Triage the 50 unverified findings from the 2026-09-03 contract audit

The contract audit of 2026-09-03 raised 76 findings across eleven lenses. Twenty six were verified adversarially, of which 17 were confirmed and are filed as work items. The remaining 50 were never verified, because the cap was set at 26 and the findings were sorted by severity first.

Every finding rated a bug was verified. The 50 unverified are 40 rated a gap and 10 rated a nit, so nothing here is expected to be serious. They are recorded so the work is not silently lost, and they need triage rather than action.

The themes, from a read of the claims: over-acceptance on open roots (a block-level tone on stats that the prose forbids), unpinned header-row position above the size threshold each schema enumerates to, limits the schemas impose that the contract never states (a 60-character cap on bar-list column names), half-enforced rules (a fallback cell holding a single empty text node passes the "no cell is ever blank" rule), and the same data divergence recorded separately as its own work item.

Triage would run the same adversarial verification over the 50 and file whatever survives. Expect a lower confirmation rate than the 65% the bug-severity set returned.

Start by reading the claims across schema/, CONTRACT.md, scripts/, and samples/, then repeat the adversarial verification described in the issue for the 50 unverified findings. Track which findings survive and file the confirmed results as work items; completion means all 50 have been triaged without silently dropping any.