06 — Gaps & Risks (Terminal 2) 2026-06-04

Independent challenges to Terminal 1's assumptions. Each is verified against live state.

Mismatches with Terminal 1

1. births +20,677 (live 1,194,727 vs checkpoint 1,174,050). Classification: EXPECTED_BACKGROUND_DRIFT. All engineering collections (approval_requests, axis_assignment, ownership, axis_registry, pivot_definitions, dot_tools, event_type_registry) MATCH exactly. The drift is background generators + KB doc uploads (including Terminal 1's ongoing run and this audit). Not a blocker, but the checkpoint's birth number is now stale — any future "before==after" birth proof must re-baseline to the current counter, not 1,174,050.

Residual / watch items (not blockers)

2. residual v_workflow_residual_evidence_hardening_v4 = 23 rows. Overlaps host_needs_review (23). These are dot/bin reconcile, throttled DB jobs, infra config needing evidence — owner/evidence work, not engineering debt.
3. stale_disabled triggers = 27 + unclassified = 8 in the trigger census. Retirement/triage backlog (D13 runbook). Owner-review, harmless while disabled.
4. 77 host triggers ADAPTER_PENDING. Not yet adapter-managed (crontab49/cron.d6/systemd4/OS18). Census roll-up only; full host adapter is future work.
5. 10 of 12 dot:kg defs contract_missing. Only DOT_KG_EXPLAIN is contract+endpoint ready. The "dot:kg REAL_RUN ladder" applies to ONE def today; the 10-process split is future.

Risks if authority acts without re-verification

• Any REAL_RUN flip must re-confirm invariant_real_run_count_zero=0 and external executor health immediately before flip (4 gates GO is necessary but not sufficient).
• UI rebase over 13-behind public main risks conflicts; operator runbook (doc 04) is mandatory — do not force-push.
• Birth-free proofs must re-baseline (see mismatch #1).

What I could NOT independently verify

• Staged SQL files under contabo:/opt/incomex/docs/mcp-writes/... — not read this run (not required for the audit; engineering numbers verified directly from live views instead). No BLOCKED_READ_CHANNEL was hit on any required check.

Net assessment

No mismatch rises to a blocker. The only correction is that the checkpoint's absolute births figure is stale by background drift; the governance/engineering conclusions are fully intact.