# FAULTLINE v1 — release verification

Recorded September 10, 2026 in America/Chicago (September 11 UTC).
This distinguishes automated execution, browser inspection, and owner acceptance.

| Check | Retained result |
| --- | --- |
| Evaluated Python core | `849ebac937fcdf81c45d1a7b86e5b4be9f358d4d`, clean when executed |
| Backend suite | 16 passed against real PostgreSQL, 4.419 seconds |
| Fixed model evaluation | 30/30 planned attempts retained; 20 verified, 8 abstentions, 2 ineffective restarts |
| Runbook comparison | 15/15 distinct scenarios repaired; both repetitions passed |
| Source branches | All 30 original incidents unchanged; all initial branch hashes match |
| Website suite | 35 passed, including existing routes and the new evidence/replay checks |
| Browser interaction matrix | 108 states: 3 incidents × 3 branches × 2 views × 3 stages × 2 widths |
| Record inspection | Duplicate entries, incompatible input, and quarantine agree with captured data |
| Browser layouts | Chromium at 1440 and 390 pixels; additional 768 and 320 pixel checks |
| Preferences and access | Light/dark modes, keyboard replay, reduced motion, 200% text enlargement checked |
| Fresh ZIP reproduction | New virtual environment and new disposable database; 16 tests passed in 4.225 seconds; all 3 runbook scenarios at seed 29 repaired |

The website suite checks every published file fingerprint and all planned model
attempts. It compares the displayed recordings with their complete raw execution
records and checks the two new routes' server-rendered evidence and metadata.
The browser matrix additionally checks displayed revenue and reconciliation gaps
against the raw JSON across every selected state. No page exceptions were observed.

The first clean-install check exposed a startup race: the PostgreSQL image's
temporary Unix-socket initialization server appeared ready before its final TCP
listener. The helper was corrected to check TCP. The replacement source bundle
was extracted into a new directory and verified with a new virtual environment
and a new database volume. The original failed reproduction log is retained locally.

The evaluated core, fixtures, protocol, dependency pins, and tests are unchanged
by that setup correction. The exported bundle is release revision
`14c9f172199044d3e8612848a81526022159b2be`. The export tool verifies the core
identity before packaging; source and evidence hashes are published separately.

The browser inspection also caught a display mapping issue where duplicate
objects were treated as pairs. The UI now labels all nine duplicate ledger
entries in the selected duplicate-delivery recording. The independent backend
verifier and evaluation totals were unaffected.

## Evidence

- [Backend test output](../backend-tests.txt)
- [Website test output](../website-tests.txt)
- [Browser matrix result](../browser-verification.json)
- [Fresh reproduction test output](../reproduction-tests.txt)
- [Full run manifest](../evidence/manifest.json)
- [Published content hashes](../files.json)

## Scope of acceptance

These are automated local/service tests and Chromium browser checks. Screenshots
were visually inspected in the development environment. They do not imply
physical-phone, Safari, independent security-audit, or owner interview acceptance.
The local model evaluation measures one bounded action decision over known
synthetic fault families. It is not a salary, hiring, or production-SLA claim.

Production publication and GitHub references are read back during the release
handoff and retained in the private owner handoff. The public site contains
synthetic exports and source; it has no connection to the private lab runtime.