TCHG-0006 Record Completeness Floor

Established and enforced a single quality floor across the entire device corpus, then raised every record to it.

• Measured the corpus against its own standard for the first time. The Device Architecture Taxonomy has described five orthogonal axes since TCHG-0004’s predecessor, but no record carried a tag on all five: heat generation was absent from 44 of 45 records and power from 43. Part numbers appeared on 11. Two records cited no primary source at all. Depth ranged 3.9× between the thinnest and richest record.
• Added scripts/audit_record_completeness.py with rules REC-01 to REC-06 and thirteen unit tests, and wired it into bin/validate_graph.sh so the floor is enforced on every build rather than remembered.
• Raised all 44 device records to the floor: every record now carries a tag on each of the five axes, a part-number row, a primary-source citation, a safety section, and a component-role or form-factor row. The audit reports zero errors and zero warnings, down from 81 errors and 51 warnings.
• Closed the archive’s only two evidence holes. TED-0021 (IOLITE Original) and TED-0022 (WISPR 2) previously rested entirely on internal research provenance with no citation a reader could follow; both now cite the manufacturer’s operation manual and the Oglesby & Butler catalytic-vaporiser patent.
• Recorded that an absent part number is stated rather than omitted: “Not published by the manufacturer” satisfies REC-02, because the reader learns the identifier does not exist instead of wondering whether the archive forgot.
• Recorded that internal research provenance is a pointer, not a citation. A record may not rest on the research corpus alone.

The unevenness this corrects was structural rather than anyone’s error: newer records carried part numbers and component roles that older ones lacked, while older records carried delivery-mode and operating-mode tags that newer ones lacked. Nothing measured the difference, so it accumulated silently. The audit now does.