cadSet/designir-pipeline/README.md

DesignIR Pipeline

This directory owns the STEP reconstruction and distillation boundary:

teacher STEP
  -> deterministic private evidence extraction
  -> reconstruction agent authors DesignIR 2.0
  -> isolated compiler rebuilds STEP without teacher access
  -> acceptance agent compares teacher and rebuilt STEP
  -> deterministic publisher promotes repeated validated experience

Security boundary

Component	Teacher STEP	Private evidence	DesignIR	Rebuilt STEP
Ingestion service	read	write	no	no
Reconstruction agent	denied	read	write	request compilation
Isolated compiler	denied	no	read	write
Acceptance agent	read	read	read	read
Publisher	denied	aggregate only	schema only	reports only

The reconstruction environment must remain functional after the teacher STEP is removed. scripts/designir_pipeline.py anti-cheat rejects source STEP references, imported B-Reps, embedded meshes, topology references, and generator source that calls STEP import APIs.

Data layout

The operator-facing layout has only three directories:

designir-pipeline/
  input/                   drop source STEP/STP files here
  runs/                    all private process artifacts
  output/library.json      verified experience consumed at runtime

input/ and runs/ are ignored by Git. Only the deterministic result in output/ is versioned. The pipeline creates these private run directories as needed:

Local path	Purpose
runs/evidence/	Deterministic evidence extracted from source STEP.
runs/designir/	DesignIR authored by Reconstruction Agent.
runs/rebuilt/	STEP rebuilt without access to source STEP.
runs/acceptance/	Independent geometry and editability reports.
runs/review/	Sanitized semantic batches and untrusted drafts.
runs/quarantine/	Failures and unsupported feature vocabulary.

The shallow directory structure is for usability, not authorization. Agent and compiler access is enforced by their contracts and by anti-cheat checks.

DesignIR 2.0

The executable contract is documented by contracts/designir-2.0.schema.json. It describes coordinate systems, datums, parameters, expressions, sketches, constraints, features, patterns, attachments, construction stages, edit interfaces, and validation contracts. It never contains teacher geometry.

The initial compiler deliberately supports a small, honest vocabulary through both Build123d and SimpleCADAPI adapters:

• extrude_circle
• extrude_rectangle
• add_cylinder
• through_hole
• polar_hole_pattern

Unsupported geometry is quarantined with explicit missing capabilities instead of embedding the source B-Rep.

The SimpleCADAPI adapter also emits its replayable *.simplecad.model.json operation graph beside the STEP. DesignIR remains the backend-neutral source of truth.

Agents

• agents/reconstruction-agent.md: Agent A authors DesignIR from private evidence and promoted methods, then requests an isolated rebuild.
• agents/acceptance-agent.md: Agent B independently checks geometry and parameter perturbations against the teacher.

Neither agent publishes experience. config/promotion-policy.json is evaluated by deterministic code after several independent cases pass.

Commands

Drop files into input/, then use the CAD Python environment:

scripts/cad-experience extract-folder
scripts/cad-experience prepare
# The semantic distiller writes runs/review/experience-draft.json.
scripts/cad-experience publish \
  --draft runs/review/experience-draft.json
scripts/cad-experience audit output/library.json

../text-to-cad/.venv/bin/python scripts/designir_pipeline.py validate part.designir.json
../text-to-cad/.venv/bin/python scripts/designir_pipeline.py compile \
  part.designir.json --output rebuilt.step
../text-to-cad/.venv/bin/python scripts/designir_pipeline.py isolated-rebuild \
  part.designir.json --output rebuilt.step --backend simplecadapi
../text-to-cad/.venv/bin/python scripts/designir_pipeline.py accept \
  --teacher teacher.step --designir part.designir.json \
  --rebuilt rebuilt.step --report acceptance.json