Production vulnerability re-ingest closeout — 2026-09-11

VULN-G2 is complete. This closeout does not repeat the already completed production re-ingest or the old database migrations. It refreshes the two gates that remained open: the four-persona Console acceptance and EXP-4 exploit-evidence coverage.

Live Console acceptance

The canonical runner completed as run-210 against the gateway-mounted Console artifact from source ea9da5b34354db6083359dc5e22b9be358805867.

The bounded operational-truth degradation/recovery journey also passed. All four role crawls had zero failures, including the Security Risk Manager advisory-source action that failed in runs 082 and 083. The four representative home screenshots were visually inspected after the machine verdict. The two-worker run completed every scheduled route with no missing or duplicate task and no worker-pool failure.

The served index, main-2JPY2LFG.js, and styles-ILSU26HA.css were byte-identical before and after the traversal and matched the read-only durable Console mount. All four disposable users were disabled with HTTP 200. The evidence scan found zero token/password patterns, and the owner-only .local-secrets/uif8/run-210 directory was removed.

Evidence: docs/qa/feature-checks/runs/2026-09-11-ui-trust-recovery/uif-8-four-role-live/run-210/.

Exploit-evidence coverage

check-exploit-evidence-coverage.sh --self-test passed all nine controls. The live read then passed at KEV generation 115, completed 2026-09-11 07:07:59.804918+00:

• upstream inventory: 1,705;
• distinct raw exploit-evidence documents: 1,705;
• curated exploited_in_wild/kev attributes: 1,705;
• distinct CVEs: 1,705;
• accepted shortfalls: 0.

This exact A/B/C/D parity closes EXP-4’s G2 closeout clause without trusting earlier counts or re-fetching an external catalog.

Capacity result and lessons

The current stellaops_vuln database is 116,703,114,943 bytes (109 GB) with 1,704,142 raw documents. The largest schema totals are ingest 51 GB, facts 38 GB, dist 9,362 MB, eventing 6,423 MB and consensus 3,988 MB. The binary schema is only 7,032 kB, so current size is dominated by retained advisory raw documents, fact provenance/current/revision data and the eventing ledgers.

This is above both the superseded 25–30 GB pre-DC-33 estimate and the 55–60 GB full-run projection recorded at the original G2 completion. Design section 9.1 already marks the old 100 GB provisioning figure unvalidated; the 109 GB measurement confirms that warning. The closeout records the drift rather than calling the current estate within the obsolete estimate. The host disk-floor check passed with 71.2 GiB free before the browser run.

The operational lessons are now concrete:

• generation ordinals can reset, so coverage uses the newest completed full generation and the four stage counts;
• a non-empty final count alone is insufficient because ingestion or normalization can be short;
• Console acceptance must bind the actual gateway-mounted, committed artifact rather than an unrelated local dist directory;
• live role evidence keeps browser/auth state under owner-only .local-secrets, uses the shared StellaOps cryptographic RNG, disables disposable users, and retains only scrubbed evidence.

Sprint 008 remains active for VULN-G6. This G2 closeout does not archive that sprint or claim the remaining source-freeze criterion.