keys()==b.keys(),'all_bitwise_equal':a.keys()==b.keys() and not unequal,'unequal_keys':unequal,'diagnostics':diagnostics,'scope':'Measured native same-seed same-protocol traces only; not a promise of determinism elsewhere. No new physical steps.'};out=A/(args.left+'--'+args.right+'-comparison.json');out.write_text(json.dumps(r,indent=2)+'\n');print(r)
+2
View File
@@ -0,0 +1,2 @@
........................................................ [100%]
56 passed in 8.06s
File diff suppressed because one or more lines are too long
+2
View File
@@ -0,0 +1,2 @@
....................................................... [100%]
55 passed in 8.30s
File diff suppressed because one or more lines are too long
+2
View File
@@ -0,0 +1,2 @@
.................................................................... [100%]
68 passed in 8.06s
File diff suppressed because one or more lines are too long
+3
View File
@@ -0,0 +1,3 @@
........................................................................ [ 93%]
..... [100%]
77 passed in 8.13s
File diff suppressed because one or more lines are too long
+45
View File
@@ -0,0 +1,45 @@
==================================== ERRORS ====================================
________ ERROR collecting tests/test_contact_cycle_live_cadence_v042.py ________
ImportError while importing test module '/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/tests/test_contact_cycle_live_cadence_v042.py'.
Hint: make sure your test modules/packages have valid Python names.
Traceback:
../../../miniconda3/envs/linker_sim_isaac/lib/python3.11/importlib/__init__.py:126: in import_module
return _bootstrap._gcd_import(name[level:], package, level)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
tests/test_contact_cycle_live_cadence_v042.py:6: in <module>
M=importlib.import_module('SkillFactory.tasks.manager_based.contact_cycle_sim.live_cadence_v042')
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
../../../miniconda3/envs/linker_sim_isaac/lib/python3.11/importlib/__init__.py:126: in import_module
return _bootstrap._gcd_import(name[level:], package, level)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
source/SkillFactory/SkillFactory/tasks/__init__.py:12: in <module>
from isaaclab_tasks.utils import import_packages
../../../opt/IsaacLab/source/isaaclab_tasks/isaaclab_tasks/__init__.py:33: in <module>
from .utils import import_packages
../../../opt/IsaacLab/source/isaaclab_tasks/isaaclab_tasks/utils/__init__.py:9: in <module>
from .parse_cfg import get_checkpoint_path, load_cfg_from_registry, parse_env_cfg
../../../opt/IsaacLab/source/isaaclab_tasks/isaaclab_tasks/utils/parse_cfg.py:17: in <module>
from isaaclab.envs import DirectRLEnvCfg, ManagerBasedRLEnvCfg
../../../opt/IsaacLab/source/isaaclab/isaaclab/envs/__init__.py:45: in <module>
from . import mdp, ui
../../../opt/IsaacLab/source/isaaclab/isaaclab/envs/mdp/__init__.py:18: in <module>
from .actions import * # noqa: F401, F403
^^^^^^^^^^^^^^^^^^^^^^
../../../opt/IsaacLab/source/isaaclab/isaaclab/envs/mdp/actions/__init__.py:8: in <module>
from .actions_cfg import *
../../../opt/IsaacLab/source/isaaclab/isaaclab/envs/mdp/actions/actions_cfg.py:8: in <module>
from isaaclab.controllers import DifferentialIKControllerCfg, OperationalSpaceControllerCfg
../../../opt/IsaacLab/source/isaaclab/isaaclab/controllers/__init__.py:14: in <module>
from .differential_ik import DifferentialIKController
../../../opt/IsaacLab/source/isaaclab/isaaclab/controllers/differential_ik.py:12: in <module>
from isaaclab.utils.math import apply_delta_pose, compute_pose_error
../../../opt/IsaacLab/source/isaaclab/isaaclab/utils/__init__.py:14: in <module>
from .mesh import *
../../../opt/IsaacLab/source/isaaclab/isaaclab/utils/mesh.py:14: in <module>
from pxr import Usd, UsdGeom
E ModuleNotFoundError: No module named 'pxr'
=========================== short test summary info ============================
ERROR tests/test_contact_cycle_live_cadence_v042.py
!!!!!!!!!!!!!!!!!!!! Interrupted: 1 error during collection !!!!!!!!!!!!!!!!!!!!
1 error in 1.10s
+38
View File
@@ -0,0 +1,38 @@
<?xml version="1.0" encoding="utf-8"?><testsuites name="pytest tests"><testsuite name="pytest" errors="1" failures="0" skipped="0" tests="1" time="1.099" timestamp="2026-10-16T14:25:01.261188+08:00" hostname="chenlin"><testcase classname="" name="tests.test_contact_cycle_live_cadence_v042" time="0.000"><error message="collection failure">ImportError while importing test module '/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/tests/test_contact_cycle_live_cadence_v042.py'.
Hint: make sure your test modules/packages have valid Python names.
Traceback:
../../../miniconda3/envs/linker_sim_isaac/lib/python3.11/importlib/__init__.py:126: in import_module
return _bootstrap._gcd_import(name[level:], package, level)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
tests/test_contact_cycle_live_cadence_v042.py:6: in &lt;module&gt;
M=importlib.import_module('SkillFactory.tasks.manager_based.contact_cycle_sim.live_cadence_v042')
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
../../../miniconda3/envs/linker_sim_isaac/lib/python3.11/importlib/__init__.py:126: in import_module
return _bootstrap._gcd_import(name[level:], package, level)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
source/SkillFactory/SkillFactory/tasks/__init__.py:12: in &lt;module&gt;
from isaaclab_tasks.utils import import_packages
../../../opt/IsaacLab/source/isaaclab_tasks/isaaclab_tasks/__init__.py:33: in &lt;module&gt;
from .utils import import_packages
../../../opt/IsaacLab/source/isaaclab_tasks/isaaclab_tasks/utils/__init__.py:9: in &lt;module&gt;
from .parse_cfg import get_checkpoint_path, load_cfg_from_registry, parse_env_cfg
../../../opt/IsaacLab/source/isaaclab_tasks/isaaclab_tasks/utils/parse_cfg.py:17: in &lt;module&gt;
from isaaclab.envs import DirectRLEnvCfg, ManagerBasedRLEnvCfg
../../../opt/IsaacLab/source/isaaclab/isaaclab/envs/__init__.py:45: in &lt;module&gt;
from . import mdp, ui
../../../opt/IsaacLab/source/isaaclab/isaaclab/envs/mdp/__init__.py:18: in &lt;module&gt;
from .actions import * # noqa: F401, F403
^^^^^^^^^^^^^^^^^^^^^^
../../../opt/IsaacLab/source/isaaclab/isaaclab/envs/mdp/actions/__init__.py:8: in &lt;module&gt;
from .actions_cfg import *
../../../opt/IsaacLab/source/isaaclab/isaaclab/envs/mdp/actions/actions_cfg.py:8: in &lt;module&gt;
from isaaclab.controllers import DifferentialIKControllerCfg, OperationalSpaceControllerCfg
../../../opt/IsaacLab/source/isaaclab/isaaclab/controllers/__init__.py:14: in &lt;module&gt;
from .differential_ik import DifferentialIKController
../../../opt/IsaacLab/source/isaaclab/isaaclab/controllers/differential_ik.py:12: in &lt;module&gt;
from isaaclab.utils.math import apply_delta_pose, compute_pose_error
../../../opt/IsaacLab/source/isaaclab/isaaclab/utils/__init__.py:14: in &lt;module&gt;
from .mesh import *
../../../opt/IsaacLab/source/isaaclab/isaaclab/utils/mesh.py:14: in &lt;module&gt;
from pxr import Usd, UsdGeom
E ModuleNotFoundError: No module named 'pxr'</error></testcase></testsuite></testsuites>
+49
View File
@@ -0,0 +1,49 @@
.......FF................................... [100%]
=================================== FAILURES ===================================
____ test_substep_two_keeps_offending_packed_frame_all_causes_and_no_render ____
make_env = <function make_env.<locals>.make at 0x76bc60275120>
def test_substep_two_keeps_offending_packed_frame_all_causes_and_no_render(make_env):
frames = [];pack = copy_function();renders = []
def evidence(event, s):
if event.kind != 'tick':return ()
f = pack({'value': torch.tensor([float('nan') if event.clock.global_tick == 2 else 1.])}, group_cpu=True)
frames.append(f)
return ('fault_one', 'fault_two') if not bool(torch.isfinite(f['value']).all()) else ()
env, hooks, _ = make_env(evidence=evidence)
env.cfg.sim.render_interval = 2;env.sim.is_rendering = True
env.sim.render = lambda **_: renders.append(True)
result = env.step(torch.zeros(1, 3))
assert result.stop.causes == ('fault_one', 'fault_two')
> assert env.sim.calls == hooks.rollout_steps == 2 and len(frames) == 2 and not renders
E assert (4 == 2)
E + where 4 = <test_contact_cycle_sim_evaluation_runtime.runtime.<locals>.Sim object at 0x76bc6026d7d0>.calls
E + where <test_contact_cycle_sim_evaluation_runtime.runtime.<locals>.Sim object at 0x76bc6026d7d0> = <SkillFactory.tasks.manager_based.contact_cycle_sim.evaluation_env.EvaluationContactCycleEnv object at 0x76bc6026ff90>.sim
E + and 2 = <SkillFactory.tasks.manager_based.contact_cycle_sim.evaluation_env.RuntimeHooks object at 0x76bc6036b450>.rollout_steps
tests/test_contact_cycle_performance_v042.py:140: AssertionError
_______________ test_transfer_exception_stops_before_third_step ________________
make_env = <function make_env.<locals>.make at 0x76bc60144540>
def test_transfer_exception_stops_before_third_step(make_env):
pack = copy_function();events = []
def evidence(event, s):
if event.kind == 'tick':
events.append(event.clock.global_tick)
if event.clock.global_tick == 2:pack({'bad': 1})
return ()
env, hooks, _ = make_env(evidence=evidence)
result = env.step(torch.zeros(1, 3))
> assert env.sim.calls == hooks.rollout_steps == 2 and events == [1, 2]
E assert (4 == 2)
E + where 4 = <test_contact_cycle_sim_evaluation_runtime.runtime.<locals>.Sim object at 0x76bc608743d0>.calls
E + where <test_contact_cycle_sim_evaluation_runtime.runtime.<locals>.Sim object at 0x76bc608743d0> = <SkillFactory.tasks.manager_based.contact_cycle_sim.evaluation_env.EvaluationContactCycleEnv object at 0x76bc60152110>.sim
E + and 2 = <SkillFactory.tasks.manager_based.contact_cycle_sim.evaluation_env.RuntimeHooks object at 0x76bc60808ad0>.rollout_steps
tests/test_contact_cycle_performance_v042.py:153: AssertionError
=========================== short test summary info ============================
FAILED tests/test_contact_cycle_performance_v042.py::test_substep_two_keeps_offending_packed_frame_all_causes_and_no_render
FAILED tests/test_contact_cycle_performance_v042.py::test_transfer_exception_stops_before_third_step
2 failed, 42 passed in 8.69s
File diff suppressed because one or more lines are too long
+23
View File
@@ -0,0 +1,23 @@
........................................................................ [ 4%]
........................................................................ [ 9%]
.........................................sss............................ [ 14%]
........................................................................ [ 18%]
........................................................................ [ 23%]
........................................................................ [ 28%]
........................................................................ [ 32%]
........................................................................ [ 37%]
........................................................................ [ 42%]
........................................................................ [ 47%]
........................................................................ [ 51%]
........................................................................ [ 56%]
.......................................s................................ [ 61%]
........................................................................ [ 65%]
.......s................................................................ [ 70%]
........................................................................ [ 75%]
....................................................s................... [ 80%]
........................................................................ [ 84%]
........................................................................ [ 89%]
........................................................................ [ 94%]
........................................................................ [ 98%]
.................. [100%]
1524 passed, 6 skipped in 114.98s (0:01:54)
File diff suppressed because one or more lines are too long
+23
View File
@@ -0,0 +1,23 @@
........................................................................ [ 4%]
........................................................................ [ 9%]
.....................................................sss................ [ 14%]
........................................................................ [ 18%]
........................................................................ [ 23%]
........................................................................ [ 28%]
........................................................................ [ 32%]
........................................................................ [ 37%]
........................................................................ [ 42%]
........................................................................ [ 46%]
........................................................................ [ 51%]
........................................................................ [ 56%]
...................................................s.................... [ 60%]
........................................................................ [ 65%]
...................s.................................................... [ 70%]
........................................................................ [ 74%]
................................................................s....... [ 79%]
........................................................................ [ 84%]
........................................................................ [ 88%]
........................................................................ [ 93%]
........................................................................ [ 98%]
.............................. [100%]
1536 passed, 6 skipped in 116.21s (0:01:56)
File diff suppressed because one or more lines are too long
+23
View File
@@ -0,0 +1,23 @@
........................................................................ [ 4%]
........................................................................ [ 9%]
........................................................................ [ 13%]
..sss................................................................... [ 18%]
........................................................................ [ 23%]
........................................................................ [ 27%]
........................................................................ [ 32%]
........................................................................ [ 36%]
........................................................................ [ 41%]
........................................................................ [ 46%]
........................................................................ [ 50%]
........................................................................ [ 55%]
........................................................................ [ 59%]
s....................................................................... [ 64%]
........................................s............................... [ 69%]
........................................................................ [ 73%]
........................................................................ [ 78%]
.............s.......................................................... [ 82%]
........................................................................ [ 87%]
........................................................................ [ 92%]
........................................................................ [ 96%]
................................................... [100%]
1557 passed, 6 skipped in 116.90s (0:01:56)
File diff suppressed because one or more lines are too long
+2
View File
@@ -0,0 +1,2 @@
............................................ [100%]
44 passed in 8.08s
File diff suppressed because one or more lines are too long
@@ -0,0 +1,6 @@
{
"preseal_observation": "First refresh asserted because generated bytecode cache remained after CPU verification. No runtime source mismatch; no archive had been sealed.",
"removed_only_package_generated_dirs": [
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/__pycache__"
]
}
+23
View File
@@ -0,0 +1,23 @@
........................................................................ [ 4%]
........................................................................ [ 9%]
........................................................................ [ 13%]
..sss................................................................... [ 18%]
........................................................................ [ 23%]
........................................................................ [ 27%]
........................................................................ [ 32%]
........................................................................ [ 36%]
........................................................................ [ 41%]
........................................................................ [ 46%]
........................................................................ [ 50%]
........................................................................ [ 55%]
........................................................................ [ 59%]
s....................................................................... [ 64%]
........................................s............................... [ 69%]
........................................................................ [ 73%]
........................................................................ [ 78%]
.............s.......................................................... [ 82%]
........................................................................ [ 87%]
........................................................................ [ 92%]
........................................................................ [ 96%]
................................................... [100%]
1557 passed, 6 skipped in 116.09s (0:01:56)
File diff suppressed because one or more lines are too long
+23
View File
@@ -0,0 +1,23 @@
........................................................................ [ 4%]
........................................................................ [ 9%]
........................................................................ [ 13%]
..sss................................................................... [ 18%]
........................................................................ [ 23%]
........................................................................ [ 27%]
........................................................................ [ 32%]
........................................................................ [ 36%]
........................................................................ [ 41%]
........................................................................ [ 46%]
........................................................................ [ 50%]
........................................................................ [ 55%]
........................................................................ [ 59%]
s....................................................................... [ 64%]
...........s............................s............................... [ 69%]
........................................................................ [ 73%]
........................................................................ [ 78%]
.............s.......................................................... [ 82%]
........................................................................ [ 87%]
........................................................................ [ 92%]
........................................................................ [ 96%]
................................................... [100%]
1556 passed, 7 skipped in 114.68s (0:01:54)
File diff suppressed because one or more lines are too long
File diff suppressed because it is too large Load Diff
File diff suppressed because one or more lines are too long
@@ -0,0 +1,20 @@
{
"first_package_CPU_result": "39failed1517passed7skipped, delivery-cpu.log/xml untouched",
"diagnosis": "36tests missing original historical storage-scope document;3tests missing original saved u0256 validation trace. Full bundled suite needs extra historical test data beyond minimal native runtime. No runtime source, safety rule, test assertion, skip or frozen weight changed.",
"added_original_fixtures": [
{
"original": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/logs/contact_cycle/evaluation_runner_recovery_v1/USER_STORAGE_SCOPE_AMENDMENT.md",
"package": "logs/contact_cycle/evaluation_runner_recovery_v1/USER_STORAGE_SCOPE_AMENDMENT.md",
"sha256": "e4fac0d36799c9beb67af7687a390a6b5e8390bde1d4b840a6e26c309578c8db",
"bytes": 1268
},
{
"original": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/logs/contact_cycle/sim_residual_training_v1/prepared01/validation_u0256_trace.pt",
"package": "logs/contact_cycle/sim_residual_training_v1/prepared01/validation_u0256_trace.pt",
"sha256": "bd77ade52943989f0bcc119da74812038d46284e0c1a01794d81db40827be438",
"bytes": 28030639
}
],
"bound41_inputs_still_equal": true,
"meaning": "These historical fixtures are data, not permission to reuse an authority or run training. New full CPU verification required before sealing."
}
@@ -0,0 +1,24 @@
{
"second_CPU": "1556passed7skipped, zero failures; one additional skip from missing P0replay fixture",
"action": "Include exact original P0initial/episodes and P1failuretrace as CPUfixtures. Portable bundled test locator points to embedded P1 trace; assertions/hash/tolerances unchanged. Not a native P1 rerun or promotion; no runtime source changed.",
"fixtures": [
{
"original": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/logs/contact_cycle/frozen_replay_p0_01/run/initial_state.pt",
"package": "logs/contact_cycle/frozen_replay_p0_01/run/initial_state.pt",
"sha256": "2e2b506b9d98d72cf64e32fcabfcc66561a694bb44a203408e95b960f29fa168",
"bytes": 18613
},
{
"original": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/logs/contact_cycle/frozen_replay_p0_01/run/first_episodes.pt",
"package": "logs/contact_cycle/frozen_replay_p0_01/run/first_episodes.pt",
"sha256": "d416095c66245f4c3359e619a04b0b092a2aca551cf0c23ca990b6b592d72149",
"bytes": 7117885
},
{
"original": "/home/cen/sf-runs/p1-01/launch-0/trial-00/trace.v1.jsonl.gz",
"package": "tests/fixtures/p1-trace.v1.jsonl.gz",
"sha256": "c943489ad035ca94a9f3edb0307113ed67af0d7bfc90543963f0eac699bfc6ee",
"bytes": 4543248
}
]
}
@@ -0,0 +1,122 @@
{
"execution_authorized": false,
"new_physics_calls": 0,
"commands": {
"validated_nominal": [
0.025,
0.02,
0.0
],
"x_bias-16mm": [
0.01485562705416415,
0.00594225082166566,
0.0
],
"x_bias-24mm": [
0.022283440581246223,
0.00891337623249849,
0.0
],
"baseline_direction-16mm": [
0.012493900951088486,
0.009995120760870788,
0.0
],
"baseline_direction-24mm": [
0.01874085142663273,
0.014992681141306183,
0.0
],
"y_bias-16mm": [
0.0096,
0.0128,
0.0
],
"y_bias-24mm": [
0.0144,
0.0192,
0.0
]
},
"sequences": {
"nominal20": [
"validated_nominal"
],
"S1": [
"x_bias-16mm",
"baseline_direction-24mm",
"y_bias-16mm"
],
"S2": [
"y_bias-24mm",
"x_bias-24mm",
"baseline_direction-16mm"
],
"S3": [
"baseline_direction-24mm",
"y_bias-24mm",
"x_bias-16mm"
]
},
"sources": {
"scripts/evaluate_cycle_functional_flow.py": "1eb8602d33ceb0284ca51ee6260a8f88ad699da38ae8e19cf6ec99320eada069",
"scripts/functional_flow_launch_v2.py": "597e87b6aae7a2a73fa9b380feb2b668b62cc4b72bc0b8d4d486bec4a8ad4e8e",
"scripts/evaluate_cycle_sim_residual.py": "a502ffd954be012ed5b038c6b5275536765d5a085b3d1c29e367e45ddea79acd",
"scripts/rootxy_launch_v2.py": "f7325fd2291ec18af61b6e10bbf136dc4df133b020d330d6b6709b4384a87aab",
"scripts/audit_contact_cycle_functional.py": "b9d9eaf51c0719d58ad0742a1055b2ad696ddc7e13d602ad354ee0c356f596c4",
"configs/contact_cycle/functional_flow_v2.json": "16c064228b30f36bbd420b7c2290602bacae6601cac2cec5deb16440f7600625",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_flow_v2.py": "9f4c84960d9dca07337527f41e8fa73fbdbf43968138ebc9872d6a3f06a79f1c",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_flow_binding_v2.py": "d47d99abde6d1944b6103007fcea0c9629cd6a299c7aeae06bf4a5c97ec05e9a",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/functional_flow_recording_v2.py": "5e577ceb7f5b060519e594e1994d73027e8a6913c745c5720987451e8cce56aa",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_env.py": "04e98455035e347633b7ea6c1fcf5372b4e6ac33645805f1beac8f7784e424c6",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_cfg.py": "62a83aa9976a4da00c49fa6d9dc61044ea705719047b2fd4f16e11fc671a3fa4",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_mdp.py": "89c46b2038733a21c97cb72a4112ff16b48ed0e15b5070ebdebcd3ce0cd8f7ac",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/recording.py": "8dc352f02a3cc84b4e2527f946a835d8c9ba5d4707d9545799cbe43c813969d0",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle/recording.py": "c718eba351d731bd4a6f864c9b8e47faeb70c7e1e583b8d108f654e0731f4224",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/evaluation_binding.py": "45c0c3c61cb968f19264398e1e6776630ed8ba8a799c7fea405a830248128fbf",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/evaluation_budget.py": "2cc58de32373df3d8508086cdaea5d7666f8d333f2822d401eccf21102225066",
"configs/contact_cycle/functional_followup_v3.json": "aaed45884eab9912db6c7bac1d82e2ab27784982c524a838b7886cdd763146f4",
"scripts/functional_followup_launch_v3.py": "9ed414f036b3aae61a4573d8cd9893be40f6c33c36e825c5de213dc05987afa3",
"scripts/functional_gpu_watch_v3.py": "c56d8e1e5e60f79ccec8d8ef1e656481d8bd316e6cde33d51b1decaf8a972484",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_followup_v3.py": "744ea0f83e7c160220959f565dd3f9ba52f7c3c224f8781afbf1d69ffb12430a",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_followup_binding_v3.py": "b10fb44971d6d267f9749de39a8e3f5b15b20cd1078df4a68e53689c4d2e539b",
"scripts/audit_cycle_functional_flow.py": "87a92b659d5a60605468c22079de49ace57d096d4f8ce89bb30d658d9234ac66",
"scripts/go2_fl_skill.py": "d95fe3c823c21c35ba45dec4221052959e3f096ef2ec311206c82774d08f12fa",
"scripts/skill_launch_v4.py": "61df4129f706b0f22d68471a81390180a1996a339d83cf49a89f37ee8340bb52",
"docs/contact_cycle_sim_skill_v4_plan.md": "ab552aa76dc30420557f4e82519fa2531677261ed7a63c5128dd22a6029c71c6",
"configs/contact_cycle/skill_v4_desktop_allowlist.json": "af8c54b2901642e6a9e9fca41e094ec95ef09dd714ccfb74a2a2869fcf92ad8d",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_v4.py": "1cfa556a44ff749677bf030dd61792fbe85d60e424646844bdace05794fd102c",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_mailbox_v4.py": "53f742f2c6d6c7ad4fbae5f1fcd1d1a41be1845567b8a0fdd4b080bf6e13fb92",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_binding_v4.py": "3efb8a8b3aac7a18111e490f2841f5205d275c99aa350116d50e15388d5de26e",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/skill_recording_v4.py": "7daf5b8eb9f84c5d2dc7ac915d16771235583bb336161339003e811c4e59770c",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/skill_gui_v041.py": "688068d6a1771076d087dc3deb4414f04922ca0bc8a296555495a07d19f57751",
"docs/contact_cycle_sim_gui_v041_plan.md": "508af7cbf5ea9b01a9f66871ed0c041c1c195e9adfb12e8289fb78b79860cb5f",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/performance_v042.py": "49bc36b4dae71141914ec5d99aa12379f5a5ce22143968dd7211bb8cbb44d47d",
"configs/contact_cycle/performance_v042.json": "039dd6b7e795e7fdac9e300a91975479bd6a93a664130acd5c18a9ed84d0aaf3",
"performance_v042/plan.md": "1d59f67d4fc56896f63aafd8fd345ec9ebccd19bce207543da8ba90783e87c4d",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/sensor_epoch_v042.py": "cd2edb7ca336f4f80a1191d6967833f84d5598fbb6712efe51c7a24df607f71a",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/packed_recording_v042.py": "c929d294623bc6b21b1f34839c7a0d59d484d63b4b35623ebe0ddbbb4a21578e",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/readonly_scope_v042.py": "024258ed6aeae33715eab0afebff92b9eb41ad97621c67f267834c6813991b7c",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/batched_kinematics_v042.py": "f41d12f48e800177520e6dec5a41137e27d4d6e80c4314026e0533f4074f7d97",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/live_cadence_v042.py": "3254a54a673fb3f27bf24f884210bc49c69c58b4fa2ac633a63ea4d699056915",
"performance_v042/presentation_amendment.md": "bbcdf6099b57bd26645674b1032e1e52e730a79cadc057cc62b88317eabdb89f"
},
"frozen": {
"candidate_path": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2/logs/contact_cycle/sim_residual_training_v1/prepared01/candidate_u0256.pt",
"candidate_sha256": "c088bb3aecaca751867118858626fd508f7a3983303802d7524353f80802cf99",
"manifest_sha256": "a70a226f176104414d2da295f08cce230f6b91ed3e2b78490c78f7ffce95ab01",
"snapshot_files_verified": 134,
"live_source_exceptions": [
{
"path": "AGENT.md",
"snapshot_sha256": "cec5e09fb709a88759e3c8b9ec9e17c95fc6bad3a24cc99b8617d2c9531dad8e",
"live_sha256": "d25b6c9db8f7f61adc2837ca227c46ad9b40f177f153346b1bfdb8d14891be91",
"reason": "Parent-approved non-runtime instruction drift only; snapshot unchanged"
}
],
"replay_sources": {
"scripts/replay_cycle_sim_residual.py": "36b44925135c5452341f43737004217a2d9036fef290a23e48aa2526c1f0aedd",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/replay.py": "e3dc56513065df6c119296eca15c2370df244674ccbe935199fcacc655448639"
}
}
}
+122
View File
@@ -0,0 +1,122 @@
{
"execution_authorized": false,
"new_physics_calls": 0,
"commands": {
"validated_nominal": [
0.025,
0.02,
0.0
],
"x_bias-16mm": [
0.01485562705416415,
0.00594225082166566,
0.0
],
"x_bias-24mm": [
0.022283440581246223,
0.00891337623249849,
0.0
],
"baseline_direction-16mm": [
0.012493900951088486,
0.009995120760870788,
0.0
],
"baseline_direction-24mm": [
0.01874085142663273,
0.014992681141306183,
0.0
],
"y_bias-16mm": [
0.0096,
0.0128,
0.0
],
"y_bias-24mm": [
0.0144,
0.0192,
0.0
]
},
"sequences": {
"nominal20": [
"validated_nominal"
],
"S1": [
"x_bias-16mm",
"baseline_direction-24mm",
"y_bias-16mm"
],
"S2": [
"y_bias-24mm",
"x_bias-24mm",
"baseline_direction-16mm"
],
"S3": [
"baseline_direction-24mm",
"y_bias-24mm",
"x_bias-16mm"
]
},
"sources": {
"scripts/evaluate_cycle_functional_flow.py": "1eb8602d33ceb0284ca51ee6260a8f88ad699da38ae8e19cf6ec99320eada069",
"scripts/functional_flow_launch_v2.py": "597e87b6aae7a2a73fa9b380feb2b668b62cc4b72bc0b8d4d486bec4a8ad4e8e",
"scripts/evaluate_cycle_sim_residual.py": "a502ffd954be012ed5b038c6b5275536765d5a085b3d1c29e367e45ddea79acd",
"scripts/rootxy_launch_v2.py": "f7325fd2291ec18af61b6e10bbf136dc4df133b020d330d6b6709b4384a87aab",
"scripts/audit_contact_cycle_functional.py": "b9d9eaf51c0719d58ad0742a1055b2ad696ddc7e13d602ad354ee0c356f596c4",
"configs/contact_cycle/functional_flow_v2.json": "16c064228b30f36bbd420b7c2290602bacae6601cac2cec5deb16440f7600625",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_flow_v2.py": "9f4c84960d9dca07337527f41e8fa73fbdbf43968138ebc9872d6a3f06a79f1c",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_flow_binding_v2.py": "d47d99abde6d1944b6103007fcea0c9629cd6a299c7aeae06bf4a5c97ec05e9a",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/functional_flow_recording_v2.py": "5e577ceb7f5b060519e594e1994d73027e8a6913c745c5720987451e8cce56aa",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_env.py": "04e98455035e347633b7ea6c1fcf5372b4e6ac33645805f1beac8f7784e424c6",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_cfg.py": "62a83aa9976a4da00c49fa6d9dc61044ea705719047b2fd4f16e11fc671a3fa4",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_mdp.py": "89c46b2038733a21c97cb72a4112ff16b48ed0e15b5070ebdebcd3ce0cd8f7ac",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/recording.py": "8dc352f02a3cc84b4e2527f946a835d8c9ba5d4707d9545799cbe43c813969d0",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle/recording.py": "c718eba351d731bd4a6f864c9b8e47faeb70c7e1e583b8d108f654e0731f4224",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/evaluation_binding.py": "45c0c3c61cb968f19264398e1e6776630ed8ba8a799c7fea405a830248128fbf",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/evaluation_budget.py": "2cc58de32373df3d8508086cdaea5d7666f8d333f2822d401eccf21102225066",
"configs/contact_cycle/functional_followup_v3.json": "aaed45884eab9912db6c7bac1d82e2ab27784982c524a838b7886cdd763146f4",
"scripts/functional_followup_launch_v3.py": "9ed414f036b3aae61a4573d8cd9893be40f6c33c36e825c5de213dc05987afa3",
"scripts/functional_gpu_watch_v3.py": "c56d8e1e5e60f79ccec8d8ef1e656481d8bd316e6cde33d51b1decaf8a972484",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_followup_v3.py": "744ea0f83e7c160220959f565dd3f9ba52f7c3c224f8781afbf1d69ffb12430a",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_followup_binding_v3.py": "b10fb44971d6d267f9749de39a8e3f5b15b20cd1078df4a68e53689c4d2e539b",
"scripts/audit_cycle_functional_flow.py": "87a92b659d5a60605468c22079de49ace57d096d4f8ce89bb30d658d9234ac66",
"scripts/go2_fl_skill.py": "d95fe3c823c21c35ba45dec4221052959e3f096ef2ec311206c82774d08f12fa",
"scripts/skill_launch_v4.py": "61df4129f706b0f22d68471a81390180a1996a339d83cf49a89f37ee8340bb52",
"docs/contact_cycle_sim_skill_v4_plan.md": "ab552aa76dc30420557f4e82519fa2531677261ed7a63c5128dd22a6029c71c6",
"configs/contact_cycle/skill_v4_desktop_allowlist.json": "af8c54b2901642e6a9e9fca41e094ec95ef09dd714ccfb74a2a2869fcf92ad8d",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_v4.py": "1cfa556a44ff749677bf030dd61792fbe85d60e424646844bdace05794fd102c",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_mailbox_v4.py": "53f742f2c6d6c7ad4fbae5f1fcd1d1a41be1845567b8a0fdd4b080bf6e13fb92",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_binding_v4.py": "3efb8a8b3aac7a18111e490f2841f5205d275c99aa350116d50e15388d5de26e",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/skill_recording_v4.py": "7daf5b8eb9f84c5d2dc7ac915d16771235583bb336161339003e811c4e59770c",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/skill_gui_v041.py": "688068d6a1771076d087dc3deb4414f04922ca0bc8a296555495a07d19f57751",
"docs/contact_cycle_sim_gui_v041_plan.md": "508af7cbf5ea9b01a9f66871ed0c041c1c195e9adfb12e8289fb78b79860cb5f",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/performance_v042.py": "49bc36b4dae71141914ec5d99aa12379f5a5ce22143968dd7211bb8cbb44d47d",
"configs/contact_cycle/performance_v042.json": "039dd6b7e795e7fdac9e300a91975479bd6a93a664130acd5c18a9ed84d0aaf3",
"performance_v042/plan.md": "1d59f67d4fc56896f63aafd8fd345ec9ebccd19bce207543da8ba90783e87c4d",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/sensor_epoch_v042.py": "cd2edb7ca336f4f80a1191d6967833f84d5598fbb6712efe51c7a24df607f71a",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/packed_recording_v042.py": "c929d294623bc6b21b1f34839c7a0d59d484d63b4b35623ebe0ddbbb4a21578e",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/readonly_scope_v042.py": "024258ed6aeae33715eab0afebff92b9eb41ad97621c67f267834c6813991b7c",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/batched_kinematics_v042.py": "f41d12f48e800177520e6dec5a41137e27d4d6e80c4314026e0533f4074f7d97",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/live_cadence_v042.py": "3254a54a673fb3f27bf24f884210bc49c69c58b4fa2ac633a63ea4d699056915",
"performance_v042/presentation_amendment.md": "bbcdf6099b57bd26645674b1032e1e52e730a79cadc057cc62b88317eabdb89f"
},
"frozen": {
"candidate_path": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2/logs/contact_cycle/sim_residual_training_v1/prepared01/candidate_u0256.pt",
"candidate_sha256": "c088bb3aecaca751867118858626fd508f7a3983303802d7524353f80802cf99",
"manifest_sha256": "a70a226f176104414d2da295f08cce230f6b91ed3e2b78490c78f7ffce95ab01",
"snapshot_files_verified": 134,
"live_source_exceptions": [
{
"path": "AGENT.md",
"snapshot_sha256": "cec5e09fb709a88759e3c8b9ec9e17c95fc6bad3a24cc99b8617d2c9531dad8e",
"live_sha256": "d25b6c9db8f7f61adc2837ca227c46ad9b40f177f153346b1bfdb8d14891be91",
"reason": "Parent-approved non-runtime instruction drift only; snapshot unchanged"
}
],
"replay_sources": {
"scripts/replay_cycle_sim_residual.py": "36b44925135c5452341f43737004217a2d9036fef290a23e48aa2526c1f0aedd",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/replay.py": "e3dc56513065df6c119296eca15c2370df244674ccbe935199fcacc655448639"
}
}
}
@@ -0,0 +1,14 @@
{
"left": "display-S1-05",
"right": "headless-S1-06",
"trace_sha256": [
"d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba",
"d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba"
],
"tensor_keys": 70,
"keysets_equal": true,
"all_bitwise_equal": true,
"unequal_keys": [],
"diagnostics": {},
"scope": "Measured native same-seed same-protocol traces only; not a promise of determinism elsewhere. No new physical steps."
}
@@ -0,0 +1,151 @@
{
"originals": {
"/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/runs/display-S1-05/campaign/run/worker-result.json": "668473b107926954758b0b85da0b0de2c882e5169101c20f38bf4943a8a21dd1",
"/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/runs/display-S1-05/campaign/campaign-result.json": "38cd9d5dcc45c0dca02d880d475c81efa6725c75657cd18d1d64ca7e5fa6f130",
"/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/runs/display-S1-05/observer/result.json": "96de9a8dead726adac1a4765b4842d1c83cabccb7496b889232cecfac800009c"
},
"worker": {
"cycles_completed": 6,
"gui": {
"camera": "/World/SkillDemoCamera",
"close_reason": "window_close",
"enabled": true,
"hold_no_physics": true,
"last_tick": 27000,
"max_final_view_seconds": 600,
"protocol_completed": true,
"render_updates": 27059,
"source": "live PhysX and actual state; no playback; original 3kg payload mesh"
},
"hook_violations": [],
"performance_installation": {
"constructor_sensor_references_routed": [
"mount_wrench"
],
"implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.sensor_epoch_v042.EpochSensor",
"jacobian": {
"all_bitwise_equal": true,
"applied": true,
"epsilon_rad": 0.0005,
"ik_parameters_changed": false,
"max_abs": 0.0,
"qualification": "sampled same-device same-dtype original-vs-batched math; not all-input proof",
"reason": "exact_samples",
"samples": 33
},
"safety_read_scope": true,
"sensors": [
"foot_contact",
"mount_wrench"
]
},
"physics_counts": {
"attempted_steps": 27000,
"completed_steps": 27000,
"composer_buffer_replacements": 27000,
"initialization_steps": 0,
"pose_write_calls": 14,
"reset_calls": 22,
"rollout_steps": 27000
},
"preclose_identity_verified": true,
"runtime_continuity_verified": true,
"runtime_stop": {
"all_sim_steps_attempted": 27000,
"all_sim_steps_completed": 27000,
"causes": [],
"clock": {
"cycle_index": 5,
"global_tick": 27000,
"local_tick": 4500
},
"completed_clock_cycles": 6,
"completed_controls": 5400,
"composer_buffer_replacements": 27000,
"fatal": false,
"initialization_steps": 0,
"partial_control_ticks": 0,
"physics_ticks": 27000,
"pose_write_calls": 14,
"reason": "protocol_complete",
"reset_calls": 22,
"rollout_steps": 27000,
"trial_index": 0
},
"shutdown_handoff_ready": true,
"success": true
},
"campaign": {
"success": true,
"actual_final_physics_calls": 27000,
"reserved_physics_calls": 27250,
"accepted_skill": false,
"optimizer_steps": 0
},
"rows": [
{
"slot": "prepare",
"success": true,
"returncode": 0,
"waited_exit": true,
"cleanup_clean": true,
"shutdown_handoff": {
"last_live_gpu": {
"approved_desktop_graphics_pids": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
],
"clean": false,
"compute_pids": [
954474
],
"scope": "owned-compute-and-graphics.plus-pinned-desktop-graphics-only",
"task_graphics_pids": [
954474
],
"uuid": "GPU-7d882fc3-f771-6d29-4e58-f0fa2a18a3a5"
},
"released": true,
"scope": "No physics after receipt; bounded owned cleanup then unchanged strict final GPU check",
"worker_result_sha256": "edd32a18be7cc48b7d03c18cf8dfd967eb2aa7143dc31520bd45c6b1a215f48b"
}
},
{
"slot": "run",
"success": true,
"returncode": 0,
"waited_exit": true,
"cleanup_clean": true,
"shutdown_handoff": {
"last_live_gpu": {
"approved_desktop_graphics_pids": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
],
"clean": false,
"compute_pids": [
956416
],
"scope": "owned-compute-and-graphics.plus-pinned-desktop-graphics-only",
"task_graphics_pids": [
956416
],
"uuid": "GPU-7d882fc3-f771-6d29-4e58-f0fa2a18a3a5"
},
"released": true,
"scope": "No physics after receipt; bounded owned cleanup then unchanged strict final GPU check",
"worker_result_sha256": "668473b107926954758b0b85da0b0de2c882e5169101c20f38bf4943a8a21dd1"
}
}
]
}
+25
View File
@@ -0,0 +1,25 @@
## Review
- **Correct:** Original safety and synchronous evidence finish before painting; physical/evidence failures suppress painting (`source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/live_cadence_v042.py:48–55`). Added-frame early close returns the literal abort cause before another step; final-tick completion is preserved (`:26–36`).
- **Correct:** Attachment is GUI-only, after `gui.start`; both combined modes receive runtime math qualification (`scripts/evaluate_cycle_functional_flow.py:122–125,270–272`). The new module is preloaded and source-bound; the amendment is included in approval sources (`:70`; `skill_binding_v4.py:18`; `scripts/skill_launch_v4.py:32–33`).
- **Correct:** Physics throughput and render-generation rate are reported separately, without claiming measured display-present FPS (`source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/performance_v042.py:90–94`).
### Finding: P1 — SDK-swallowed visualizer failures permit silent missing paints
**Location:** `source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/live_cadence_v042.py:28–36`.
The painter treats a returning `sim.render()` as successful, increments `additional_renders`, and checks only GUI/app closure. However, the installed SDK catches exceptions from `viz.step()`, closes/removes that visualizer, and returns normally:
- `/home/cen/IsaacLab/source/isaaclab/isaaclab/sim/simulation_context.py:829–839`
- `render_generation` still increments at `:776`; subsequent updates with no visualizers return immediately at `:793–794`.
Consequently, a Kit visualizer failure during an added paint can leave the app running, allow the next physical step, and count subsequent no-op renders as successful paints. This contradicts the new every-accepted-tick-paint contract. The direct render-throw test (`tests/test_contact_cycle_live_cadence_v042.py:78–83`) does not cover this SDK-swallowed failure.
**Smallest fix:** Retain and validate the expected Kit visualizer around added renders, failing closed if it is removed/closed. Check its health before subsequent physics as well, covering failures in retained interval20 renders. Add a negative case where rendering returns normally after removing the visualizer.
- **Fixed:** None; read-only review.
- **Merge verdict: BLOCK** — address this exception gap before spending the fresh full-S1 GUI qualification run.
### Qualification limits
The existing XML reports **68 passed, zero failures/errors**, including 12 presentation cases; nothing was rerun. CPU tests do not establish native acceptance. GUI/headless trace equivalence, measured presentation responsiveness/RTF, native early-close behavior, and renderer physical side effects remain empirical qualification work. In particular, SDK rendering invokes kinematic/fabric refresh (`simulation_context.py:800–801`; `physx_manager.py:329–335`), although the inspected path does not integrate another physics step.
@@ -0,0 +1,9 @@
# Display review P1 correction before native qualification
The reviewer found SimulationContext.update_visualizers swallows exceptions, removes the failing visualizer, and still increments render_generation. Native render-call return or generation count alone is NOT successful painting proof.
VerifiedTickPainter now retains exactly one expected concrete Kit instance. It checks identity membership, closed/running status, rendering-paused and training-paused flags at construction, before an added render, and AFTER its return. Only then increments additional_renders. A swallowed removal/close/paused state fails closed. No replacement visualizer is admitted. This does not claim display-present timestamps.
Evaluator invokes the same check before EVERY next physical call and immediately after every env.step/control, covering failures in original interval20 renders (including the final protocol tick). Final viewing uses the same guard per iteration, with an additional post-hold check. Headless/other modes without a painter have no new visualizer requirement. Existing user-close abort checks remain first in before_physics; intentional canceled close at full completion remains distinct from lost visualizer.
Added CPU cases for normal-return/swallowed removal with app still alive, removal in original interval20 at both midprotocol/final tick preventing the next physical call, closed/not-running/render-paused/training-paused/replaced concrete instances, and no-painter headless. Presentation/math/original runtime suite77passed (cpu-display-health.log/.xml). Native equivalence and actual responsiveness remain unqualified until the new real run. Existing failed collection log remains preserved; no GPU retry has occurred for these fixes.
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,37 @@
cycle,command,phase,Fn_min_N,Fn_mean_N,Fn_max_N,mean_load_W,gap_min_m,gap_max_m,dx_m,dy_m,dz_m,root_xy_max_m
1,x_bias-16mm,support,30.63823699951172,52.85984420776367,74.09165954589844,1.0000022649765015,-1.010112464427948e-05,9.723007678985596e-07,0.0017836317420005798,0.0006739199161529541,8.530914783477783e-07,0.029242604970932007
1,x_bias-16mm,unload,35.07340621948242,48.76642990112305,53.128562927246094,0.9225630760192871,-2.6773661375045776e-05,-7.014721632003784e-06,-0.002730610780417919,-0.0031594932079315186,-1.9758939743041992e-05,0.07960376888513565
1,x_bias-16mm,lift,0.0,10.591720581054688,35.157493591308594,0.20037412643432617,-2.9793009161949158e-05,0.044680267572402954,0.06119457259774208,-0.009405404329299927,0.044706881046295166,0.08049342036247253
1,x_bias-16mm,swing,0.0,0.0,0.0,0.0,0.04415256530046463,0.05414176732301712,0.01172223687171936,0.004630729556083679,-0.0005735680460929871,0.030786940827965736
1,x_bias-16mm,landing,0.0,0.8175418376922607,6.757045269012451,0.015466248616576195,-3.1776726245880127e-06,0.04414748400449753,-1.6469508409500122e-05,-0.001996748149394989,-0.0441494956612587,0.0022682934068143368
1,x_bias-16mm,reload,6.254490375518799,32.64689636230469,53.67607116699219,0.6176137328147888,-6.200745701789856e-06,-1.7527490854263306e-06,0.0016945339739322662,0.0006249770522117615,-1.3224780559539795e-07,0.016857551410794258
2,x_bias-16mm,support,52.44454574584961,52.84786605834961,52.95899963378906,0.9997755885124207,-2.386048436164856e-06,-1.8700957298278809e-06,-2.793222665786743e-05,-3.3736228942871094e-05,-3.203749656677246e-07,0.010354658588767052
2,x_bias-16mm,unload,8.834904670715332,33.76447677612305,52.96793746948242,0.6387562155723572,-7.227063179016113e-06,-2.0116567611694336e-06,-0.0016921795904636383,-0.0010888651013374329,-1.948326826095581e-06,0.01555535290390253
2,x_bias-16mm,lift,0.0,1.4353010654449463,8.878164291381836,0.02715301886200905,-4.244968295097351e-06,0.04408973455429077,0.00015369057655334473,0.0028083696961402893,0.044093795120716095,0.014773398637771606
2,x_bias-16mm,swing,0.0,0.0,0.0,0.0,0.044129468500614166,0.0542534664273262,-0.014362409710884094,-0.006014391779899597,0.00036966055631637573,0.005341693293303251
2,x_bias-16mm,landing,0.0,0.7480207085609436,6.374540328979492,0.014151052571833134,-4.002824425697327e-06,0.04449426382780075,-3.477931022644043e-05,-0.001990102231502533,-0.04449593275785446,0.0018029295606538653
2,x_bias-16mm,reload,5.9372711181640625,32.19451904296875,54.12532043457031,0.6090556979179382,-5.416572093963623e-06,-4.3958425521850586e-07,0.002125866711139679,0.000569663941860199,-1.564621925354004e-07,0.014224872924387455
3,baseline_direction-24mm,support,52.238529205322266,52.86407470703125,53.99081039428711,1.0000823736190796,-1.8570572137832642e-06,-7.972121238708496e-07,-3.374367952346802e-05,-5.480647087097168e-05,5.774199962615967e-07,0.01038394682109356
3,baseline_direction-24mm,unload,8.168989181518555,33.66178512573242,53.045509338378906,0.6368134617805481,-6.029382348060608e-06,-1.16787850856781e-06,-0.0021336041390895844,-0.0009494572877883911,-1.3634562492370605e-06,0.013789398595690727
3,baseline_direction-24mm,lift,0.0,1.2718819379806519,8.22209358215332,0.024061456322669983,-2.64681875705719e-06,0.04435984790325165,-6.523355841636658e-05,0.002624101936817169,0.044362492859363556,0.012191063724458218
3,baseline_direction-24mm,swing,0.0,0.0,0.0,0.0,0.043728090822696686,0.054023608565330505,0.01837150752544403,0.014431692659854889,-0.0006711855530738831,0.004302798304706812
3,baseline_direction-24mm,landing,0.0,0.9398022294044495,7.396997451782227,0.017779173329472542,-4.2729079723358154e-06,0.043723247945308685,-0.00018042325973510742,-0.00211150199174881,-0.04372534155845642,0.0028196307830512524
3,baseline_direction-24mm,reload,6.851399898529053,33.19783020019531,53.04985809326172,0.628036379814148,-6.3944607973098755e-06,-1.3932585716247559e-06,0.0013345815241336823,0.0007220804691314697,6.277114152908325e-07,0.01937531679868698
4,baseline_direction-24mm,support,52.748748779296875,52.85877227783203,52.925045013427734,0.9999818205833435,-1.7173588275909424e-06,-1.3485550880432129e-06,-1.5985220670700073e-05,-4.0940940380096436e-05,-5.960464477539063e-08,0.010775136761367321
4,baseline_direction-24mm,unload,9.798057556152344,34.15670394897461,52.934261322021484,0.6461763381958008,-7.964670658111572e-06,-1.434236764907837e-06,-0.0012961290776729584,-0.001278109848499298,-2.898275852203369e-06,0.01769513636827469
4,baseline_direction-24mm,lift,0.0,1.6734437942504883,9.832220077514648,0.031658198684453964,-4.449859261512756e-06,0.04365993291139603,0.00011078640818595886,0.002964310348033905,0.04366418346762657,0.01744244620203972
4,baseline_direction-24mm,swing,0.0,0.0,0.0,0.0,0.043699897825717926,0.054017066955566406,-0.018175136297941208,-0.014890573918819427,0.000783778727054596,0.006423341576009989
4,baseline_direction-24mm,landing,0.0,0.7480117678642273,6.37589693069458,0.014150884002447128,-3.807246685028076e-06,0.04447878897190094,-3.225356340408325e-05,-0.0019776523113250732,-0.0444803461432457,0.001810165704227984
4,baseline_direction-24mm,reload,5.937798023223877,32.19753646850586,54.13505935668945,0.6091127991676331,-5.3998082876205444e-06,-3.5390257835388184e-07,0.0021260157227516174,0.000570446252822876,-1.9744038581848145e-07,0.014237570576369762
5,y_bias-16mm,support,52.23505783081055,52.86397171020508,53.985328674316406,1.0000804662704468,-1.7434358596801758e-06,-7.487833499908447e-07,-3.3818185329437256e-05,-5.445629358291626e-05,2.1047890186309814e-07,0.010374261997640133
5,y_bias-16mm,unload,8.171730995178223,33.66255569458008,53.0526123046875,0.6368280649185181,-6.103888154029846e-06,-1.1138617992401123e-06,-0.0021327808499336243,-0.0009502843022346497,-1.5534460544586182e-06,0.013793233782052994
5,y_bias-16mm,lift,0.0,1.272587776184082,8.224916458129883,0.02407480776309967,-2.6654452085494995e-06,0.04435988515615463,-6.387010216712952e-05,0.002628639340400696,0.04436245560646057,0.012200022116303444
5,y_bias-16mm,swing,0.0,0.0,0.0,0.0,0.04386032372713089,0.05408100038766861,0.009489979594945908,0.012295417487621307,-0.000539138913154602,0.0043068528175354
5,y_bias-16mm,landing,0.0,0.9074799418449402,7.242805480957031,0.017167698591947556,-3.287568688392639e-06,0.04385525733232498,-0.00011193007230758667,-0.002096906304359436,-0.04385732114315033,0.0025537440087646246
5,y_bias-16mm,reload,6.725765228271484,33.07767105102539,53.55429458618164,0.6257631778717041,-6.126239895820618e-06,-1.2926757335662842e-06,0.0016064122319221497,0.0006933882832527161,5.345791578292847e-07,0.018027793616056442
6,y_bias-16mm,support,52.486968994140625,52.852115631103516,52.9252815246582,0.9998559951782227,-1.7229467630386353e-06,-1.2200325727462769e-06,-2.7876347303390503e-05,-3.699213266372681e-05,-1.6205012798309326e-07,0.010842877440154552
6,y_bias-16mm,unload,9.509719848632812,34.18946075439453,52.94326400756836,0.6467959880828857,-6.4838677644729614e-06,-1.3951212167739868e-06,-0.001573760062456131,-0.001220114529132843,-1.8905848264694214e-06,0.016642587259411812
6,y_bias-16mm,lift,0.0,1.597585916519165,9.55402946472168,0.030223118141293526,-3.6247074604034424e-06,0.04377070814371109,2.1435320377349854e-05,0.002900734543800354,0.04377419129014015,0.016067897900938988
6,y_bias-16mm,swing,0.0,0.0,0.0,0.0,0.04381057620048523,0.05407753586769104,-0.009249720722436905,-0.012734860181808472,0.0006802380084991455,0.0058596753515303135
6,y_bias-16mm,landing,0.0,0.7479984164237976,6.375153541564941,0.014150630682706833,-3.933906555175781e-06,0.044485852122306824,-3.6288052797317505e-05,-0.0019793733954429626,-0.044487517327070236,0.0018060122383758426
6,y_bias-16mm,reload,5.937589168548584,32.19605255126953,54.14176940917969,0.6090847849845886,-5.345791578292847e-06,-4.172325134277344e-07,0.0021268315613269806,0.0005712136626243591,-6.51925802230835e-08,0.01423043105751276
1 cycle command phase Fn_min_N Fn_mean_N Fn_max_N mean_load_W gap_min_m gap_max_m dx_m dy_m dz_m root_xy_max_m
2 1 x_bias-16mm support 30.63823699951172 52.85984420776367 74.09165954589844 1.0000022649765015 -1.010112464427948e-05 9.723007678985596e-07 0.0017836317420005798 0.0006739199161529541 8.530914783477783e-07 0.029242604970932007
3 1 x_bias-16mm unload 35.07340621948242 48.76642990112305 53.128562927246094 0.9225630760192871 -2.6773661375045776e-05 -7.014721632003784e-06 -0.002730610780417919 -0.0031594932079315186 -1.9758939743041992e-05 0.07960376888513565
4 1 x_bias-16mm lift 0.0 10.591720581054688 35.157493591308594 0.20037412643432617 -2.9793009161949158e-05 0.044680267572402954 0.06119457259774208 -0.009405404329299927 0.044706881046295166 0.08049342036247253
5 1 x_bias-16mm swing 0.0 0.0 0.0 0.0 0.04415256530046463 0.05414176732301712 0.01172223687171936 0.004630729556083679 -0.0005735680460929871 0.030786940827965736
6 1 x_bias-16mm landing 0.0 0.8175418376922607 6.757045269012451 0.015466248616576195 -3.1776726245880127e-06 0.04414748400449753 -1.6469508409500122e-05 -0.001996748149394989 -0.0441494956612587 0.0022682934068143368
7 1 x_bias-16mm reload 6.254490375518799 32.64689636230469 53.67607116699219 0.6176137328147888 -6.200745701789856e-06 -1.7527490854263306e-06 0.0016945339739322662 0.0006249770522117615 -1.3224780559539795e-07 0.016857551410794258
8 2 x_bias-16mm support 52.44454574584961 52.84786605834961 52.95899963378906 0.9997755885124207 -2.386048436164856e-06 -1.8700957298278809e-06 -2.793222665786743e-05 -3.3736228942871094e-05 -3.203749656677246e-07 0.010354658588767052
9 2 x_bias-16mm unload 8.834904670715332 33.76447677612305 52.96793746948242 0.6387562155723572 -7.227063179016113e-06 -2.0116567611694336e-06 -0.0016921795904636383 -0.0010888651013374329 -1.948326826095581e-06 0.01555535290390253
10 2 x_bias-16mm lift 0.0 1.4353010654449463 8.878164291381836 0.02715301886200905 -4.244968295097351e-06 0.04408973455429077 0.00015369057655334473 0.0028083696961402893 0.044093795120716095 0.014773398637771606
11 2 x_bias-16mm swing 0.0 0.0 0.0 0.0 0.044129468500614166 0.0542534664273262 -0.014362409710884094 -0.006014391779899597 0.00036966055631637573 0.005341693293303251
12 2 x_bias-16mm landing 0.0 0.7480207085609436 6.374540328979492 0.014151052571833134 -4.002824425697327e-06 0.04449426382780075 -3.477931022644043e-05 -0.001990102231502533 -0.04449593275785446 0.0018029295606538653
13 2 x_bias-16mm reload 5.9372711181640625 32.19451904296875 54.12532043457031 0.6090556979179382 -5.416572093963623e-06 -4.3958425521850586e-07 0.002125866711139679 0.000569663941860199 -1.564621925354004e-07 0.014224872924387455
14 3 baseline_direction-24mm support 52.238529205322266 52.86407470703125 53.99081039428711 1.0000823736190796 -1.8570572137832642e-06 -7.972121238708496e-07 -3.374367952346802e-05 -5.480647087097168e-05 5.774199962615967e-07 0.01038394682109356
15 3 baseline_direction-24mm unload 8.168989181518555 33.66178512573242 53.045509338378906 0.6368134617805481 -6.029382348060608e-06 -1.16787850856781e-06 -0.0021336041390895844 -0.0009494572877883911 -1.3634562492370605e-06 0.013789398595690727
16 3 baseline_direction-24mm lift 0.0 1.2718819379806519 8.22209358215332 0.024061456322669983 -2.64681875705719e-06 0.04435984790325165 -6.523355841636658e-05 0.002624101936817169 0.044362492859363556 0.012191063724458218
17 3 baseline_direction-24mm swing 0.0 0.0 0.0 0.0 0.043728090822696686 0.054023608565330505 0.01837150752544403 0.014431692659854889 -0.0006711855530738831 0.004302798304706812
18 3 baseline_direction-24mm landing 0.0 0.9398022294044495 7.396997451782227 0.017779173329472542 -4.2729079723358154e-06 0.043723247945308685 -0.00018042325973510742 -0.00211150199174881 -0.04372534155845642 0.0028196307830512524
19 3 baseline_direction-24mm reload 6.851399898529053 33.19783020019531 53.04985809326172 0.628036379814148 -6.3944607973098755e-06 -1.3932585716247559e-06 0.0013345815241336823 0.0007220804691314697 6.277114152908325e-07 0.01937531679868698
20 4 baseline_direction-24mm support 52.748748779296875 52.85877227783203 52.925045013427734 0.9999818205833435 -1.7173588275909424e-06 -1.3485550880432129e-06 -1.5985220670700073e-05 -4.0940940380096436e-05 -5.960464477539063e-08 0.010775136761367321
21 4 baseline_direction-24mm unload 9.798057556152344 34.15670394897461 52.934261322021484 0.6461763381958008 -7.964670658111572e-06 -1.434236764907837e-06 -0.0012961290776729584 -0.001278109848499298 -2.898275852203369e-06 0.01769513636827469
22 4 baseline_direction-24mm lift 0.0 1.6734437942504883 9.832220077514648 0.031658198684453964 -4.449859261512756e-06 0.04365993291139603 0.00011078640818595886 0.002964310348033905 0.04366418346762657 0.01744244620203972
23 4 baseline_direction-24mm swing 0.0 0.0 0.0 0.0 0.043699897825717926 0.054017066955566406 -0.018175136297941208 -0.014890573918819427 0.000783778727054596 0.006423341576009989
24 4 baseline_direction-24mm landing 0.0 0.7480117678642273 6.37589693069458 0.014150884002447128 -3.807246685028076e-06 0.04447878897190094 -3.225356340408325e-05 -0.0019776523113250732 -0.0444803461432457 0.001810165704227984
25 4 baseline_direction-24mm reload 5.937798023223877 32.19753646850586 54.13505935668945 0.6091127991676331 -5.3998082876205444e-06 -3.5390257835388184e-07 0.0021260157227516174 0.000570446252822876 -1.9744038581848145e-07 0.014237570576369762
26 5 y_bias-16mm support 52.23505783081055 52.86397171020508 53.985328674316406 1.0000804662704468 -1.7434358596801758e-06 -7.487833499908447e-07 -3.3818185329437256e-05 -5.445629358291626e-05 2.1047890186309814e-07 0.010374261997640133
27 5 y_bias-16mm unload 8.171730995178223 33.66255569458008 53.0526123046875 0.6368280649185181 -6.103888154029846e-06 -1.1138617992401123e-06 -0.0021327808499336243 -0.0009502843022346497 -1.5534460544586182e-06 0.013793233782052994
28 5 y_bias-16mm lift 0.0 1.272587776184082 8.224916458129883 0.02407480776309967 -2.6654452085494995e-06 0.04435988515615463 -6.387010216712952e-05 0.002628639340400696 0.04436245560646057 0.012200022116303444
29 5 y_bias-16mm swing 0.0 0.0 0.0 0.0 0.04386032372713089 0.05408100038766861 0.009489979594945908 0.012295417487621307 -0.000539138913154602 0.0043068528175354
30 5 y_bias-16mm landing 0.0 0.9074799418449402 7.242805480957031 0.017167698591947556 -3.287568688392639e-06 0.04385525733232498 -0.00011193007230758667 -0.002096906304359436 -0.04385732114315033 0.0025537440087646246
31 5 y_bias-16mm reload 6.725765228271484 33.07767105102539 53.55429458618164 0.6257631778717041 -6.126239895820618e-06 -1.2926757335662842e-06 0.0016064122319221497 0.0006933882832527161 5.345791578292847e-07 0.018027793616056442
32 6 y_bias-16mm support 52.486968994140625 52.852115631103516 52.9252815246582 0.9998559951782227 -1.7229467630386353e-06 -1.2200325727462769e-06 -2.7876347303390503e-05 -3.699213266372681e-05 -1.6205012798309326e-07 0.010842877440154552
33 6 y_bias-16mm unload 9.509719848632812 34.18946075439453 52.94326400756836 0.6467959880828857 -6.4838677644729614e-06 -1.3951212167739868e-06 -0.001573760062456131 -0.001220114529132843 -1.8905848264694214e-06 0.016642587259411812
34 6 y_bias-16mm lift 0.0 1.597585916519165 9.55402946472168 0.030223118141293526 -3.6247074604034424e-06 0.04377070814371109 2.1435320377349854e-05 0.002900734543800354 0.04377419129014015 0.016067897900938988
35 6 y_bias-16mm swing 0.0 0.0 0.0 0.0 0.04381057620048523 0.05407753586769104 -0.009249720722436905 -0.012734860181808472 0.0006802380084991455 0.0058596753515303135
36 6 y_bias-16mm landing 0.0 0.7479984164237976 6.375153541564941 0.014150630682706833 -3.933906555175781e-06 0.044485852122306824 -3.6288052797317505e-05 -0.0019793733954429626 -0.044487517327070236 0.0018060122383758426
37 6 y_bias-16mm reload 5.937589168548584 32.19605255126953 54.14176940917969 0.6090847849845886 -5.345791578292847e-06 -4.172325134277344e-07 0.0021268315613269806 0.0005712136626243591 -6.51925802230835e-08 0.01423043105751276
+39
View File
@@ -0,0 +1,39 @@
{
"campaign_original_success": true,
"owned_identities_checked": 11,
"owned_alive": [],
"finite_bound_files_checked": 470,
"binding_file_mismatches": [],
"protected_file_mismatches": [],
"gpu_clients": {
"compute": [],
"graphics": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
]
},
"gpu_review": {
"uuid": "GPU-7d882fc3-f771-6d29-4e58-f0fa2a18a3a5",
"compute_pids": [],
"approved_desktop_graphics_pids": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
],
"task_graphics_pids": [],
"clean": true,
"scope": "owned-compute-and-graphics.plus-pinned-desktop-graphics-only"
},
"static_check_not_same_cfg_proof": true,
"boottime": 361576.443472762,
"passed": true
}
@@ -0,0 +1,90 @@
{
"originals": {
"worker": "performance_v042/runs/display-S1-05/campaign/run/worker-result.json",
"campaign": "performance_v042/runs/display-S1-05/campaign/campaign-result.json",
"observer": "performance_v042/runs/display-S1-05/observer/result.json"
},
"worker": {
"cycles_completed": 6,
"gui": {
"camera": "/World/SkillDemoCamera",
"close_reason": "window_close",
"enabled": true,
"hold_no_physics": true,
"last_tick": 27000,
"max_final_view_seconds": 600,
"protocol_completed": true,
"render_updates": 27059,
"source": "live PhysX and actual state; no playback; original 3kg payload mesh"
},
"hook_violations": [],
"performance_installation": {
"constructor_sensor_references_routed": [
"mount_wrench"
],
"implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.sensor_epoch_v042.EpochSensor",
"jacobian": {
"all_bitwise_equal": true,
"applied": true,
"epsilon_rad": 0.0005,
"ik_parameters_changed": false,
"max_abs": 0.0,
"qualification": "sampled same-device same-dtype original-vs-batched math; not all-input proof",
"reason": "exact_samples",
"samples": 33
},
"safety_read_scope": true,
"sensors": [
"foot_contact",
"mount_wrench"
]
},
"physics_counts": {
"attempted_steps": 27000,
"completed_steps": 27000,
"composer_buffer_replacements": 27000,
"initialization_steps": 0,
"pose_write_calls": 14,
"reset_calls": 22,
"rollout_steps": 27000
},
"preclose_identity_verified": true,
"runtime_continuity_verified": true,
"runtime_stop": {
"all_sim_steps_attempted": 27000,
"all_sim_steps_completed": 27000,
"causes": [],
"clock": {
"cycle_index": 5,
"global_tick": 27000,
"local_tick": 4500
},
"completed_clock_cycles": 6,
"completed_controls": 5400,
"composer_buffer_replacements": 27000,
"fatal": false,
"initialization_steps": 0,
"partial_control_ticks": 0,
"physics_ticks": 27000,
"pose_write_calls": 14,
"reason": "protocol_complete",
"reset_calls": 22,
"rollout_steps": 27000,
"trial_index": 0
},
"shutdown_handoff_ready": true,
"success": true
},
"campaign": {
"reserved_physics_calls": 27250,
"success": true
},
"rows": [
{
"success": true
},
{
"success": true
}
]
}
@@ -0,0 +1,15 @@
{
"paths": [
"/home/cen/sf-runs/skill-gui-v041-gui-complete11/S1/run/trace.pt",
"performance_v042/runs/display-S1-05/campaign/run/trace.pt"
],
"sha256": [
"d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba",
"d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba"
],
"keys": 70,
"keysets_equal": true,
"all_bitwise_equal": true,
"differences": {},
"scope": "Actual native six-cycle S1 same-seed trace; different source versions and render cadence. No all-input determinism or complete opaque PhysX-state claim."
}
@@ -0,0 +1,16 @@
## Review
- **Correct — new P1 fix:** `live_cadence_v042.py:18–46` retains the concrete Kit identity, checks membership/liveness/pause state before and after added rendering, and increments the paint count only after validation. This closes the SDK-swallowed visualizer-removal gap identified in `performance_v042/display-code-review.md`.
- **Correct — integration:** `scripts/evaluate_cycle_functional_flow.py:175–179,226–231,291–295,328–331` applies that guard before physics, after controls—including final interval20 rendering—and during/post final viewing. `live_cadence_v042.py:63–76` preserves evidence-before-paint ordering and headless no-painter behavior.
- **Correct — regression evidence:** `tests/test_contact_cycle_live_cadence_v042.py:95–137` covers swallowed removal, original interval20 failure, unhealthy/replaced identities and headless operation. Retained logs and XML agree: targeted **77 passed**; full suite **1557 passed, 6 skipped**, zero failures/errors. Nothing was rerun.
- **Correct — bounded native evidence:** Both review extracts report original campaign/worker success, six cycles and 27000 ticks. Original observer receipts show actual caller wait0; GUI additionally records real WM_DELETE, unchanged hold counters and window absence. Both physical audits report zero new steps, passing functional checks and exact world targets (`actual-audit.json:1010–1026`).
- **Correct — performance and equality claims:** `performance-result.json` supports GUI **764.863s / 0.07060×**, headless **294.679s / 0.18325×**, and **25650 added + 1350 original renders**. Both comparison JSONs report 70 equal tensors and identical whole-trace SHA256 `d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba`. The inspected PNG visibly shows **33.98 FPS, tick2519**. Documentation correctly separates render submissions, capture sampling and simulation throughput (`docs/contact_cycle_sim_performance_v042_result.md:15–21`).
- **Correct — disclosure:** The independent headless postexit first failure remains preserved with unknown cause; the second clean observation is not presented as an explanation or physical rerun. Failed baseline01, initial CPU collection failure, historical 30mm RootXY failure, qualification limits and mandatory `accepted_skill=false` remain explicit (`docs/contact_cycle_sim_performance_v042_result.md:43–55`).
- **Fixed:** None by this read-only reviewer.
No issues found.
- **Merge verdict: OK with notes** — no blocker found for proceeding to packaging of the bounded analytic-S1 result. This is **not delivery-complete approval**.
### Residual risks
New-cadence native early-abort, injected visualizer failure, startup-close, safe-stop and 600s timeout remain untested. S2/S3/nominal20/u256, all-input bitwise equivalence and global native-state equality are not qualified. Package checksums, final inventory/protected/process checks and sealing remain pending.
+43
View File
@@ -0,0 +1,43 @@
#!/usr/bin/python3.12
"""Read-only runtime evidence aggregation, never reclassify failed attempts."""
import pathlib,json,hashlib,datetime,time,xml.etree.ElementTree as ET
A=pathlib.Path(__file__).resolve().parent
read=lambda p:json.loads(p.read_text())
sha=lambda p:hashlib.sha256(p.read_bytes()).hexdigest()
def differences(a,b,path=''):
if type(a)!=type(b):return [path]
if isinstance(a,dict):
return [p for k in sorted(a.keys()|b.keys()) for p in ([path+'/'+k] if k not in a or k not in b else differences(a[k],b[k],path+'/'+k))]
if isinstance(a,list):
return [path] if len(a)!=len(b) else [p for i,(x,y) in enumerate(zip(a,b)) for p in differences(x,y,path+'/'+str(i))]
return [] if a==b else [path]
summary=read(A/'performance-summary-pre-headless.json')
summary['headless-S1-06']={'loop':{k:v for k,v in read(A/'runs/headless-S1-06/campaign/run/performance-profile.json').items() if k!='functions_by_cumulative'}}
rows=[];identities=[]
for root in sorted((A/'runs').iterdir()):
c=read(root/'campaign/campaign-result.json');ap=read(root/'campaign/approval.json');identities += [x for row in c['rows'] for x in row['owned_identities']]
rows.append({'name':root.name,'original_success':c['success'],'Kit_instances':len(c['rows']),'actual_physics_calls':c['actual_final_physics_calls'],'reserved_physics_calls':c['reserved_physics_calls'],'original_campaign_sha256':sha(root/'campaign/campaign-result.json'),'waited_cleanup':all(x['waited_exit'] and x['cleanup_clean'] for x in c['rows']), 'source_inputs':len(ap['sources'])})
alive=[]
for pid,birth in identities:
try:
s=pathlib.Path(f'/proc/{pid}/stat').read_text()
if int(s[s.rfind(')')+2:].split()[19])==birth:alive.append([pid,birth])
except FileNotFoundError:pass
scope=read(A/'scope.json');counts={'Kit_instances':sum(x['Kit_instances'] for x in rows),'actual_physics_calls':sum(x['actual_physics_calls'] for x in rows),'reserved_physics_calls':sum(x['reserved_physics_calls'] for x in rows),'optimizer_steps':0,'owned_identities_checked':len(identities),'owned_alive':alive,'elapsed_seconds':time.time()-datetime.datetime.fromisoformat(scope['start']).timestamp()}
assert counts['Kit_instances']<=scope['limits']['kit_launches'] and counts['reserved_physics_calls']<=scope['limits']['physics_calls'] and not alive
G=read(A/'runs/display-S1-05/campaign/prepare/binding.json');H=read(A/'runs/headless-S1-06/campaign/prepare/binding.json');diff=differences(G['effective_config'],H['effective_config']);assert len(diff)==1 and diff[0].startswith('/fields/mapping/sim/fields/mapping/visualizer_cfgs/'),diff
src=G['sources'];assert src==H['sources']
authority=read(A/'runs/display-S1-05/campaign/approval.json')['sources'];assert authority==read(A/'runs/headless-S1-06/campaign/approval.json')['sources']
P=pathlib.Path(read(A/'staging.json')['root']);assert all(sha(P/n)==h for n,h in authority.items())
summary.update(cases=rows,accounting=counts,effective_config_differences=diff,current_source_inputs_equal=True,binding_base_source_input_count=len(src),authority_source_input_count=len(authority),accepted_skill=False,training_steps=0)
summary['failure_disclosure']='baseline-profile01 failed PREP before physics and remains failed. Both reviewP1findings fixed before native qualification, CPU collection failure preserved. First independent headless06 postexit observer raised GPU admission mismatch without PID evidence; exact cause unknown. Original runtime success unchanged. After callerwait0 and clean nvidia-smi observation, unchanged strict verifier second observation passed; no whitelist/source changes, no new Kit/physics. Shared native cache monitoring not a hard quota.'
summary['qualification_scope']='Current-source analytic S1 only: GUI6cycles27000ticks and headless6cycles27000ticks,70tensors byte-identical to each other and prior0.4.1GUI11 including entire trace-fileSHA. No current-source nominal20/S2/S3/u256 qualification, native new-cadence early-abort/visualizer-failure injection/startup-close/GUI-safe-stop/600s viewing-timeout test, global determinism or opaque native-state completeness claim. Earlyclose/visualizerfailure protections CPUnegative-tested.'
summary['postexit_reports']=['display05-postexit.json','headless06-postexit-first-failure.json','headless06-postexit-second.json']
summary['cpu_acceptance']={'passed':1557,'skipped':6,'failed':0,'log':'cpu-full03.log','xml':'cpu-full03.xml'}
(A/'performance-result.json').write_text(json.dumps(summary,indent=2)+'\n')
# Compact receipts with original paths/hashes; originals remain untouched.
for name in ['display-S1-05','headless-S1-06']:
r=A/'runs'/name/'campaign';w=read(r/'run/worker-result.json');c=read(r/'campaign-result.json');e={'originals':{str(p):sha(p) for p in [r/'run/worker-result.json',r/'campaign-result.json',r.parent/'observer/result.json']},'worker':{k:v for k,v in w.items() if k in ['success','error','runtime_stop','cycles_completed','physics_counts','hook_violations','runtime_continuity_verified','preclose_identity_verified','performance_installation','gui','shutdown_handoff_ready']},'campaign':{k:c[k] for k in ['success','actual_final_physics_calls','reserved_physics_calls','accepted_skill','optimizer_steps']},'rows':[{k:x[k] for k in ['slot','success','returncode','waited_exit','cleanup_clean','shutdown_handoff']} for x in c['rows']]};(A/(name+'-review-extract.json')).write_text(json.dumps(e,indent=2)+'\n')
print(json.dumps({'accounting':counts,'config_differences':diff,'binding_base_source_inputs':len(src),'authority_source_inputs':len(authority),'cpu':summary['cpu_acceptance']}))
+49
View File
@@ -0,0 +1,49 @@
#!/usr/bin/python3.12
"""Final independent static/process/GPU/protected/accounting checks. No physics."""
import pathlib,json,hashlib,sys,time,datetime,ctypes
A=pathlib.Path(__file__).resolve().parent;D=A.parent
read=lambda p:json.loads(p.read_text());sha=lambda p:hashlib.sha256(p.read_bytes()).hexdigest()
seal=read(A/'package-verification.json');P=pathlib.Path(seal['package']);S=pathlib.Path(read(A/'staging.json')['root']);scope=read(A/'scope.json');assert not (A/'task-complete.json').exists()
assert sha(pathlib.Path(seal['archive']))==seal['archive_sha256']
for line in (P/'SHA256SUMS').read_text().splitlines():
h,n=line.split(' ',1);assert sha(P/n)==h,n
auth=read(A/'runs/display-S1-05/campaign/approval.json')['sources'];assert all(sha(P/n)==sha(S/n)==sha(D/n)==h for n,h in auth.items())
protected=read(A/'protected-registration.json')['protected'];bad=[n for n,h in protected.items() if not pathlib.Path(n).is_file() or sha(pathlib.Path(n))!=h];assert not bad,bad
identities=[];intervals=[];cases=[]
for r in (A/'runs').iterdir():
c=read(r/'campaign/campaign-result.json');o=read(r/'observer/result.json');identities += [i for row in c['rows'] for i in row['owned_identities']];intervals.append({'case':r.name,'begin_wall':o['began_wall'],'end_wall':o['ended_wall'],'Kit':len(c['rows'])});cases.append(c)
intervals.sort(key=lambda x:x['begin_wall']);assert all(x['end_wall']<=y['begin_wall'] for x,y in zip(intervals,intervals[1:])),intervals
alive=[]
for pid,birth in identities:
try:
st=pathlib.Path(f'/proc/{pid}/stat').read_text()
if int(st[st.rfind(')')+2:].split()[19])==birth:alive.append([pid,birth])
except FileNotFoundError:pass
assert not alive,alive
sys.path.insert(0,str(S/'scripts'));import functional_flow_launch_v2 as f
identity=read(A/'runs/headless-S1-06/campaign/approval.json')['gpu_identity'];gpu=f.ownership.GPUOwnership(identity);review=gpu.check(type('EmptyOwned',(),{'discover':lambda self:{}})(),final=True);clients={k:sorted(v) for k,v in gpu.clients().items()};assert not clients['compute']
# This scan is a final owned-storage inventory, not a claim to monitor all
# shared caches continuously. pytest root is conservatively counted even if
# it includes unrelated tests. Symlinks are listed but never followed.
entries=[];seen=set();bytes_=0;gaps=[]
roots=[A,pathlib.Path('/tmp/pytest-of-cen')]
for root in roots:
if not root.exists():continue
for p in root.rglob('*'):
if p.is_symlink():continue
try:
if p.is_file():
n=str(p.resolve())
if n not in seen:
seen.add(n);s=p.stat().st_size;bytes_+=s;entries.append({'path':n,'bytes':s,'sha256':sha(p) if root==A else None,'scope':'task-root' if root==A else 'conservative-shared-pytest-temp'})
except (FileNotFoundError,PermissionError) as e:gaps.append({'path':str(p),'error':type(e).__name__})
for p in [D/'setup-preflight.json',*(D/n for n in auth),D/'docs/contact_cycle_sim_performance_v042_result.md']:
n=str(p.resolve())
if n not in seen:seen.add(n);s=p.stat().st_size;bytes_+=s;entries.append({'path':n,'bytes':s,'sha256':sha(p),'scope':'additional-source/probe/result'})
reserve=16*1024**2;accounted=bytes_+reserve;elapsed=time.time()-datetime.datetime.fromisoformat(scope['start']).timestamp();kits=sum(len(c['rows']) for c in cases);actual=sum(c['actual_final_physics_calls'] for c in cases);reserved=sum(c['reserved_physics_calls'] for c in cases)
assert kits==11 and actual==67500 and reserved==73500
assert kits<=scope['limits']['kit_launches'] and reserved<=scope['limits']['physics_calls'] and accounted<=scope['limits']['owned_bytes'] and elapsed<scope['limits']['seconds'] and not gaps
inventory={'generated':datetime.datetime.now().astimezone().isoformat(),'entries':entries,'files':len(entries),'observed_logical_bytes':bytes_,'receipt_reserve_bytes':reserve,'accounted_bytes_including_reserve':accounted,'scope_roots':list(map(str,roots)),'gaps':gaps,'shared_native_cache_hard_quota_claim':False,'intervals':intervals,'serial_campaign_intervals':True,'owned_identities_checked':len(identities),'owned_alive':alive,'protected_files':len(protected),'protected_mismatches':bad,'gpu_clients':clients,'gpu_strict_final_review':review,'archive_sha256':seal['archive_sha256']}
(A/'final-inventory.json').write_text(json.dumps(inventory,indent=2)+'\n');assert (A/'final-inventory.json').stat().st_size<reserve-65536
complete={'completed':datetime.datetime.now().astimezone().isoformat(),'task':'bounded application performance + analyticS1 result sealed','accepted_skill':False,'training_steps':0,'Kit_instances':kits,'reserved_physics_calls':reserved,'actual_physics_calls':actual,'elapsed_seconds':elapsed,'accounted_owned_bytes_including_reserve':accounted,'scope_limits':scope['limits'],'protected_files_unchanged':len(protected),'owned_identities_checked':len(identities),'owned_alive':alive,'strict_final_GPU_clients':clients,'package':str(P),'archive':seal['archive'],'archive_sha256':seal['archive_sha256'],'archive_bytes':seal['archive_bytes'],'package_verification_sha256':sha(A/'package-verification.json'),'final_inventory_sha256':sha(A/'final-inventory.json'),'GUI_render_submissions_per_second':35.30043787880234,'GUI_window_overlay_observed_FPS':33.98,'GUI_sim_time_ratio':0.07060087575760468,'headless_sim_time_ratio':0.18325004295397335,'qualification':'current-source analyticS1GUI/headless6cycles each;70tensorbyte-identical to0.4.1GUI11; not1x/fullGPU/fullmatrix/u256/nativefaultinjection/globaldeterminism','unresolved_scope_limits':'README / RESULTS; firstindependentheadlesspostexitGPUadmissionexceptioncauseunknownpreserved; laterstrictchecknotretroactiveexplanation; firstpackagingCPUfixturefailurepreservedandpayloadrepaired','no_more_automatic_native_work':True}
(A/'task-complete.json').write_text(json.dumps(complete,indent=2)+'\n');print(json.dumps(complete,indent=2))
@@ -0,0 +1,136 @@
{
"originals": {
"/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/runs/headless-S1-06/campaign/run/worker-result.json": "def0264012d6f172746efda456434c124c5d88d5847776221a5937893bdd6cb5",
"/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/runs/headless-S1-06/campaign/campaign-result.json": "dd247da7ae4e830ad4420e0ea7006db6390642cc8269079496f22f3356f6141f",
"/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/runs/headless-S1-06/observer/result.json": "60f4a406366c39b029a98c110168ad5110f76d04b3d9a9f7a225d2032c9aba81"
},
"worker": {
"cycles_completed": 6,
"hook_violations": [],
"performance_installation": {
"constructor_sensor_references_routed": [
"mount_wrench"
],
"implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.sensor_epoch_v042.EpochSensor",
"jacobian": {
"all_bitwise_equal": true,
"applied": true,
"epsilon_rad": 0.0005,
"ik_parameters_changed": false,
"max_abs": 0.0,
"qualification": "sampled same-device same-dtype original-vs-batched math; not all-input proof",
"reason": "exact_samples",
"samples": 33
},
"safety_read_scope": true,
"sensors": [
"foot_contact",
"mount_wrench"
]
},
"physics_counts": {
"attempted_steps": 27000,
"completed_steps": 27000,
"composer_buffer_replacements": 27000,
"initialization_steps": 0,
"pose_write_calls": 14,
"reset_calls": 22,
"rollout_steps": 27000
},
"preclose_identity_verified": true,
"runtime_continuity_verified": true,
"runtime_stop": {
"all_sim_steps_attempted": 27000,
"all_sim_steps_completed": 27000,
"causes": [],
"clock": {
"cycle_index": 5,
"global_tick": 27000,
"local_tick": 4500
},
"completed_clock_cycles": 6,
"completed_controls": 5400,
"composer_buffer_replacements": 27000,
"fatal": false,
"initialization_steps": 0,
"partial_control_ticks": 0,
"physics_ticks": 27000,
"pose_write_calls": 14,
"reason": "protocol_complete",
"reset_calls": 22,
"rollout_steps": 27000,
"trial_index": 0
},
"shutdown_handoff_ready": true,
"success": true
},
"campaign": {
"success": true,
"actual_final_physics_calls": 27000,
"reserved_physics_calls": 27250,
"accepted_skill": false,
"optimizer_steps": 0
},
"rows": [
{
"slot": "prepare",
"success": true,
"returncode": 0,
"waited_exit": true,
"cleanup_clean": true,
"shutdown_handoff": {
"last_live_gpu": {
"approved_desktop_graphics_pids": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
],
"clean": false,
"compute_pids": [
1005680
],
"scope": "owned-compute-and-graphics.plus-pinned-desktop-graphics-only",
"task_graphics_pids": [],
"uuid": "GPU-7d882fc3-f771-6d29-4e58-f0fa2a18a3a5"
},
"released": true,
"scope": "No physics after receipt; bounded owned cleanup then unchanged strict final GPU check",
"worker_result_sha256": "0027e4228f3d004b61752912bb9dcbe52fa7a3db553de6301434bc4ce5a667c6"
}
},
{
"slot": "run",
"success": true,
"returncode": 0,
"waited_exit": true,
"cleanup_clean": true,
"shutdown_handoff": {
"last_live_gpu": {
"approved_desktop_graphics_pids": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
],
"clean": false,
"compute_pids": [
1006161
],
"scope": "owned-compute-and-graphics.plus-pinned-desktop-graphics-only",
"task_graphics_pids": [],
"uuid": "GPU-7d882fc3-f771-6d29-4e58-f0fa2a18a3a5"
},
"released": true,
"scope": "No physics after receipt; bounded owned cleanup then unchanged strict final GPU check",
"worker_result_sha256": "def0264012d6f172746efda456434c124c5d88d5847776221a5937893bdd6cb5"
}
}
]
}
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,37 @@
cycle,command,phase,Fn_min_N,Fn_mean_N,Fn_max_N,mean_load_W,gap_min_m,gap_max_m,dx_m,dy_m,dz_m,root_xy_max_m
1,x_bias-16mm,support,30.63823699951172,52.85984420776367,74.09165954589844,1.0000022649765015,-1.010112464427948e-05,9.723007678985596e-07,0.0017836317420005798,0.0006739199161529541,8.530914783477783e-07,0.029242604970932007
1,x_bias-16mm,unload,35.07340621948242,48.76642990112305,53.128562927246094,0.9225630760192871,-2.6773661375045776e-05,-7.014721632003784e-06,-0.002730610780417919,-0.0031594932079315186,-1.9758939743041992e-05,0.07960376888513565
1,x_bias-16mm,lift,0.0,10.591720581054688,35.157493591308594,0.20037412643432617,-2.9793009161949158e-05,0.044680267572402954,0.06119457259774208,-0.009405404329299927,0.044706881046295166,0.08049342036247253
1,x_bias-16mm,swing,0.0,0.0,0.0,0.0,0.04415256530046463,0.05414176732301712,0.01172223687171936,0.004630729556083679,-0.0005735680460929871,0.030786940827965736
1,x_bias-16mm,landing,0.0,0.8175418376922607,6.757045269012451,0.015466248616576195,-3.1776726245880127e-06,0.04414748400449753,-1.6469508409500122e-05,-0.001996748149394989,-0.0441494956612587,0.0022682934068143368
1,x_bias-16mm,reload,6.254490375518799,32.64689636230469,53.67607116699219,0.6176137328147888,-6.200745701789856e-06,-1.7527490854263306e-06,0.0016945339739322662,0.0006249770522117615,-1.3224780559539795e-07,0.016857551410794258
2,x_bias-16mm,support,52.44454574584961,52.84786605834961,52.95899963378906,0.9997755885124207,-2.386048436164856e-06,-1.8700957298278809e-06,-2.793222665786743e-05,-3.3736228942871094e-05,-3.203749656677246e-07,0.010354658588767052
2,x_bias-16mm,unload,8.834904670715332,33.76447677612305,52.96793746948242,0.6387562155723572,-7.227063179016113e-06,-2.0116567611694336e-06,-0.0016921795904636383,-0.0010888651013374329,-1.948326826095581e-06,0.01555535290390253
2,x_bias-16mm,lift,0.0,1.4353010654449463,8.878164291381836,0.02715301886200905,-4.244968295097351e-06,0.04408973455429077,0.00015369057655334473,0.0028083696961402893,0.044093795120716095,0.014773398637771606
2,x_bias-16mm,swing,0.0,0.0,0.0,0.0,0.044129468500614166,0.0542534664273262,-0.014362409710884094,-0.006014391779899597,0.00036966055631637573,0.005341693293303251
2,x_bias-16mm,landing,0.0,0.7480207085609436,6.374540328979492,0.014151052571833134,-4.002824425697327e-06,0.04449426382780075,-3.477931022644043e-05,-0.001990102231502533,-0.04449593275785446,0.0018029295606538653
2,x_bias-16mm,reload,5.9372711181640625,32.19451904296875,54.12532043457031,0.6090556979179382,-5.416572093963623e-06,-4.3958425521850586e-07,0.002125866711139679,0.000569663941860199,-1.564621925354004e-07,0.014224872924387455
3,baseline_direction-24mm,support,52.238529205322266,52.86407470703125,53.99081039428711,1.0000823736190796,-1.8570572137832642e-06,-7.972121238708496e-07,-3.374367952346802e-05,-5.480647087097168e-05,5.774199962615967e-07,0.01038394682109356
3,baseline_direction-24mm,unload,8.168989181518555,33.66178512573242,53.045509338378906,0.6368134617805481,-6.029382348060608e-06,-1.16787850856781e-06,-0.0021336041390895844,-0.0009494572877883911,-1.3634562492370605e-06,0.013789398595690727
3,baseline_direction-24mm,lift,0.0,1.2718819379806519,8.22209358215332,0.024061456322669983,-2.64681875705719e-06,0.04435984790325165,-6.523355841636658e-05,0.002624101936817169,0.044362492859363556,0.012191063724458218
3,baseline_direction-24mm,swing,0.0,0.0,0.0,0.0,0.043728090822696686,0.054023608565330505,0.01837150752544403,0.014431692659854889,-0.0006711855530738831,0.004302798304706812
3,baseline_direction-24mm,landing,0.0,0.9398022294044495,7.396997451782227,0.017779173329472542,-4.2729079723358154e-06,0.043723247945308685,-0.00018042325973510742,-0.00211150199174881,-0.04372534155845642,0.0028196307830512524
3,baseline_direction-24mm,reload,6.851399898529053,33.19783020019531,53.04985809326172,0.628036379814148,-6.3944607973098755e-06,-1.3932585716247559e-06,0.0013345815241336823,0.0007220804691314697,6.277114152908325e-07,0.01937531679868698
4,baseline_direction-24mm,support,52.748748779296875,52.85877227783203,52.925045013427734,0.9999818205833435,-1.7173588275909424e-06,-1.3485550880432129e-06,-1.5985220670700073e-05,-4.0940940380096436e-05,-5.960464477539063e-08,0.010775136761367321
4,baseline_direction-24mm,unload,9.798057556152344,34.15670394897461,52.934261322021484,0.6461763381958008,-7.964670658111572e-06,-1.434236764907837e-06,-0.0012961290776729584,-0.001278109848499298,-2.898275852203369e-06,0.01769513636827469
4,baseline_direction-24mm,lift,0.0,1.6734437942504883,9.832220077514648,0.031658198684453964,-4.449859261512756e-06,0.04365993291139603,0.00011078640818595886,0.002964310348033905,0.04366418346762657,0.01744244620203972
4,baseline_direction-24mm,swing,0.0,0.0,0.0,0.0,0.043699897825717926,0.054017066955566406,-0.018175136297941208,-0.014890573918819427,0.000783778727054596,0.006423341576009989
4,baseline_direction-24mm,landing,0.0,0.7480117678642273,6.37589693069458,0.014150884002447128,-3.807246685028076e-06,0.04447878897190094,-3.225356340408325e-05,-0.0019776523113250732,-0.0444803461432457,0.001810165704227984
4,baseline_direction-24mm,reload,5.937798023223877,32.19753646850586,54.13505935668945,0.6091127991676331,-5.3998082876205444e-06,-3.5390257835388184e-07,0.0021260157227516174,0.000570446252822876,-1.9744038581848145e-07,0.014237570576369762
5,y_bias-16mm,support,52.23505783081055,52.86397171020508,53.985328674316406,1.0000804662704468,-1.7434358596801758e-06,-7.487833499908447e-07,-3.3818185329437256e-05,-5.445629358291626e-05,2.1047890186309814e-07,0.010374261997640133
5,y_bias-16mm,unload,8.171730995178223,33.66255569458008,53.0526123046875,0.6368280649185181,-6.103888154029846e-06,-1.1138617992401123e-06,-0.0021327808499336243,-0.0009502843022346497,-1.5534460544586182e-06,0.013793233782052994
5,y_bias-16mm,lift,0.0,1.272587776184082,8.224916458129883,0.02407480776309967,-2.6654452085494995e-06,0.04435988515615463,-6.387010216712952e-05,0.002628639340400696,0.04436245560646057,0.012200022116303444
5,y_bias-16mm,swing,0.0,0.0,0.0,0.0,0.04386032372713089,0.05408100038766861,0.009489979594945908,0.012295417487621307,-0.000539138913154602,0.0043068528175354
5,y_bias-16mm,landing,0.0,0.9074799418449402,7.242805480957031,0.017167698591947556,-3.287568688392639e-06,0.04385525733232498,-0.00011193007230758667,-0.002096906304359436,-0.04385732114315033,0.0025537440087646246
5,y_bias-16mm,reload,6.725765228271484,33.07767105102539,53.55429458618164,0.6257631778717041,-6.126239895820618e-06,-1.2926757335662842e-06,0.0016064122319221497,0.0006933882832527161,5.345791578292847e-07,0.018027793616056442
6,y_bias-16mm,support,52.486968994140625,52.852115631103516,52.9252815246582,0.9998559951782227,-1.7229467630386353e-06,-1.2200325727462769e-06,-2.7876347303390503e-05,-3.699213266372681e-05,-1.6205012798309326e-07,0.010842877440154552
6,y_bias-16mm,unload,9.509719848632812,34.18946075439453,52.94326400756836,0.6467959880828857,-6.4838677644729614e-06,-1.3951212167739868e-06,-0.001573760062456131,-0.001220114529132843,-1.8905848264694214e-06,0.016642587259411812
6,y_bias-16mm,lift,0.0,1.597585916519165,9.55402946472168,0.030223118141293526,-3.6247074604034424e-06,0.04377070814371109,2.1435320377349854e-05,0.002900734543800354,0.04377419129014015,0.016067897900938988
6,y_bias-16mm,swing,0.0,0.0,0.0,0.0,0.04381057620048523,0.05407753586769104,-0.009249720722436905,-0.012734860181808472,0.0006802380084991455,0.0058596753515303135
6,y_bias-16mm,landing,0.0,0.7479984164237976,6.375153541564941,0.014150630682706833,-3.933906555175781e-06,0.044485852122306824,-3.6288052797317505e-05,-0.0019793733954429626,-0.044487517327070236,0.0018060122383758426
6,y_bias-16mm,reload,5.937589168548584,32.19605255126953,54.14176940917969,0.6090847849845886,-5.345791578292847e-06,-4.172325134277344e-07,0.0021268315613269806,0.0005712136626243591,-6.51925802230835e-08,0.01423043105751276
1 cycle command phase Fn_min_N Fn_mean_N Fn_max_N mean_load_W gap_min_m gap_max_m dx_m dy_m dz_m root_xy_max_m
2 1 x_bias-16mm support 30.63823699951172 52.85984420776367 74.09165954589844 1.0000022649765015 -1.010112464427948e-05 9.723007678985596e-07 0.0017836317420005798 0.0006739199161529541 8.530914783477783e-07 0.029242604970932007
3 1 x_bias-16mm unload 35.07340621948242 48.76642990112305 53.128562927246094 0.9225630760192871 -2.6773661375045776e-05 -7.014721632003784e-06 -0.002730610780417919 -0.0031594932079315186 -1.9758939743041992e-05 0.07960376888513565
4 1 x_bias-16mm lift 0.0 10.591720581054688 35.157493591308594 0.20037412643432617 -2.9793009161949158e-05 0.044680267572402954 0.06119457259774208 -0.009405404329299927 0.044706881046295166 0.08049342036247253
5 1 x_bias-16mm swing 0.0 0.0 0.0 0.0 0.04415256530046463 0.05414176732301712 0.01172223687171936 0.004630729556083679 -0.0005735680460929871 0.030786940827965736
6 1 x_bias-16mm landing 0.0 0.8175418376922607 6.757045269012451 0.015466248616576195 -3.1776726245880127e-06 0.04414748400449753 -1.6469508409500122e-05 -0.001996748149394989 -0.0441494956612587 0.0022682934068143368
7 1 x_bias-16mm reload 6.254490375518799 32.64689636230469 53.67607116699219 0.6176137328147888 -6.200745701789856e-06 -1.7527490854263306e-06 0.0016945339739322662 0.0006249770522117615 -1.3224780559539795e-07 0.016857551410794258
8 2 x_bias-16mm support 52.44454574584961 52.84786605834961 52.95899963378906 0.9997755885124207 -2.386048436164856e-06 -1.8700957298278809e-06 -2.793222665786743e-05 -3.3736228942871094e-05 -3.203749656677246e-07 0.010354658588767052
9 2 x_bias-16mm unload 8.834904670715332 33.76447677612305 52.96793746948242 0.6387562155723572 -7.227063179016113e-06 -2.0116567611694336e-06 -0.0016921795904636383 -0.0010888651013374329 -1.948326826095581e-06 0.01555535290390253
10 2 x_bias-16mm lift 0.0 1.4353010654449463 8.878164291381836 0.02715301886200905 -4.244968295097351e-06 0.04408973455429077 0.00015369057655334473 0.0028083696961402893 0.044093795120716095 0.014773398637771606
11 2 x_bias-16mm swing 0.0 0.0 0.0 0.0 0.044129468500614166 0.0542534664273262 -0.014362409710884094 -0.006014391779899597 0.00036966055631637573 0.005341693293303251
12 2 x_bias-16mm landing 0.0 0.7480207085609436 6.374540328979492 0.014151052571833134 -4.002824425697327e-06 0.04449426382780075 -3.477931022644043e-05 -0.001990102231502533 -0.04449593275785446 0.0018029295606538653
13 2 x_bias-16mm reload 5.9372711181640625 32.19451904296875 54.12532043457031 0.6090556979179382 -5.416572093963623e-06 -4.3958425521850586e-07 0.002125866711139679 0.000569663941860199 -1.564621925354004e-07 0.014224872924387455
14 3 baseline_direction-24mm support 52.238529205322266 52.86407470703125 53.99081039428711 1.0000823736190796 -1.8570572137832642e-06 -7.972121238708496e-07 -3.374367952346802e-05 -5.480647087097168e-05 5.774199962615967e-07 0.01038394682109356
15 3 baseline_direction-24mm unload 8.168989181518555 33.66178512573242 53.045509338378906 0.6368134617805481 -6.029382348060608e-06 -1.16787850856781e-06 -0.0021336041390895844 -0.0009494572877883911 -1.3634562492370605e-06 0.013789398595690727
16 3 baseline_direction-24mm lift 0.0 1.2718819379806519 8.22209358215332 0.024061456322669983 -2.64681875705719e-06 0.04435984790325165 -6.523355841636658e-05 0.002624101936817169 0.044362492859363556 0.012191063724458218
17 3 baseline_direction-24mm swing 0.0 0.0 0.0 0.0 0.043728090822696686 0.054023608565330505 0.01837150752544403 0.014431692659854889 -0.0006711855530738831 0.004302798304706812
18 3 baseline_direction-24mm landing 0.0 0.9398022294044495 7.396997451782227 0.017779173329472542 -4.2729079723358154e-06 0.043723247945308685 -0.00018042325973510742 -0.00211150199174881 -0.04372534155845642 0.0028196307830512524
19 3 baseline_direction-24mm reload 6.851399898529053 33.19783020019531 53.04985809326172 0.628036379814148 -6.3944607973098755e-06 -1.3932585716247559e-06 0.0013345815241336823 0.0007220804691314697 6.277114152908325e-07 0.01937531679868698
20 4 baseline_direction-24mm support 52.748748779296875 52.85877227783203 52.925045013427734 0.9999818205833435 -1.7173588275909424e-06 -1.3485550880432129e-06 -1.5985220670700073e-05 -4.0940940380096436e-05 -5.960464477539063e-08 0.010775136761367321
21 4 baseline_direction-24mm unload 9.798057556152344 34.15670394897461 52.934261322021484 0.6461763381958008 -7.964670658111572e-06 -1.434236764907837e-06 -0.0012961290776729584 -0.001278109848499298 -2.898275852203369e-06 0.01769513636827469
22 4 baseline_direction-24mm lift 0.0 1.6734437942504883 9.832220077514648 0.031658198684453964 -4.449859261512756e-06 0.04365993291139603 0.00011078640818595886 0.002964310348033905 0.04366418346762657 0.01744244620203972
23 4 baseline_direction-24mm swing 0.0 0.0 0.0 0.0 0.043699897825717926 0.054017066955566406 -0.018175136297941208 -0.014890573918819427 0.000783778727054596 0.006423341576009989
24 4 baseline_direction-24mm landing 0.0 0.7480117678642273 6.37589693069458 0.014150884002447128 -3.807246685028076e-06 0.04447878897190094 -3.225356340408325e-05 -0.0019776523113250732 -0.0444803461432457 0.001810165704227984
25 4 baseline_direction-24mm reload 5.937798023223877 32.19753646850586 54.13505935668945 0.6091127991676331 -5.3998082876205444e-06 -3.5390257835388184e-07 0.0021260157227516174 0.000570446252822876 -1.9744038581848145e-07 0.014237570576369762
26 5 y_bias-16mm support 52.23505783081055 52.86397171020508 53.985328674316406 1.0000804662704468 -1.7434358596801758e-06 -7.487833499908447e-07 -3.3818185329437256e-05 -5.445629358291626e-05 2.1047890186309814e-07 0.010374261997640133
27 5 y_bias-16mm unload 8.171730995178223 33.66255569458008 53.0526123046875 0.6368280649185181 -6.103888154029846e-06 -1.1138617992401123e-06 -0.0021327808499336243 -0.0009502843022346497 -1.5534460544586182e-06 0.013793233782052994
28 5 y_bias-16mm lift 0.0 1.272587776184082 8.224916458129883 0.02407480776309967 -2.6654452085494995e-06 0.04435988515615463 -6.387010216712952e-05 0.002628639340400696 0.04436245560646057 0.012200022116303444
29 5 y_bias-16mm swing 0.0 0.0 0.0 0.0 0.04386032372713089 0.05408100038766861 0.009489979594945908 0.012295417487621307 -0.000539138913154602 0.0043068528175354
30 5 y_bias-16mm landing 0.0 0.9074799418449402 7.242805480957031 0.017167698591947556 -3.287568688392639e-06 0.04385525733232498 -0.00011193007230758667 -0.002096906304359436 -0.04385732114315033 0.0025537440087646246
31 5 y_bias-16mm reload 6.725765228271484 33.07767105102539 53.55429458618164 0.6257631778717041 -6.126239895820618e-06 -1.2926757335662842e-06 0.0016064122319221497 0.0006933882832527161 5.345791578292847e-07 0.018027793616056442
32 6 y_bias-16mm support 52.486968994140625 52.852115631103516 52.9252815246582 0.9998559951782227 -1.7229467630386353e-06 -1.2200325727462769e-06 -2.7876347303390503e-05 -3.699213266372681e-05 -1.6205012798309326e-07 0.010842877440154552
33 6 y_bias-16mm unload 9.509719848632812 34.18946075439453 52.94326400756836 0.6467959880828857 -6.4838677644729614e-06 -1.3951212167739868e-06 -0.001573760062456131 -0.001220114529132843 -1.8905848264694214e-06 0.016642587259411812
34 6 y_bias-16mm lift 0.0 1.597585916519165 9.55402946472168 0.030223118141293526 -3.6247074604034424e-06 0.04377070814371109 2.1435320377349854e-05 0.002900734543800354 0.04377419129014015 0.016067897900938988
35 6 y_bias-16mm swing 0.0 0.0 0.0 0.0 0.04381057620048523 0.05407753586769104 -0.009249720722436905 -0.012734860181808472 0.0006802380084991455 0.0058596753515303135
36 6 y_bias-16mm landing 0.0 0.7479984164237976 6.375153541564941 0.014150630682706833 -3.933906555175781e-06 0.044485852122306824 -3.6288052797317505e-05 -0.0019793733954429626 -0.044487517327070236 0.0018060122383758426
37 6 y_bias-16mm reload 5.937589168548584 32.19605255126953 54.14176940917969 0.6090847849845886 -5.345791578292847e-06 -4.172325134277344e-07 0.0021268315613269806 0.0005712136626243591 -6.51925802230835e-08 0.01423043105751276
@@ -0,0 +1,10 @@
{
"kind": "post-exit observer failure; not a runtime qualification retry",
"original_campaign_success": true,
"attempt": 1,
"exception": "RuntimeError: GPU admission compute/approved-desktop mismatch",
"location": "postexit_verify.py:19 -> rootxy_launch_v2.py:315, GPUOwnership constructor",
"pid_evidence": "The constructor exception did not report actual PID sets; the exact transient cause is unestablished. Do not claim an owned teardown race or a foreign client as proven.",
"followup_observation": "2026-10-16 14:59:29 local nvidia-smi showed compute empty and exactly graphics2420,2618,3268,4486,574096,602685,696142, matching unchanged approval; external observer/result.json confirms actual callerwait0 and campaign/worker success.",
"action": "Preserve this failure, rerun only the same read-only strict postexit verifier after actual wait/clean observation, no new Kit/physics, no whitelist/source changes, no process termination. Later success is a later observation, not retroactive explanation of this exception."
}
@@ -0,0 +1,39 @@
{
"campaign_original_success": true,
"owned_identities_checked": 8,
"owned_alive": [],
"finite_bound_files_checked": 463,
"binding_file_mismatches": [],
"protected_file_mismatches": [],
"gpu_clients": {
"compute": [],
"graphics": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
]
},
"gpu_review": {
"uuid": "GPU-7d882fc3-f771-6d29-4e58-f0fa2a18a3a5",
"compute_pids": [],
"approved_desktop_graphics_pids": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
],
"task_graphics_pids": [],
"clean": true,
"scope": "owned-compute-and-graphics.plus-pinned-desktop-graphics-only"
},
"static_check_not_same_cfg_proof": true,
"boottime": 362094.201891648,
"passed": true
}
@@ -0,0 +1,23 @@
{
"version": "0.4.2",
"accepted_skill": false,
"package": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2",
"archive": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2.tar.gz",
"archive_bytes": 96636332,
"archive_sha256": "16c855f6086a90aedc3f50eb510d44f64f802254505146a61c83402df7d99151",
"checksum_file": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2.tar.gz.sha256",
"payload_files_including_manifest": 1329,
"sha256sums_entries": 1328,
"authority_inputs_match_native_staging": 41,
"all_manifest_files_match": true,
"every_archive_payload_readback_verified": true,
"no_illegal_members": true,
"native_run_at_delivery_path": false,
"native_qualification_scope": "analyticS1GUI/headless27000ticks each; source-identical copied artifact, not fullmatrix/u256/nativefaultinjection",
"CPU_package": {
"passed": 1557,
"skipped": 6,
"failed": 0
},
"validation_receipt": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/delivery-validation-final.json"
}
+429
View File
@@ -0,0 +1,429 @@
{
"baseline-profile02": {
"loop": {
"active_loop_wall_seconds": 113.54154362500412,
"actual_render_generations_per_wall_second": 1.9816535236046455,
"control_calls": 900,
"cprofile_enabled": true,
"mode": "baseline",
"physics_samples": 4500,
"physics_ticks_per_wall_second": 39.63307047209291,
"render_generations": 225,
"scope": "Actual env.step loop incl gates/recording/UI. Startup and final viewing excluded. cProfile adds overhead. Render generations are submissions, not display-present GPU timings.",
"simulation_seconds_per_wall_second": 0.07926614094418583
},
"gpu_window_basis": "estimated: profile_file_mtime minus measured_loop_duration; slight serialization-offset uncertainty",
"device_wide_gpu_samples": {
"utilization.gpu [%]": {
"n": 227,
"mean": 16.198237885462554,
"min": 1.0,
"max": 24.0
},
"utilization.memory [%]": {
"n": 227,
"mean": 2.488986784140969,
"min": 1.0,
"max": 8.0
},
"power.draw [W]": {
"n": 227,
"mean": 75.59013215859031,
"min": 59.14,
"max": 77.94
},
"power.limit [W]": {
"n": 227,
"mean": 360.0,
"min": 360.0,
"max": 360.0
},
"clocks.current.sm [MHz]": {
"n": 227,
"mean": 2895.9030837004407,
"min": 2647.0,
"max": 2910.0
},
"memory.used [MiB]": {
"n": 227,
"mean": 6310.920704845815,
"min": 5990.0,
"max": 6391.0
}
}
},
"sensor-profile03": {
"loop": {
"active_loop_ended_wall": 1792129722.1940768,
"active_loop_started_wall": 1792129630.083723,
"active_loop_wall_seconds": 92.10295005096123,
"actual_render_generations_per_wall_second": 2.4429184936585187,
"capture_implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.recording.SimCycleRecorder",
"control_calls": 900,
"cprofile_enabled": true,
"mode": "sensor_epoch",
"physics_samples": 4500,
"physics_ticks_per_wall_second": 48.858369873170375,
"render_generations": 225,
"scope": "Actual env.step loop incl gates/recording/UI. Startup and final viewing excluded. cProfile adds overhead. Render generations are submissions, not display-present GPU timings.",
"sensor_epochs": {
"foot_contact": {
"fetches": 4501,
"invalidations": 4501,
"requests": 57614
},
"mount_wrench": {
"fetches": 0,
"invalidations": 4501,
"requests": 0
}
},
"simulation_seconds_per_wall_second": 0.09771673974634075
},
"gpu_window_basis": "recorded_loop_wall",
"device_wide_gpu_samples": {
"utilization.gpu [%]": {
"n": 184,
"mean": 17.60326086956522,
"min": 1.0,
"max": 25.0
},
"utilization.memory [%]": {
"n": 184,
"mean": 2.3206521739130435,
"min": 1.0,
"max": 8.0
},
"power.draw [W]": {
"n": 184,
"mean": 76.3308152173913,
"min": 58.61,
"max": 78.19
},
"power.limit [W]": {
"n": 184,
"mean": 360.0,
"min": 360.0,
"max": 360.0
},
"clocks.current.sm [MHz]": {
"n": 184,
"mean": 2896.4239130434785,
"min": 2647.0,
"max": 2910.0
},
"memory.used [MiB]": {
"n": 184,
"mean": 6313.483695652174,
"min": 5994.0,
"max": 6395.0
}
}
},
"combined-profile04": {
"loop": {
"active_loop_ended_wall": 1792130851.711445,
"active_loop_started_wall": 1792130775.9220383,
"active_loop_wall_seconds": 75.78353441797663,
"actual_render_generations_per_wall_second": 2.9689826652717812,
"capture_implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.packed_recording_v042.ScopedPackedSimCycleRecorder",
"control_calls": 900,
"cprofile_enabled": true,
"jacobian_qualification": {
"all_bitwise_equal": true,
"applied": true,
"epsilon_rad": 0.0005,
"ik_parameters_changed": false,
"max_abs": 0.0,
"qualification": "sampled same-device same-dtype original-vs-batched math; not all-input proof",
"reason": "exact_samples",
"samples": 33
},
"mode": "combined",
"physics_samples": 4500,
"physics_ticks_per_wall_second": 59.379653305435625,
"render_generations": 225,
"scope": "Actual env.step loop incl gates/recording/UI. Startup and final viewing excluded. cProfile adds overhead. Render generations are submissions, not display-present GPU timings.",
"sensor_epochs": {
"foot_contact": {
"fetches": 4501,
"invalidations": 4501,
"requests": 21606
},
"mount_wrench": {
"fetches": 4501,
"invalidations": 4501,
"requests": 9903
}
},
"simulation_seconds_per_wall_second": 0.11875930661087125
},
"gpu_window_basis": "recorded_loop_wall",
"device_wide_gpu_samples": {
"utilization.gpu [%]": {
"n": 152,
"mean": 19.723684210526315,
"min": 1.0,
"max": 28.0
},
"utilization.memory [%]": {
"n": 152,
"mean": 3.0921052631578947,
"min": 1.0,
"max": 8.0
},
"power.draw [W]": {
"n": 152,
"mean": 77.5408552631579,
"min": 58.7,
"max": 79.91
},
"power.limit [W]": {
"n": 152,
"mean": 360.0,
"min": 360.0,
"max": 360.0
},
"clocks.current.sm [MHz]": {
"n": 152,
"mean": 2894.2302631578946,
"min": 2640.0,
"max": 2910.0
},
"memory.used [MiB]": {
"n": 152,
"mean": 6427.3223684210525,
"min": 6108.0,
"max": 6506.0
}
}
},
"display-S1-05": {
"loop": {
"active_loop_ended_wall": 1792133436.1219404,
"active_loop_started_wall": 1792132671.2588954,
"active_loop_wall_seconds": 764.8630335039925,
"actual_render_generations_per_wall_second": 35.30043787880234,
"capture_implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.packed_recording_v042.ScopedPackedSimCycleRecorder",
"control_calls": 5400,
"cprofile_enabled": false,
"jacobian_qualification": {
"all_bitwise_equal": true,
"applied": true,
"epsilon_rad": 0.0005,
"ik_parameters_changed": false,
"max_abs": 0.0,
"qualification": "sampled same-device same-dtype original-vs-batched math; not all-input proof",
"reason": "exact_samples",
"samples": 33
},
"mode": "combined_display",
"physics_samples": 27000,
"physics_ticks_per_wall_second": 35.30043787880234,
"presentation": {
"additional_renders": 25650,
"adds_physics": false,
"interpolation": false,
"last_verified_tick": 27000,
"original_render_interval": 20,
"policy": "one current-state render per verified physical tick; no frame cap",
"presentation_vs_throughput_tradeoff": true
},
"render_generations": 27000,
"scope": "Actual env.step loop incl gates/recording/UI. Startup and final viewing excluded. cProfile adds overhead. Render generations are submissions, not display-present GPU timings.",
"sensor_epochs": {
"foot_contact": {
"fetches": 27001,
"invalidations": 27001,
"requests": 129606
},
"mount_wrench": {
"fetches": 27001,
"invalidations": 27001,
"requests": 59403
}
},
"simulation_seconds_per_wall_second": 0.07060087575760468
},
"gpu_window_basis": "recorded_loop_wall",
"device_wide_gpu_samples": {
"utilization.gpu [%]": {
"n": 1528,
"mean": 39.73560209424084,
"min": 1.0,
"max": 49.0
},
"utilization.memory [%]": {
"n": 1528,
"mean": 20.138089005235603,
"min": 1.0,
"max": 30.0
},
"power.draw [W]": {
"n": 1528,
"mean": 133.430222513089,
"min": 61.58,
"max": 137.97
},
"power.limit [W]": {
"n": 1528,
"mean": 360.0,
"min": 360.0,
"max": 360.0
},
"clocks.current.sm [MHz]": {
"n": 1528,
"mean": 2887.8219895287957,
"min": 2707.0,
"max": 2902.0
},
"memory.used [MiB]": {
"n": 1528,
"mean": 6308.526832460733,
"min": 6271.0,
"max": 6360.0
}
}
},
"limits": "Whole-device GPU samples incl baseline desktop. Like-profiled single-cycle baseline02/combined04 supports algorithm speed comparison; disabled-profiler fullS1 display05 includes120s30HzCPUrecording and early concurrent CPUtests, not a controlled pure-algorithm comparison. Render generations are SDK submissions, not independent display-present timestamps. Actual window overlay at capturedtick2519 reads33.98FPS.",
"like_profiled_speed_ratio": 1.4982349991592754,
"trace_sha256": {
"baseline-profile02": "2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94",
"sensor-profile03": "2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94",
"combined-profile04": "2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94",
"display-S1-05": "d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba"
},
"headless-S1-06": {
"loop": {
"active_loop_ended_wall": 1792133870.5507057,
"active_loop_started_wall": 1792133575.8713605,
"active_loop_wall_seconds": 294.67933065403486,
"actual_render_generations_per_wall_second": 0.0,
"capture_implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.packed_recording_v042.ScopedPackedSimCycleRecorder",
"control_calls": 5400,
"cprofile_enabled": false,
"jacobian_qualification": {
"all_bitwise_equal": true,
"applied": true,
"epsilon_rad": 0.0005,
"ik_parameters_changed": false,
"max_abs": 0.0,
"qualification": "sampled same-device same-dtype original-vs-batched math; not all-input proof",
"reason": "exact_samples",
"samples": 33
},
"mode": "combined_display",
"physics_samples": 27000,
"physics_ticks_per_wall_second": 91.62502147698667,
"presentation": null,
"render_generations": 0,
"scope": "Actual env.step loop incl gates/recording/UI. Startup and final viewing excluded. cProfile adds overhead. Render generations are submissions, not display-present GPU timings.",
"sensor_epochs": {
"foot_contact": {
"fetches": 27001,
"invalidations": 27001,
"requests": 129606
},
"mount_wrench": {
"fetches": 27001,
"invalidations": 27001,
"requests": 59403
}
},
"simulation_seconds_per_wall_second": 0.18325004295397335
}
},
"cases": [
{
"name": "baseline-profile01",
"original_success": false,
"Kit_instances": 1,
"actual_physics_calls": 0,
"reserved_physics_calls": 4750,
"original_campaign_sha256": "7b1779f0f734c03d9bdd95aed47b95ee486ed8b0b2c1c1b24850a5b2bfe3d532",
"waited_cleanup": true,
"source_inputs": 35
},
{
"name": "baseline-profile02",
"original_success": true,
"Kit_instances": 2,
"actual_physics_calls": 4500,
"reserved_physics_calls": 4750,
"original_campaign_sha256": "5e446923f62cd62ae38154ba825b6448518bc8578a957e46bcf6fd0290c741b7",
"waited_cleanup": true,
"source_inputs": 35
},
{
"name": "combined-profile04",
"original_success": true,
"Kit_instances": 2,
"actual_physics_calls": 4500,
"reserved_physics_calls": 4750,
"original_campaign_sha256": "eb80f898a2abf31ed6e98b24b56721390a5b4ea516cdf7ae85b556dff25a00ef",
"waited_cleanup": true,
"source_inputs": 39
},
{
"name": "display-S1-05",
"original_success": true,
"Kit_instances": 2,
"actual_physics_calls": 27000,
"reserved_physics_calls": 27250,
"original_campaign_sha256": "38cd9d5dcc45c0dca02d880d475c81efa6725c75657cd18d1d64ca7e5fa6f130",
"waited_cleanup": true,
"source_inputs": 41
},
{
"name": "headless-S1-06",
"original_success": true,
"Kit_instances": 2,
"actual_physics_calls": 27000,
"reserved_physics_calls": 27250,
"original_campaign_sha256": "dd247da7ae4e830ad4420e0ea7006db6390642cc8269079496f22f3356f6141f",
"waited_cleanup": true,
"source_inputs": 41
},
{
"name": "sensor-profile03",
"original_success": true,
"Kit_instances": 2,
"actual_physics_calls": 4500,
"reserved_physics_calls": 4750,
"original_campaign_sha256": "fd85fcdc45e0084f4473092ec6bffc1366351ca8806ff4f428b10193f0ea96e5",
"waited_cleanup": true,
"source_inputs": 37
}
],
"accounting": {
"Kit_instances": 11,
"actual_physics_calls": 67500,
"reserved_physics_calls": 73500,
"optimizer_steps": 0,
"owned_identities_checked": 57,
"owned_alive": [],
"elapsed_seconds": 6537.71110534668
},
"effective_config_differences": [
"/fields/mapping/sim/fields/mapping/visualizer_cfgs/list"
],
"current_source_inputs_equal": true,
"binding_base_source_input_count": 30,
"authority_source_input_count": 41,
"accepted_skill": false,
"training_steps": 0,
"failure_disclosure": "baseline-profile01 failed PREP before physics and remains failed. Both reviewP1findings fixed before native qualification, CPU collection failure preserved. First independent headless06 postexit observer raised GPU admission mismatch without PID evidence; exact cause unknown. Original runtime success unchanged. After callerwait0 and clean nvidia-smi observation, unchanged strict verifier second observation passed; no whitelist/source changes, no new Kit/physics. Shared native cache monitoring not a hard quota.",
"qualification_scope": "Current-source analytic S1 only: GUI6cycles27000ticks and headless6cycles27000ticks,70tensors byte-identical to each other and prior0.4.1GUI11 including entire trace-fileSHA. No current-source nominal20/S2/S3/u256 qualification, native new-cadence early-abort/visualizer-failure injection/startup-close/GUI-safe-stop/600s viewing-timeout test, global determinism or opaque native-state completeness claim. Earlyclose/visualizerfailure protections CPUnegative-tested.",
"postexit_reports": [
"display05-postexit.json",
"headless06-postexit-first-failure.json",
"headless06-postexit-second.json"
],
"cpu_acceptance": {
"passed": 1557,
"skipped": 6,
"failed": 0,
"log": "cpu-full03.log",
"xml": "cpu-full03.xml"
}
}
@@ -0,0 +1,296 @@
{
"baseline-profile02": {
"loop": {
"active_loop_wall_seconds": 113.54154362500412,
"actual_render_generations_per_wall_second": 1.9816535236046455,
"control_calls": 900,
"cprofile_enabled": true,
"mode": "baseline",
"physics_samples": 4500,
"physics_ticks_per_wall_second": 39.63307047209291,
"render_generations": 225,
"scope": "Actual env.step loop incl gates/recording/UI. Startup and final viewing excluded. cProfile adds overhead. Render generations are submissions, not display-present GPU timings.",
"simulation_seconds_per_wall_second": 0.07926614094418583
},
"gpu_window_basis": "estimated: profile_file_mtime minus measured_loop_duration; slight serialization-offset uncertainty",
"device_wide_gpu_samples": {
"utilization.gpu [%]": {
"n": 227,
"mean": 16.198237885462554,
"min": 1.0,
"max": 24.0
},
"utilization.memory [%]": {
"n": 227,
"mean": 2.488986784140969,
"min": 1.0,
"max": 8.0
},
"power.draw [W]": {
"n": 227,
"mean": 75.59013215859031,
"min": 59.14,
"max": 77.94
},
"power.limit [W]": {
"n": 227,
"mean": 360.0,
"min": 360.0,
"max": 360.0
},
"clocks.current.sm [MHz]": {
"n": 227,
"mean": 2895.9030837004407,
"min": 2647.0,
"max": 2910.0
},
"memory.used [MiB]": {
"n": 227,
"mean": 6310.920704845815,
"min": 5990.0,
"max": 6391.0
}
}
},
"sensor-profile03": {
"loop": {
"active_loop_ended_wall": 1792129722.1940768,
"active_loop_started_wall": 1792129630.083723,
"active_loop_wall_seconds": 92.10295005096123,
"actual_render_generations_per_wall_second": 2.4429184936585187,
"capture_implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.recording.SimCycleRecorder",
"control_calls": 900,
"cprofile_enabled": true,
"mode": "sensor_epoch",
"physics_samples": 4500,
"physics_ticks_per_wall_second": 48.858369873170375,
"render_generations": 225,
"scope": "Actual env.step loop incl gates/recording/UI. Startup and final viewing excluded. cProfile adds overhead. Render generations are submissions, not display-present GPU timings.",
"sensor_epochs": {
"foot_contact": {
"fetches": 4501,
"invalidations": 4501,
"requests": 57614
},
"mount_wrench": {
"fetches": 0,
"invalidations": 4501,
"requests": 0
}
},
"simulation_seconds_per_wall_second": 0.09771673974634075
},
"gpu_window_basis": "recorded_loop_wall",
"device_wide_gpu_samples": {
"utilization.gpu [%]": {
"n": 184,
"mean": 17.60326086956522,
"min": 1.0,
"max": 25.0
},
"utilization.memory [%]": {
"n": 184,
"mean": 2.3206521739130435,
"min": 1.0,
"max": 8.0
},
"power.draw [W]": {
"n": 184,
"mean": 76.3308152173913,
"min": 58.61,
"max": 78.19
},
"power.limit [W]": {
"n": 184,
"mean": 360.0,
"min": 360.0,
"max": 360.0
},
"clocks.current.sm [MHz]": {
"n": 184,
"mean": 2896.4239130434785,
"min": 2647.0,
"max": 2910.0
},
"memory.used [MiB]": {
"n": 184,
"mean": 6313.483695652174,
"min": 5994.0,
"max": 6395.0
}
}
},
"combined-profile04": {
"loop": {
"active_loop_ended_wall": 1792130851.711445,
"active_loop_started_wall": 1792130775.9220383,
"active_loop_wall_seconds": 75.78353441797663,
"actual_render_generations_per_wall_second": 2.9689826652717812,
"capture_implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.packed_recording_v042.ScopedPackedSimCycleRecorder",
"control_calls": 900,
"cprofile_enabled": true,
"jacobian_qualification": {
"all_bitwise_equal": true,
"applied": true,
"epsilon_rad": 0.0005,
"ik_parameters_changed": false,
"max_abs": 0.0,
"qualification": "sampled same-device same-dtype original-vs-batched math; not all-input proof",
"reason": "exact_samples",
"samples": 33
},
"mode": "combined",
"physics_samples": 4500,
"physics_ticks_per_wall_second": 59.379653305435625,
"render_generations": 225,
"scope": "Actual env.step loop incl gates/recording/UI. Startup and final viewing excluded. cProfile adds overhead. Render generations are submissions, not display-present GPU timings.",
"sensor_epochs": {
"foot_contact": {
"fetches": 4501,
"invalidations": 4501,
"requests": 21606
},
"mount_wrench": {
"fetches": 4501,
"invalidations": 4501,
"requests": 9903
}
},
"simulation_seconds_per_wall_second": 0.11875930661087125
},
"gpu_window_basis": "recorded_loop_wall",
"device_wide_gpu_samples": {
"utilization.gpu [%]": {
"n": 152,
"mean": 19.723684210526315,
"min": 1.0,
"max": 28.0
},
"utilization.memory [%]": {
"n": 152,
"mean": 3.0921052631578947,
"min": 1.0,
"max": 8.0
},
"power.draw [W]": {
"n": 152,
"mean": 77.5408552631579,
"min": 58.7,
"max": 79.91
},
"power.limit [W]": {
"n": 152,
"mean": 360.0,
"min": 360.0,
"max": 360.0
},
"clocks.current.sm [MHz]": {
"n": 152,
"mean": 2894.2302631578946,
"min": 2640.0,
"max": 2910.0
},
"memory.used [MiB]": {
"n": 152,
"mean": 6427.3223684210525,
"min": 6108.0,
"max": 6506.0
}
}
},
"display-S1-05": {
"loop": {
"active_loop_ended_wall": 1792133436.1219404,
"active_loop_started_wall": 1792132671.2588954,
"active_loop_wall_seconds": 764.8630335039925,
"actual_render_generations_per_wall_second": 35.30043787880234,
"capture_implementation": "SkillFactory.tasks.manager_based.contact_cycle_sim.packed_recording_v042.ScopedPackedSimCycleRecorder",
"control_calls": 5400,
"cprofile_enabled": false,
"jacobian_qualification": {
"all_bitwise_equal": true,
"applied": true,
"epsilon_rad": 0.0005,
"ik_parameters_changed": false,
"max_abs": 0.0,
"qualification": "sampled same-device same-dtype original-vs-batched math; not all-input proof",
"reason": "exact_samples",
"samples": 33
},
"mode": "combined_display",
"physics_samples": 27000,
"physics_ticks_per_wall_second": 35.30043787880234,
"presentation": {
"additional_renders": 25650,
"adds_physics": false,
"interpolation": false,
"last_verified_tick": 27000,
"original_render_interval": 20,
"policy": "one current-state render per verified physical tick; no frame cap",
"presentation_vs_throughput_tradeoff": true
},
"render_generations": 27000,
"scope": "Actual env.step loop incl gates/recording/UI. Startup and final viewing excluded. cProfile adds overhead. Render generations are submissions, not display-present GPU timings.",
"sensor_epochs": {
"foot_contact": {
"fetches": 27001,
"invalidations": 27001,
"requests": 129606
},
"mount_wrench": {
"fetches": 27001,
"invalidations": 27001,
"requests": 59403
}
},
"simulation_seconds_per_wall_second": 0.07060087575760468
},
"gpu_window_basis": "recorded_loop_wall",
"device_wide_gpu_samples": {
"utilization.gpu [%]": {
"n": 1528,
"mean": 39.73560209424084,
"min": 1.0,
"max": 49.0
},
"utilization.memory [%]": {
"n": 1528,
"mean": 20.138089005235603,
"min": 1.0,
"max": 30.0
},
"power.draw [W]": {
"n": 1528,
"mean": 133.430222513089,
"min": 61.58,
"max": 137.97
},
"power.limit [W]": {
"n": 1528,
"mean": 360.0,
"min": 360.0,
"max": 360.0
},
"clocks.current.sm [MHz]": {
"n": 1528,
"mean": 2887.8219895287957,
"min": 2707.0,
"max": 2902.0
},
"memory.used [MiB]": {
"n": 1528,
"mean": 6308.526832460733,
"min": 6271.0,
"max": 6360.0
}
}
},
"limits": "Whole-device GPU samples incl baseline desktop. Like-profiled single-cycle baseline02/combined04 supports algorithm speed comparison; disabled-profiler fullS1 display05 includes120s30HzCPUrecording and early concurrent CPUtests, not a controlled pure-algorithm comparison. Render generations are SDK submissions, not independent display-present timestamps. Actual window overlay at capturedtick2519 reads33.98FPS.",
"like_profiled_speed_ratio": 1.4982349991592754,
"trace_sha256": {
"baseline-profile02": "2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94",
"sensor-profile03": "2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94",
"combined-profile04": "2ce511b4c82b6555db30a99bb70dd0231b43ed06ebf7657d57f93c7b1a950e94",
"display-S1-05": "d0934245e50f3d7e174b24c109f167236c4693a023d1da01bbadfd71e41b4dba"
}
}
+16
View File
@@ -0,0 +1,16 @@
# 0.4.2 application hot-path profiling and optimization
Operator: current request explicitly permits full use of RTX5080 and authorizes artifacts under this repository. No more output-path questions. Existing AGENT.md authorizes bounded autonomous work and read-only specialists.
This is an Isaac Lab application-loop performance task, not USD geometry optimization. The omniverse-usd-performance-tuning entry triages it to application/runtime profiling; no USD restructuring, simplification, SO/AV operation or standalone install is required or claimed. Runtime is the already-used local IsaacLab /home/cen/IsaacLab + Isaac Sim /home/cen/isaacsim-6.0.1 on CUDA0 RTX5080, not a CPU fallback. Runtime metadata/probe is recorded in ../setup-preflight.json. A missing SO/AV blocks those unrequested operations, not this Python hot-path benchmark. All generated performance artifacts are under this operator-selected repository (this subdirectory and scripts/).
Bounded task: at most12Kit instances,120000 reserved/actual target-environment calls,24GiB owned logical storage,6hours from scope.json; zero training/parameter search, no artificial GPU power/clock cap, no foreign process termination. Serial runs only; unchanged strict GPU admission/monitoring/final ownership and exact-SHA shutdown handoff. New session/grant/output each run, whole-task parent accounting in addition to per-session ledgers. Do not reuse completed0.4.1 authority.
Preserve old0.4.0 and0.4.1 packages/archive/checksums and all historical evidence. Parent sole writer, no source/config edits while a run is active. Snapshot each attempt's approved code, including unsuccessful attempts. Frozen candidate and original physics/controller/limits/gates/500Hz physics100Hz controls/render_interval20 stay unchanged. Do not thin500Hz traces, postpone failure stopping, bypass provenance, fake FPS via prerecorded/interpolated output, add dummy GPU load, enlarge the task to multiple legs/envs, or optimize RootXY.
1. Bind a CPU cProfile/perf-counter observer before PREP, measure a fresh analytic GUI run with requested safe stop after first complete cycle. Keep startup, active loop, control/physics throughput, real render cadence and CPU profiling overhead separate. Sample real GPU util/power/clocks externally without changing settings. No guesses about utilization.
2. Optimize evidenced CPU/GPU transfers/duplicate read-only computation while keeping fail-before-next-step semantics and exact every-tick values. Prefer batching/gathering to omitting checks. CPU equivalence/negative tests, original source hashes and reviewer before actual optimized run.
3. Compare same-seed same-protocol real traces and observed native counters. Report unchanged gate results and equality/tolerances honestly; CPU profile comparisons share profiler overhead, final benchmark disables profiling.
4. Run bounded actual full analytic S1 GUI and headless regression on accepted implementation, plus early-close as needed for affected behavior. Preserve current source/version independently; seal only evidence-qualified artifact, disclose residual bottlenecks and do not promiseGPU100% or1x before measured.
No USD stage mutation chain is planned. No optimized USD/report is claimed from SO/AV. This task produces application profiling JSON/Markdown, raw traces and a versioned callable implementation if accepted.
@@ -0,0 +1,11 @@
# Presentation-cadence refinement within the authorized FPS task
The original plan/scope/caps and sealed0.4.0/0.4.1 remain unchanged. The four frozen physical/controller/protocol files and original sim.render_interval=20 remain unchanged. This document is separately authenticated in each subsequent approval.
Measured pure application optimization:4500tick GUI baseline113.5415s → epoch92.1030s → combined75.7835s, all with cProfile. All three complete trace files have identical SHA256;70tensors. Combined sampled strict float32-bit Jacobian qualification33samples passed; not an all-input proof. Full CPU1536passed6skipped. Combined original campaign/strict postexit passed,4427protectedfiles unchanged. Render submissions are still only225 per9simulatedseconds, so about3 per WALLsecond. Faster loop alone is not adequate GUI responsiveness.
New combined_display mode retains original interval20 renders and adds a genuine render on other ticks, only after the original safety checks and synchronous complete500Hz evidence succeed. Thus each moving render represents a NEW actually integrated and checked physical tick; no interpolation, repeated-state filler frames, synthetic FPS, movie playback, sleep/frame cap, environment enlargement or dummy GPU load. It calls sim.render, never sim.step, scene.update/reset, controller refresh, time scaling, teleport or parameter mutation. The pinned KitVisualizer disables player physics while pumping the app. The expected concrete Kit visualizer must remain a member, running, not closed and not paused before/after added rendering, before each subsequent physical call, after each control/original interval20 rendering, and during/following final viewing. SDK-swallowed visualizer failures must fail closed without counting missing paints. Original GUIclose finalization and exactSHA/liveGPU shutdown handoff remain mandatory. Early close during an added frame before total_ticks returns the original GUIabort cause before another step. Final-tick close retains completion semantics.
This intentionally separates UI responsiveness from simulation throughput. More rendering may LOWER physics ticks/wallsecond; actual real-time factor and render submissions/wallsecond MUST both be reported, with startup/finalhold/profile overhead separated. Do not label submissions as independently measured display-present FPS. Genuine Kit-window screenshots/video/overlay and sampled GPUutil/power/clocks are separate evidence. This does not promise1x physical time,60displayFPS or100%GPU utilization.
Before qualification: CPU sequencing/negative tests and read-only review. Then one fresh serial full analyticS1GUI run (27000actualticks,profiledisabled), full trace/audit/counter/continuity/source checks against old sameprotocol evidence, actual finalview/close/wait and strict cleanup. Same-source analyticS1headless regression follows, subject to original aggregate12Kit/120000calls/24GiB/6h. No training/u256/current-source fullmatrix qualification is implied. No source/config/plan edits during an active bound run.
@@ -0,0 +1,88 @@
# Read-only hot-path review
## Review
- **Correct:** Every physics tick completes latch updates, safety checks and synchronous evidence capture before rendering or another physics step (`source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_env.py:447–469`). Preserve this ordering.
- **Correct:** RuntimeHooks retains pre-step admission, counters, unique wrench ownership and reset/write guards (`evaluation_env.py:114–201`).
- **Fixed:** None; read-only review.
- **No issues found.**
- **Merge verdict: OK with notes** for the profiling approach—not approval of an optimized implementation or a performance claim.
Inspected the requested original source seams, observer/binding integration, and relevant CPU tests. No commands, tests, GPU operations or edits were performed. No `performance-profile.json` was available when checked; no measured hot-function ranking is claimed.
Below, task-module paths are relative to `source/SkillFactory/SkillFactory/tasks/manager_based/`.
## Why a single leg can look slow
The implementation contains substantial **serial CPU/GPU handshaking around small tensor workloads**. Training throughput across many environments and single-environment GUI frame rate measure different things; aggregate training throughput does not establish single-leg interactive performance.
The environment requests rendering every 20 physics ticks. At 500 Hz simulation time, that is **25 render requests per simulated second**, not necessarily 25 per wall second. Slow execution of those intervening physics/check/recording ticks also slows GUI updates without the renderer itself being the dominant cost.
These are code-supported explanations, **not a measured attribution of the reported 2–3 FPS**. Current evidence does not establish GPU utilization, rendering cost, or an achievable speedup. Maximizing useful execution speed also does not require—or guarantee—100% RTX5080 utilization for this small serial workload.
## Evidenced optimization opportunities
| Location | Code evidence | Advice |
|---|---|---|
| `contact_cycle/recording.py:109–139`; `contact_cycle_sim/recording.py:14–26` | Snapshot builds many tensor fields and independently executes `detach().cpu().clone()` for each. Simulation extras repeat this pattern. Some metadata is already CPU-resident. | **First candidate:** pack GPU snapshot fields by exact dtype/device, copy each packed group synchronously once, then reconstruct the original CPU fields. Keep CPU metadata separate. |
| `contact_cycle_sim/evaluation_mdp.py:119–168` | Safety retrieves failure flags to Python and converts numerous individual GPU predicates to Python booleans. It also recreates limit tensors each call. | Subsequently batch boolean predicate results into one synchronous CPU transfer. Preserve each predicate’s arithmetic, thresholds, cause ordering and overlapping causes. Cache immutable limit tensors per state instance. |
| `contact_cycle_sim/evaluation_env.py:325–332`; `evaluation_mdp.py:25–26, 86–104` | Each event converts two reference vectors to CPU tuples; protocol-command construction reads GPU weight as a Python float. Reference/port-reference helpers repeatedly reconstruct commands and device tensors. | Secondary candidate only. Reuse read-only results within a precisely bounded snapshot/reference epoch; never cache merely by global tick. |
| `contact_cycle/recording.py:122`; `contact_cycle/mdp.py:261–275` | Snapshot calls both `failed()` and `failure_flags()`; `failed()` recomputes the flags. | Compute flags once within that snapshot and derive `would_fail = flags.any(-1)`. Do not remove the separate safety gate. |
| `contact_cycle_sim/recording.py:22–23`; `contact_cycle_sim/mdp.py:106–113` | Position and velocity cost helpers each compute both components through `pose_cost`. | Reuse that read-only result inside the snapshot, retaining the original arithmetic and outputs. |
| `contact_cycle_sim/functional_flow_recording_v2.py:175–198` | Every frame receives full finite, bounds, composition and identity checks; additional nominal CPU copying occurs during composition validation. | Retain every check. Consider instance-local immutable nominal CPU data only after equivalence qualification. |
| `contact_cycle_sim/skill_gui_v041.py:105–137` | GUI updates run each control call, including GPU scalar reads for tick/contact. Other displayed physical values already come from CPU evidence. | Lower-priority opportunity: read matching tick/contact evidence from the same completed frame where available. Do not change render cadence or evidence frequency. |
### Introspection: no established per-tick culprit
`RuntimeHooks.isaac()` enumerates methods during installation; guard setup and `functools.wraps` are not repeated discovery operations on every tick (`evaluation_env.py:69–120, 161–201`).
There is **no explicit per-tick `inspect.signature`, `inspect.stack` or source inspection in the requested application hot paths**. Installed Isaac code contains introspection elsewhere, but its existence does not establish active-loop cost. Do not remove provenance checks, bypass wrappers, or globally cache/patch introspection without a profile identifying the actual caller.
## Lowest-risk first implementation
Implement a **new bound packed recorder**, leaving the original recorder and frozen physical/controller sources untouched.
1. Construct the same snapshot values using the same GPU expressions.
2. Flatten/group GPU tensors by **exact dtype and device**. Never cast counters, booleans or integer phase fields into floating-point storage.
3. Perform synchronous device-to-host copies before the evidence callback returns.
4. Reconstruct the same keys, shapes, dtypes, values and independent snapshot ownership.
5. Run all existing frame checks and window/endpoint audits before another physics step.
Do **not** defer copies to a later control, buffer unchecked ticks, or use unfinished asynchronous transfers. Keep initial snapshots, offending samples, next-reference evidence and seam fingerprints.
This first version should optimize **transfer packaging**, not simultaneously rewrite safety predicates, command caching and GUI behavior. Smaller scope makes exact-equivalence failures easier to localize. Packing adds its own concatenation/allocation work; improvement remains unmeasured.
The observer currently accepts `mode="packed"` but does not itself select an implementation (`contact_cycle_sim/performance_v042.py:13–18`); environment/recorder construction remains explicit in `scripts/evaluate_cycle_functional_flow.py:219–231`. A mode label alone is not evidence that packing occurred.
Any new implementation must join authenticated source/import inventories before PREP. Existing integration points are:
- `source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_binding_v4.py:8–17, 25–56`
- `scripts/skill_launch_v4.py:19–40`
## Required qualification before acceptance
**Parent-run tests only, after the active run ends:**
- Differential original/packed snapshots: identical keys, dtype, shape and values; include noncontiguous tensors, CPU metadata, booleans, integers, empty fields and nonfinite values.
- Snapshot ownership: mutating source tensors or later snapshots must not alter earlier evidence.
- Failure injection at physics substep 2: retain the offending frame, report all original causes in order, and execute no third step or post-failure render.
- Transfer/reconstruction exceptions must fail closed before another physics step.
- Preserve exact window decisions at ticks 1499/1999/2999; endpoint evidence, admission, carry/RNG identity and exactly one refresh at seams.
- Compare original/optimized complete traces and gate results; disclose any tolerance rather than calling unequal results exact.
- Negative provenance tests: changed implementation, mode/config or unbound imports must reject before physics. Retain original GPU admission and SHA-bound shutdown ownership.
- GUI/headless regression and early-close/final-hold checks: no added physics, changed render interval or altered cleanup semantics.
Existing suites to extend/rerun include:
`test_contact_cycle_sim_evaluation_runtime.py`,
`test_contact_cycle_functional_flow_runtime_v2.py`,
`test_contact_cycle_skill_v4.py`,
`test_contact_cycle_gui_v041.py`, and
`test_contact_cycle_gui_shutdown_v041.py`.
Benchmark matched protocols with profiling enabled on both sides for diagnosis, then disabled for final wall-throughput reporting. `LoopProfile` correctly separates active-loop timing from startup/final viewing and labels render generations as submissions—not displayed frames (`contact_cycle_sim/performance_v042.py:40–67`).
## Residual risks
- Dominant cost remains unknown until baseline profiling completes; physics, rendering, CPU recording or synchronization may dominate.
- Packed copies can reduce transfer fan-out yet lose time to packing/allocation.
- CPU equivalence tests cannot establish native CUDA/PhysX trace equality.
- Command caching is particularly sensitive to endpoint/next-reference clocks and live pair-command changes; defer it initially.
- No recommendation changes the floating-root fixture, physical parameters, limits, controller math, 500 Hz trace, gates, render settings, GPU ownership or protected historical artifacts.
@@ -0,0 +1,29 @@
#!/usr/bin/python3.12
import pathlib,json,hashlib,shutil
A=pathlib.Path(__file__).resolve().parent;D=A.parent;P=pathlib.Path(json.loads((A/'assembly.json').read_text())['package']);sha=lambda p:hashlib.sha256(p.read_bytes()).hexdigest()
I={x['package']:x for x in json.loads((P/'evidence-inventory.json').read_text())}
def copy(src,dst):
dst.parent.mkdir(parents=True,exist_ok=True);shutil.copy2(src,dst);I[str(dst.relative_to(P))]={'original':str(src),'package':str(dst.relative_to(P)),'sha256':sha(src),'bytes':src.stat().st_size}
for src in A.iterdir():
if src.is_file() and src.suffix in {'.json','.md','.log','.xml','.py','.txt'}:copy(src,P/'evidence/v042/parent'/src.name)
for name in ['README.md','current_status.md','next_stage.md','evidence_index.md','contact_cycle_sim_performance_v042_result.md']:
shutil.copy2(D/'docs'/name,P/'docs'/name)
shutil.copy2(D/'docs/contact_cycle_sim_performance_v042_result.md',P/'RESULTS.md')
(P/'evidence-inventory.json').write_text(json.dumps(list(I.values()),indent=2)+'\n')
# Remove only this generated package's test/bytecode caches, never user/shared
# caches or any source/evidence. First preseal cache assertion failure is retained.
shutil.rmtree(P/'.pytest_cache',ignore_errors=True)
caches=list(P.rglob('__pycache__'))
if caches:
receipt={'preseal_observation':'First refresh asserted because generated bytecode cache remained after CPU verification. No runtime source mismatch; no archive had been sealed.',
'removed_only_package_generated_dirs':[str(p.relative_to(P)) for p in caches]}
(A/'delivery-cache-cleanup.json').write_text(json.dumps(receipt,indent=2)+'\n')
for p in caches:
assert not p.is_symlink();shutil.rmtree(p)
assert not list(P.rglob('__pycache__'))
a=json.loads((A/'runs/display-S1-05/campaign/approval.json').read_text())['sources'];assert all(sha(P/n)==h for n,h in a.items())
if (A/'delivery-cache-cleanup.json').exists():
copy(A/'delivery-cache-cleanup.json',P/'evidence/v042/parent/delivery-cache-cleanup.json')
(P/'evidence-inventory.json').write_text(json.dumps(list(I.values()),indent=2)+'\n')
assert all(sha(P/x['package'])==x['sha256'] for x in I.values())
print('evidence inventory verified',len(I),'authority41 unchanged')
+60
View File
@@ -0,0 +1,60 @@
#!/usr/bin/env python3
"""New finite public-API run + read-only GPU/X11 observer. No device tuning."""
import argparse,hashlib,importlib.util,json,os,pathlib,signal,subprocess,sys,time,traceback
A=pathlib.Path(__file__).resolve().parent;ROOT=A.parent;P=pathlib.Path(json.loads((A/'staging.json').read_text())['root'])
def read(p):
try:return json.loads(p.read_text())
except FileNotFoundError:return {}
def main():
ap=argparse.ArgumentParser();ap.add_argument('--name',required=True);ap.add_argument('--sequence');ap.add_argument('--cycles',type=int,default=1);ap.add_argument('--headless',action='store_true');ap.add_argument('--video-seconds',type=int,default=0);args=ap.parse_args();assert 0<=args.video_seconds<=240
scope=read(A/'scope.json');cases=list((A/'runs').glob('*/campaign/campaign-result.json'));used=sum(len(read(p)['rows']) for p in cases);reserved=sum(read(p)['reserved_physics_calls'] for p in cases);assert used+2<=scope['limits']['kit_launches'];assert reserved+(27250 if args.sequence else args.cycles*4500+250)<=scope['limits']['physics_calls']
O=A/'runs'/args.name;O.mkdir(parents=True);E=O/'observer';E.mkdir();cli=P/'scripts/go2_fl_skill.py';reg=O/'session/registration.json';out=O/'campaign'
env=os.environ.copy();env.update(PYTHONPATH='',PYTHONDONTWRITEBYTECODE='1',DISPLAY=':1',XAUTHORITY='/run/user/1000/gdm/Xauthority',XDG_RUNTIME_DIR='/run/user/1000')
commands=[['/usr/bin/python3.12',str(cli),'authorize','--registration',str(reg)],['/usr/bin/python3.12',str(cli),'run','--registration',str(reg),'--output',str(out),'--controller','analytic']]
commands[1]+=(['--sequence',args.sequence] if args.sequence else ['--command','x_bias-16mm','--cycles',str(args.cycles)])
if not args.headless:commands[1].append('--gui')
helper=pathlib.Path('/home/cen/sf-runs/simskill-gui-v041-authority-01/desktop_trial.py');spec=importlib.util.spec_from_file_location('external_x11',helper);x=importlib.util.module_from_spec(spec);spec.loader.exec_module(x);x.P=P
r={'commands':commands,'began_wall':time.time(),'prior_kit_launches':used,'scope_limit':scope['limits'],'source_observer_sha256':hashlib.sha256(pathlib.Path(__file__).read_bytes()).hexdigest()};closed=False;child=None;gpu=None;video=None;w=None;view=None;probed=False;shot=False
try:
with (E/'authorize.log').open('w') as f:subprocess.run(commands[0],cwd=P,env=env,stdout=f,stderr=subprocess.STDOUT,check=True)
gpu=subprocess.Popen(['nvidia-smi','--id=0','--query-gpu=timestamp,utilization.gpu,utilization.memory,memory.used,power.draw,power.limit,clocks.current.sm,clocks.current.memory,pstate,temperature.gpu','--format=csv','-lms','500'],stdout=(E/'gpu.csv').open('w'),stderr=(E/'gpu.stderr.log').open('w'))
child=subprocess.Popen(commands[1],cwd=P,env=env,stdout=(E/'public-run.log').open('w'),stderr=subprocess.STDOUT)
deadline=time.monotonic()+1400
while child.poll() is None and time.monotonic()<deadline:
prep=out/'prepare/worker-result.json'
if not probed and read(prep).get('kit_runtime_metadata'):
subprocess.run(['/usr/bin/python3.12',str(ROOT/'scripts/probe_setup.py'),'--kit-metadata',str(prep)],env=env,stdout=(E/'runtime-header.json').open('w'),check=True);probed=True
if not args.headless:
current=x.window()
if current is not None:w=current
status=read(out/'mailbox/status.json');g=read(out/'run/gui-status.json')
if w is not None and g.get('last_tick',0)>=500 and args.video_seconds and video is None:
vcmd=[str(x.FF),'-hide_banner','-y','-f','x11grab','-framerate','30','-video_size',f"{w['width']//2*2}x{w['height']//2*2}",'-i',f"{env['DISPLAY']}+{w['x']},{w['y']}",'-an','-c:v','libx264','-preset','ultrafast','-crf','24','-pix_fmt','yuv420p','-t',str(args.video_seconds),str(E/'live-kit-window.mp4')]
video=subprocess.Popen(vcmd,stdout=(E/'screen-video.log').open('w'),stderr=subprocess.STDOUT);r['video_command']=vcmd;r['video_started_wall']=time.time();r['video_start_gui']=g;r['video_scope']='Owned live X11 window;30Hz capture sampling NOT a claim of render or display FPS; observer overhead included in active loop measurement'
if w is not None and g.get('last_tick',0)>2500 and not shot:
x.screenshot(w,E/'actual-moving-window.png');r['moving_capture']={'window':w,'gui':g,'wall':time.time()};shot=True
if w is not None and status.get('state')=='viewing_final' and not closed:
if view is None:view=time.monotonic();r['final_counts_before']=read(out/'run/physics-counts.json')
elif time.monotonic()-view>=2:
x.screenshot(w,E/'actual-final-window.png');r['final_counts_after']=read(out/'run/physics-counts.json');x.wm_close(w);closed=True;r['wm_delete_wall']=time.time()
time.sleep(.2)
if child.poll() is None:
if w is not None and not closed:x.wm_close(w);closed=True
raise TimeoutError('Observer wall deadline; no physical retry')
r['waited_exit']=child.wait();r['window_absent_after_wait']=True if args.headless else x.window() is None;r['campaign_success']=read(out/'campaign-result.json').get('success');r['worker_success']=read(out/'run/worker-result.json').get('success')
except BaseException:
r['observer_error']=traceback.format_exc()
if child is not None and child.poll() is None:
if w is not None and not closed:x.wm_close(w)
try:r['waited_exit']=child.wait(timeout=150)
except subprocess.TimeoutExpired:r['cleanup_wait_timeout']=True
finally:
if video is not None:
if video.poll() is None:video.send_signal(signal.SIGINT)
try:r['video_waited_exit']=video.wait(timeout=30)
except subprocess.TimeoutExpired:video.terminate();r['video_waited_exit']=video.wait(timeout=10)
if gpu is not None:
gpu.terminate();r['gpu_observer_exit']=gpu.wait(timeout=10)
r['ended_wall']=time.time();(E/'result.json').write_text(json.dumps(r,indent=2)+'\n')
print(json.dumps(r,indent=2))
if __name__=='__main__':main()
+13
View File
@@ -0,0 +1,13 @@
{
"start": "2026-10-16T13:17:31.933004+08:00",
"limits": {
"kit_launches": 12,
"physics_calls": 120000,
"owned_bytes": 25769803776,
"seconds": 21600
},
"training": false,
"gpu": "NVIDIA GeForce RTX 5080 / cuda:0",
"user_output_path": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory",
"plan_sha256": "1d59f67d4fc56896f63aafd8fd345ec9ebccd19bce207543da8ba90783e87c4d"
}
+31
View File
@@ -0,0 +1,31 @@
#!/usr/bin/python3.12
"""Seal and independently read back every archive payload; no native/GPU run."""
import pathlib,json,hashlib,tarfile,xml.etree.ElementTree as ET
A=pathlib.Path(__file__).resolve().parent;P=pathlib.Path(json.loads((A/'assembly.json').read_text())['package']);archive=P.with_suffix('.tar.gz')
# with_suffix would truncate the dotted version; use the whole directory name.
archive=P.parent/(P.name+'.tar.gz');assert not archive.exists();assert not (P/'SHA256SUMS').exists()
sha=lambda p:hashlib.sha256(p.read_bytes()).hexdigest()
auth=json.loads((A/'runs/display-S1-05/campaign/approval.json').read_text())['sources'];assert len(auth)==41 and all(sha(P/n)==h for n,h in auth.items())
xml=ET.parse(A/'delivery-cpu-final.xml');s=next(xml.iter('testsuite'));assert int(s.attrib['failures'])==0 and int(s.attrib['errors'])==0 and int(s.attrib['skipped'])==6 and int(s.attrib['tests'])==1563
validation=json.loads((A/'delivery-validation-final.json').read_text());assert validation['execution_authorized'] is False and validation['new_physics_calls']==0 and validation['sources']==auth
assert validation['frozen']['snapshot_files_verified']==134 and validation['frozen']['candidate_sha256']=='c088bb3aecaca751867118858626fd508f7a3983303802d7524353f80802cf99'
files=sorted(p for p in P.rglob('*') if p.is_file());assert not any(p.is_symlink() for p in P.rglob('*'))
manifest={str(p.relative_to(P)):sha(p) for p in files}
(P/'SHA256SUMS').write_text(''.join(h+' '+n+'\n' for n,h in manifest.items()))
manifest['SHA256SUMS']=sha(P/'SHA256SUMS')
with tarfile.open(archive,'w:gz',compresslevel=6) as t:t.add(P,arcname=P.name)
readback={}
with tarfile.open(archive,'r:gz') as t:
for m in t:
path=pathlib.PurePosixPath(m.name)
assert not path.is_absolute() and '..' not in path.parts and path.parts[0]==P.name
assert m.isdir() or m.isfile(),m.name
if m.isfile():
name=str(path.relative_to(P.name));assert name not in readback
h=hashlib.sha256();f=t.extractfile(m)
while chunk:=f.read(1024*1024):h.update(chunk)
readback[name]=h.hexdigest()
assert readback==manifest,'Archive payload differs from sealed package'
h=sha(archive);side=pathlib.Path(str(archive)+'.sha256');side.write_text(h+' '+archive.name+'\n')
r={'version':'0.4.2','accepted_skill':False,'package':str(P),'archive':str(archive),'archive_bytes':archive.stat().st_size,'archive_sha256':h,'checksum_file':str(side),'payload_files_including_manifest':len(manifest),'sha256sums_entries':len(manifest)-1,'authority_inputs_match_native_staging':41,'all_manifest_files_match':True,'every_archive_payload_readback_verified':True,'no_illegal_members':True,'native_run_at_delivery_path':False,'native_qualification_scope':'analyticS1GUI/headless27000ticks each; source-identical copied artifact, not fullmatrix/u256/nativefaultinjection','CPU_package':{'passed':1557,'skipped':6,'failed':0},'validation_receipt':str(A/'delivery-validation-final.json')}
(A/'package-verification.json').write_text(json.dumps(r,indent=2)+'\n');print(json.dumps(r,indent=2))
@@ -0,0 +1,21 @@
## Review
Paths below use `T = source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/`.
- **Correct — epoch freshness:** `T/sensor_epoch_v042.py:26–48` caches only successful getters, returns the same live object, and invalidates before and after every delegated update/reset—including zero-dt calls and exceptions.
- **Correct — installation/reset safety:** installation replaces scene dictionary entries without modifying original sensor methods (`sensor_epoch_v042.py:51–66`). Constructor-installed source reset guards remain effective (`T/evaluation_env.py:316–326`). Integration installs after construction and before the initial reset (`scripts/evaluate_cycle_functional_flow.py:223–248`).
- **Correct — fail-before-next-step:** each physics step is followed by scene update, latch/state updates, safety and synchronous evidence; failures return before rendering or another step (`T/evaluation_env.py:457–471`). The proxy introduces no deferred work.
- **Correct — SDK premise:** inspected installed `SensorBase._update_outdated_buffers` and PhysX `ContactSensor._update_buffers_impl`: repeated getters invoke native fetches even when downstream kernels skip false-mask entries. Those kernels preserve already-captured values when masked out.
- **Correct — implementation selection/provenance:** `sensor_epoch` selects the original recorder (`T/performance_v042.py:34–44`). Both new modules are preloaded, source-bound and included in authority inputs (`scripts/evaluate_cycle_functional_flow.py:58–60`, `scripts/skill_launch_v4.py:17–29`, `source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_binding_v4.py:8–16,44–49`).
- **Fixed:** None; read-only review.
No issues found.
**Merge verdict: OK with notes** for the first bounded `sensor_epoch` qualification run—not final performance or native-equivalence acceptance.
### Validation and residual risks
- Inspected `tests/test_contact_cycle_performance_v042.py`; supplied `performance_v042/cpu-second.log` and XML report **44 passed**. No tests, commands, Kit/GPU operations or edits were performed during this review.
- Native trace/gate equivalence, constructor/reset behavior under Kit, and speedup remain unqualified until the optimized run.
- This is not a general transparent sensor replacement: attribute writes are not forwarded, and previously captured source references bypass the proxy. Current inspected consumers do not introduce conflicting sensor mutations. In particular, `LoadedState.sensor` retains the original mount sensor (`tasks/manager_based/loaded_leg/mdp.py:18,70–72`); epoch counters therefore describe wrapper traffic, not all sensor reads.
- Frozen-file hashes and stopped-process/protected-artifact claims were not independently recomputed.
+39
View File
@@ -0,0 +1,39 @@
{
"campaign_original_success": true,
"owned_identities_checked": 11,
"owned_alive": [],
"finite_bound_files_checked": 467,
"binding_file_mismatches": [],
"protected_file_mismatches": [],
"gpu_clients": {
"compute": [],
"graphics": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
]
},
"gpu_review": {
"uuid": "GPU-7d882fc3-f771-6d29-4e58-f0fa2a18a3a5",
"compute_pids": [],
"approved_desktop_graphics_pids": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
],
"task_graphics_pids": [],
"clean": true,
"scope": "owned-compute-and-graphics.plus-pinned-desktop-graphics-only"
},
"static_check_not_same_cfg_proof": true,
"boottime": 358046.608404712,
"passed": true
}
+41
View File
@@ -0,0 +1,41 @@
{
"root": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/staging/go2_fl_contact_cycle_sim-0.4.2-work",
"base": "/home/cen/sf-deliveries/go2_fl_contact_cycle_sim-0.4.1",
"source_files": [
"configs/contact_cycle/functional_flow_v2.json",
"configs/contact_cycle/functional_followup_v3.json",
"configs/contact_cycle/skill_v4_desktop_allowlist.json",
"docs/contact_cycle_sim_gui_v041_plan.md",
"docs/contact_cycle_sim_skill_v4_plan.md",
"scripts/audit_contact_cycle_functional.py",
"scripts/audit_cycle_functional_flow.py",
"scripts/evaluate_cycle_functional_flow.py",
"scripts/evaluate_cycle_sim_residual.py",
"scripts/functional_flow_launch_v2.py",
"scripts/functional_followup_launch_v3.py",
"scripts/functional_gpu_watch_v3.py",
"scripts/go2_fl_skill.py",
"scripts/rootxy_launch_v2.py",
"scripts/skill_launch_v4.py",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/evaluation_binding.py",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/evaluation_budget.py",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_flow_binding_v2.py",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_flow_v2.py",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_followup_binding_v3.py",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/functional_followup_v3.py",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_binding_v4.py",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_mailbox_v4.py",
"source/SkillFactory/SkillFactory/legs/contact_cycle_sim/skill_v4.py",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle/recording.py",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_cfg.py",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_env.py",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/evaluation_mdp.py",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/functional_flow_recording_v2.py",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/recording.py",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/skill_gui_v041.py",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/skill_recording_v4.py",
"source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/performance_v042.py",
"configs/contact_cycle/performance_v042.json",
"performance_v042/plan.md"
]
}
+22
View File
@@ -0,0 +1,22 @@
#!/usr/bin/python3.12
"""Sync only with all prior campaign observers terminal; preserve approved bytes."""
import pathlib,json,hashlib,shutil
A=pathlib.Path(__file__).resolve().parent;D=A.parent;P=pathlib.Path(json.loads((A/'staging.json').read_text())['root'])
for root in (A/'runs').iterdir():
assert (root/'campaign/campaign-result.json').exists() and (root/'observer/result.json').exists(),str(root)
approval=json.loads((root/'campaign/approval.json').read_text())
snapshot=A/'approved-sources'/root.name
for n,h in approval['sources'].items():
dest=snapshot/n
if dest.exists():assert hashlib.sha256(dest.read_bytes()).hexdigest()==h;continue
candidates=[P/n,A/'failed-baseline01-approved-source'/n,A/'baseline02-approved-source'/n]
source=next((p for p in candidates if p.exists() and hashlib.sha256(p.read_bytes()).hexdigest()==h),None)
assert source is not None,(root,n)
dest.parent.mkdir(parents=True,exist_ok=True);shutil.copy2(source,dest)
# Older sources already archived in named preserved paths may not match latest staging.
files=json.loads((A/'staging.json').read_text())['source_files']
files+=['source/SkillFactory/SkillFactory/tasks/manager_based/contact_cycle_sim/'+name+'.py' for name in ('sensor_epoch_v042','packed_recording_v042','readonly_scope_v042','batched_kinematics_v042','live_cadence_v042')]
files+=['performance_v042/presentation_amendment.md']
for n in dict.fromkeys(files):
(P/n).parent.mkdir(parents=True,exist_ok=True);shutil.copy2(D/n,P/n)
print('synced',len(set(files)),'bound source inputs')
+45
View File
@@ -0,0 +1,45 @@
{
"completed": "2026-10-16T15:38:11.211717+08:00",
"task": "bounded application performance + analyticS1 result sealed",
"accepted_skill": false,
"training_steps": 0,
"Kit_instances": 11,
"reserved_physics_calls": 73500,
"actual_physics_calls": 67500,
"elapsed_seconds": 8439.208175182343,
"accounted_owned_bytes_including_reserve": 1454357029,
"scope_limits": {
"kit_launches": 12,
"physics_calls": 120000,
"owned_bytes": 25769803776,
"seconds": 21600
},
"protected_files_unchanged": 4427,
"owned_identities_checked": 57,
"owned_alive": [],
"strict_final_GPU_clients": {
"compute": [],
"graphics": [
2420,
2618,
3268,
4486,
574096,
602685,
696142
]
},
"package": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2",
"archive": "/home/cen/Embodied_Workspace/Isaac_Projects/SkillFactory/performance_v042/deliveries/go2_fl_contact_cycle_sim-0.4.2.tar.gz",
"archive_sha256": "16c855f6086a90aedc3f50eb510d44f64f802254505146a61c83402df7d99151",
"archive_bytes": 96636332,
"package_verification_sha256": "3f80d0b0108061098bf5908558b9b00b22aff83e1919249c42d00d9d1699b77c",
"final_inventory_sha256": "4e9fb49c5befa6af4027a78de93f8912d43a5360e808110a94daf96f72c4f8d5",
"GUI_render_submissions_per_second": 35.30043787880234,
"GUI_window_overlay_observed_FPS": 33.98,
"GUI_sim_time_ratio": 0.07060087575760468,
"headless_sim_time_ratio": 0.18325004295397335,
"qualification": "current-source analyticS1GUI/headless6cycles each;70tensorbyte-identical to0.4.1GUI11; not1x/fullGPU/fullmatrix/u256/nativefaultinjection/globaldeterminism",
"unresolved_scope_limits": "README / RESULTS; firstindependentheadlesspostexitGPUadmissionexceptioncauseunknownpreserved; laterstrictchecknotretroactiveexplanation; firstpackagingCPUfixturefailurepreservedandpayloadrepaired",
"no_more_automatic_native_work": true
}
+185
View File
@@ -0,0 +1,185 @@
#!/usr/bin/env python3
"""CPU-only function-first trace audit. Never starts Kit or grants execution.
Checks one or three *globally clocked* cycles; never joins separate episodes.
A tensor trace cannot prove the absence of simulator state writes. Therefore
trace_functional_passed is NOT a runtime-continuity or new simulation claim.
"""
import argparse
import hashlib
import importlib
import importlib.util
import json
import sys
from pathlib import Path
import torch
from audit_contact_cycle_functional import evaluate as evaluate_cycle
ROOT = Path(__file__).resolve().parents[1]
PHASES = ('support', 'unload', 'lift', 'swing', 'landing', 'reload')
FAILURES = ('nonfinite', 'contact_acquisition_timeout', 'premature_separation', 'lift_deadline',
'swing_clearance', 'reload_deadline', 'low_height', 'tilt')
DEFAULT_CONFIG = ROOT / 'configs/contact_cycle/functional_flow_v2.json'
def cpu_protocol():
"""Private CPU namespace only; NEVER a runtime/binding import alias."""
name = '_functional_flow_cpu'
directory = ROOT / 'source/SkillFactory/SkillFactory/legs/contact_cycle_sim'
if name not in sys.modules:
spec = importlib.util.spec_from_file_location(name, directory / '__init__.py',
submodule_search_locations=[str(directory)])
module = importlib.util.module_from_spec(spec)
sys.modules[name] = module
spec.loader.exec_module(module)
return importlib.import_module(name + '.functional_flow_v2')
def root_error(trace):
for key in ('root_position_error_m', 'root_position_error'):
if key in trace:
return trace[key]
if 'policy_snapshot' in trace and trace['policy_snapshot'].shape[-1] == 46:
return trace['policy_snapshot'][..., 32:35]
return None
def phase_summary(trace):
result = {}
error = root_error(trace)
for phase, name in enumerate(PHASES):
mask = trace['phase'][:, 0] == phase
if not mask.any():
result[name] = {'samples': 0, 'status': 'not_evaluated'}
continue
fn = trace['normal'][mask, 0, 2]
gap = trace['gap'][mask, 0]
foot = trace['foot'][mask, 0]
result[name] = {
'samples': int(mask.sum()),
'normal_N_min_mean_max': [float(fn.min()), float(fn.mean()), float(fn.max())],
'mean_load_W': float((fn / trace['weight'][mask, 0]).mean()),
'gap_m_min_max': [float(gap.min()), float(gap.max())],
'foot_net_xyz_m': (foot[-1] - foot[0]).tolist(),
'root_xy_error_max_m': None if error is None else float(error[mask, 0, :2].norm(dim=-1).max()),
}
return result
def evaluate_flow(trace, cfg, controller_kind, cycles):
api = cpu_protocol()
api.validate_contract(cfg)
protocol = api.FunctionalFlowProtocol(cfg, cycles=cycles, controller_kind=controller_kind)
out = {'format': cfg['format'], 'controller_kind': controller_kind, 'requested_cycles': cycles,
'trace_functional_passed': False, 'runtime_continuity_verified': False,
'accepted_skill': False, 'new_simulation_steps': 0, 'new_optimizer_steps': 0,
'gates': {}, 'cycles': [],
'limitation': 'Offline tensor evidence only. Full runtime pass additionally requires fresh '
'no-autoreset hooks, seam carry-state, budget and clean terminal evidence.'}
if 'time' not in trace or not trace['time'].numel():
out['error'] = 'No physics samples'
return out
required = {'time', 'phase', 'scenario', 'weight', 'normal', 'foot', 'target', 'gap',
'signed_height_error', 'tilt', 'external', 'guide', 'auxiliary', 'joint_excess',
'limit_ratio', 'contact_acquired', 'actor_metadata_available', 'control_action_available'}
if not required <= trace.keys():
out['missing_fields'] = sorted(required - trace.keys())
return out
n = len(trace['time'])
if any(not isinstance(v, torch.Tensor) or v.shape[:2] != (n, 1) for v in trace.values()):
out['error'] = 'Require single-trial [T,1,...] tensors'
return out
if not all(bool(torch.isfinite(v).all()) for v in trace.values()):
out['error'] = 'Nonfinite raw evidence; no passing prefix substituted'
return out
t = trace['time'][:, 0]
expected = torch.arange(1, cycles * 4500 + 1, dtype=t.dtype) * cfg['physics_dt_s']
out['gates']['global_clock_without_restart'] = n == len(expected) and bool(
torch.allclose(t.cpu(), expected, atol=1e-5, rtol=0))
out['observed_end_s'] = float(t[-1])
runtime_fields = ('failure_flags', 'would_fail', 'acquisition_failed', 'premature_separation')
out['runtime_failure_fields_missing'] = [k for k in runtime_fields if k not in trace]
out['gates']['no_recorded_runtime_failure'] = all(
not bool(trace[k].any()) for k in runtime_fields if k in trace)
out['first_recorded_failure'] = None
if 'failure_flags' in trace:
flags = trace['failure_flags'][:, 0]
if flags.shape != (n, len(FAILURES)):
out['error'] = 'Invalid failure flag schema'
return out
hits = flags.any(-1).nonzero().flatten()
if len(hits):
i = int(hits[0])
out['first_recorded_failure'] = {'time_s': float(t[i]), 'phase': PHASES[int(trace['phase'][i, 0])],
'causes': [k for k, hit in zip(FAILURES, flags[i]) if hit]}
out['controller_kind_is_metadata_only'] = True
out['recorded_max_effective_residual_rad'] = (None if 'effective_residual_rad' not in trace else
float(trace['effective_residual_rad'].abs().max()))
if controller_kind == 'analytic_reference':
out['gates']['no_nonzero_recorded_reference_residual'] = all(
not bool(trace[k].abs().gt(1e-6).any()) for k in
('requested_action', 'effective_residual_rad', 'residual_request_normalized') if k in trace)
if controller_kind == 'learned_policy' and {'requested_action', 'actor_mean'} <= trace.keys():
out['gates']['recorded_deterministic_mean_action'] = bool(torch.allclose(
trace['requested_action'], trace['actor_mean'].tanh(), atol=1e-6, rtol=0))
error = root_error(trace)
out['root_xy_error_max_m'] = None if error is None else float(error[..., :2].norm(dim=-1).max())
out['root_xy_30mm_exceeded_diagnostic_only'] = None if error is None else bool(
(error[..., :2].norm(dim=-1) > .03).any())
# No resampling, truncation-to-success, reanchoring or phase relabeling.
# Only after the GLOBAL clock is verified, derive local time from integer
# ticks: float32 (18 + 4.6) - 18 can otherwise misclassify an exact boundary.
legacy = {'format': 'skillfactory.contact-cycle-simulation-functional.v1',
'physics_dt_s': cfg['physics_dt_s'], 'phase_durations_s': cfg['phase_durations_s'],
'nominal_scenario_id': 0, 'gates': cfg['functional_gates']}
for c in range(cycles):
start, end = c * 4500, min((c + 1) * 4500, n)
if start >= n:
out['cycles'].append({'cycle_index': c, 'status': 'not_evaluated'})
continue
local = {k: v[start:end].clone() for k, v in trace.items()}
local['time'] -= c * 9.0
if out['gates']['global_clock_without_restart']:
local['time'] = (torch.arange(1, end - start + 1, dtype=t.dtype,
device=t.device) * cfg['physics_dt_s'])[:, None]
audit = evaluate_cycle(local, legacy, controller_kind)
out['cycles'].append({'cycle_index': c, 'audit': audit, 'phases': phase_summary(local)})
out['trace_functional_passed'] = bool(all(out['gates'].values()) and all(
row.get('audit', {}).get('sim_functional_passed', False) for row in out['cycles']))
return out
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('--trace', type=Path, required=True)
parser.add_argument('--config', type=Path, default=DEFAULT_CONFIG)
parser.add_argument('--controller-kind', choices=('analytic_reference', 'learned_policy'), required=True)
parser.add_argument('--cycles', type=int, choices=(1, 3), required=True)
parser.add_argument('--output', type=Path, required=True)
args = parser.parse_args()
if args.output.exists():
raise FileExistsError('Never overwrite an earlier audit')
cfg = cpu_protocol().load_contract(args.config)
raw = torch.load(args.trace, map_location='cpu', weights_only=True)
traces = ({str(e): {k: v[:, e:e+1] for k, v in raw.items()}
for e in range(raw['time'].shape[1])} if 'time' in raw else raw)
result = {'config_sha256': hashlib.sha256(args.config.read_bytes()).hexdigest(),
'audit_sha256': hashlib.sha256(Path(__file__).read_bytes()).hexdigest(),
'trace': str(args.trace.resolve()),
'trace_sha256': hashlib.sha256(args.trace.read_bytes()).hexdigest(),
'rows': {e: evaluate_flow(d, cfg, args.controller_kind, args.cycles) for e, d in traces.items()},
'new_simulation_steps': 0, 'new_optimizer_steps': 0, 'runtime_continuity_verified': False}
args.output.parent.mkdir(parents=True, exist_ok=True)
with args.output.open('x') as stream:
json.dump(result, stream, indent=2, allow_nan=False)
stream.write('\n')
print(json.dumps({'output': str(args.output), 'rows': len(result['rows']),
'trace_functional_passed': sum(r['trace_functional_passed'] for r in result['rows'].values()),
'runtime_continuity_verified': False}))
return 0 if all(r['trace_functional_passed'] for r in result['rows'].values()) else 2
if __name__ == '__main__':
raise SystemExit(main())

Some files were not shown because too many files have changed in this diff Show More