# Round 1 gate audit

Status: historical evidence

Audit date: 21 July 2026

This page records the original gate audit that preceded the corrected public reanalysis. The original evaluator required several implementation details that the briefs did not declare.

## What failed

- Brand universally failed `C02 C03 C05 B01 B02`; those checks depended on hidden dialog, menu and error hooks.
- Operations universally failed `C02 C03 O01 O02 O03`; those checks prescribed hook placement, hidden result/metric/error attributes and an exact activity structure.
- Launch universally failed `C02 C03 C05 L01 L02 L03 L04 L05`; those checks prescribed private state attributes and CSS-3D structure even though the brief allowed CSS, SVG, canvas or WebGL.

This does not mean every run passed. Twenty-one runs also recorded at least one independent build/load, serious/critical accessibility, or overflow failure. Three Brand runs recorded only contract-dependent gate failures:

- Grok 4.5, repetition 1
- Claude Fable 5, repetition 1
- Claude Fable 5, repetition 2

They are priorities for corrected re-testing, not automatic passes.

## How to read the page

- Raw and component scores remain useful descriptive evidence.
- Original gate records and capped official scores are retained for provenance.
- Qualification, capped scores and gate-dependent ranking were part of the original audit record.
- The corrected reanalysis is published separately in `ROUND1_REANALYSIS.md`; this page remains historical evidence.

The corrected evaluator accepts equivalent semantic controls, interactive descendants and the rendering techniques permitted by the briefs. Alternative positive controls passed before candidate outputs were re-tested.