Estimate Pipeline

Estimate Pipeline

steel-estimate coordinates the individual skills without turning review findings into silent assumptions.

estimate input
    ↓
canonical validation
    ↓
plate nesting
    ↓
draft RFQ compilation
    ↓
QA report + lineage + immutable run manifest

Outcomes

Outcome	Meaning
ready	Contract and geometry checks passed for the generated package
needs_review	Artifacts are available, but a person must resolve or acknowledge findings
blocked	A required contract failed; a blocked run does not contain an RFQ workbook

Exit status 0 is geometry-verified, 2 requires review, and 3 is blocked. Automation should read the JSON outcome instead of relying only on console text.

Run isolation

Every invocation publishes into an immutable runs/<run-id>/ directory beneath the requested output root. latest-run.json is a pointer, not the source of truth. Consumers should resolve it and then read that run's manifest and QA report together.

Human acknowledgements

Review decisions are explicit records tied to the input hash and finding ID. They do not erase the original finding. Use skills/steel-estimate/scripts/acknowledge-finding.py --help for the canonical command contract.