Engineering Evidence Toolkit

Small, inspectable building blocks for turning engineering and simulation outputs into source-backed, reproducible evidence. The v0.2 contract accepts the v0.1 example while validating lifecycle precedence, hexadecimal SHA-256 digests, unique IDs, resolvable claims, and typed frame/time markers. The toolkit is designed for human-reviewed workflows: it records scope, provenance, solver state, units, coordinates, frames, and claim readiness without pretending that a green solver exit code proves physical correctness.

This is a clean-room, Apache-2.0 repository. The examples are synthetic and contain no ODB, CAE, manuscript, credential, or private project data.

What is included

Skill	Purpose
evidence-contract	Define the smallest testable contract before collecting results.
provenance-ledger	Bind every input and observation to an identifier, digest, unit system, and transform history.
solver-status-gate	Separate datacheck, solver completion, physical review, and publication readiness.
result-audit	Audit signs, units, coordinates, frames, regions, baselines, and simple bounds.
claim-readiness-audit	Map each prose claim to evidence and expose blocked or conditional links.
reproducible-reporting	Assemble a deterministic manifest, environment record, command log, and handoff report.

The standard-library checker at scripts/evidence_check.py validates a compact JSON contract. It is a structural gate, not a replacement for an Abaqus run, an ODB review, or an engineering sign-off.

Quick start

python -B scripts/evidence_check.py examples/synthetic/contract.json
python -B -m unittest discover -s tests -v

The synthetic contract should print PASS. Changing solver_status.physics_review to not_started demonstrates the CONDITIONAL exit code (2). Missing source records or unbound evidence returns BLOCKED.

Contract shape

Every contract records:

1. an artifact_id and explicit scope (model, step, region, and other boundaries);
2. non-empty inputs with source_id, SHA-256, and units;
3. separate solver_status values for datacheck, analysis, and physics_review;
4. field/history observations with source IDs, field names, units, coordinate system, and valid non-negative frame or finite non-negative time; static-audit/document rows are exempt from frame/time;
5. claims whose evidence_ids resolve to observations.

The contract intentionally keeps the final decision visible. A failed status in any of datacheck, analysis, or physics_review blocks the contract; otherwise any pending/required/not-run state is conditional; only complete/passed/reviewed states across all three can pass. A calculation can be complete while a claim remains conditional or blocked until physical review, baseline comparison, or source reconciliation is complete.

Abaqus report adapter

The deterministic from_abaqus_audit.py adapter accepts the companion Abaqus report shape (input_digest plus findings) and emits a static-only 0.2 contract. It can also hash a model input file with --model-input:

python -B scripts/from_abaqus_audit.py examples/cross-repo/abaqus-agent-report.json contract.json
python -B scripts/evidence_check.py contract.json

The public cross-repository example is synthetic. Repeating the conversion produces byte-identical output and the checker returns CONDITIONAL because no solver or physical-review gate is being asserted.

Documentation

• Architecture
• Quickstart and extension guide
• Claim lifecycle
• Contributing
• Security and data boundary
• Roadmap
• Citation metadata

Relationship to Abaqus skills

The companion abaqus-agent-skills repository focuses on model construction and review. This repository is deliberately solver-agnostic so that the same evidence contract can wrap Abaqus, other finite-element solvers, laboratory measurements, or analytical baselines.

License

Apache-2.0. See LICENSE and NOTICE.