E03: bill of materials export (bom-export) as CSV - #111

Verdict: approved

E03 bill-of-materials promotion, reviewed as the six-commit delta on top of the A09 branch it stacks on (the shared A09 tail is reviewed in #110).

• crates/engine/src/bom.rs is pure and owns the report rules: capped row count refused before rendering, deterministic sort (kind, profile, mark), grouping with mass_unknown counted and unknown mass excluded from totals and never rendered as zero, RFC-4180 quoting applied only where needed. All pinned by tests, including the CSV header and the empty-cell contract.
• The CLI adapter (apps/beamcraft-cli/src/bom.rs) derives area as volume/thickness for plates and volume/length for members with explicit denominator guards, resolves profile/material labels from the pinned definitions (designation, then display name, then the digest as the visible last resort), and takes the mark from the A09 marks of the same stack with the element-id fallback the issue requires.
• cdf95ae reserves both output files with create_new before any bytes are written and removes the earlier reservations when a later reserve fails; the refusal test covers the taken-summary path. One benign gap, optional to address: if the first write itself fails after both files are reserved, the later empty file is left behind — a retry then refuses with "exists", visible rather than destructive, and the real fix (write-to-temp-then-install) is what the store already does for blobs.
• The dedup commit (34bba58) leaves the module registrations and dispatch clean for the upstream base, and the design note names the adapter module correctly (784d6b4).

No CI checks have reported on the branch yet; the approval rests on the review above and on this code having run in the fork since its E03 slice.

Verdict: approved

E03 rebased from the stacked fork-A09 branch onto the natively merged A09 (#103); the delta against the version I approved is exactly the adaptation, and it is right:

• The mark source moved from the fork marks API to the native one: the adapter takes &[StoredMark] from numbering_approval::head_marks, renders through the reader rule (mark.text(kind) with the proper GroupKind), and falls back to the element id for unnumbered elements — same contract as before, now on the upstream type.
• head_marks was widened from pub(crate) to pub with a doc comment naming the BOM/IFC readers as the reason; no logic in the native A09 adapter changed.
• The command cross-checks revisions: the marks read is rejected unless mark_revision equals the inspected revision id, so the report cannot mix a scene from one head with marks of another.
• The engine tables, CSV rendering, reserve-both-files-then-write flow and the mass-unknown rules are unchanged from what I approved; no schema descriptor or pin changes, which is correct for a slice that adds no attributes.
• The test suite grew with the native-marks cases (415 lines vs 384).

No CI checks have reported on the branch yet; the approval rests on the diff review and on the earlier approved version of the same code.

Verdict: approved

Re-reviewed the new head against the previously approved one: the entire delta is a rustfmt line-wrap inside one test assertion (groups[0][3] in apps/beamcraft-cli/tests/bom.rs); the adapter, command, engine tables, marks integration through head_marks with the revision cross-check, and the reserve-both-files flow are byte-identical to the approved version. Nothing to change.

Verdict: approved

Re-re base review: the branch is rebased onto the master that now carries the merged #108 and #109, and the E03 content is byte-identical to the previously approved version (b8d0c62) — the adapter, the command with the marks revision cross-check, the engine tables, tests and design note all diff clean.

The only changes in the delta are the base-wide registration lines the merged A05 brings in: the bolt module in lib.rs, the bolt_cmd module and the help-literal additions (create-bolt | bolts after the welds segment) in main.rs — consistent with how the branch already dispatched its own command, no ordering or surface surprises.

The marks source remains the native A09 head_marks with the revision-id cross-check; no pin or descriptor changes, correct for a slice that adds no schema lines.