# Audit repairs — 14 September 2026

This update implements the application repairs authorised after the original audit. Student account creation is deferred at the teacher’s request. The existing outcome workspace remains the assessment foundation.

| Finding | Delivered change | Limit or follow-up |
|---|---|---|
| A1 — persistence | One atomic application snapshot covers outcomes, rosters, Studio and settings. Failed writes preserve durable state, show failure and retain recovery data. Stale tabs cannot overwrite a newer revision. | Browser storage still requires dated external JSON backups. |
| A2 — restore validation | Nested records, methods, dates, levels, rubric fields and shared references are validated before replacement. Damaged stored data is write-blocked with a recovery download. | Restore a full backup when recovering connected records. |
| A3 — duplicate evidence | One stable evidence ID and artifact reference can support a skill and a separately judged outcome. Outcome capture also creates a shared event. Artifact corrections retain prior references. | Earlier records retain their original history; they are not silently merged. |
| A4 — source fidelity | Five active published Sound outcomes, a meaning-based crosswalk and code-based portable IDs. Historical Sound records remain available. IB targets are visibly local and linked to the official guide. | School/session adoption and individual historical judgments remain teacher decisions. |
| A5 — diagnostic rehearsal | Actual A/B production files were built and reopened in Photoshop 27.10, with mask faults and repairs. A classroom timing/moderation worksheet is supplied. | Typography defect in first files; corrected script awaits rerun after Photoshop control timed out. Timed learner trials and teacher moderation remain outstanding. |
| A6 — fabrication | Course-aware briefs plus twelve original creative Photoshop proposals, a source catalogue and a reusable fabrication prompt. | Proposals are clearly distinguished from fabricated and rehearsed lessons. |
| A7 — student portal | Teacher-managed Student OS and private standalone passport remain. Skill-use counts, reflection and next steps are improved. | Accounts, authentication and automatic submissions deferred by request. |
| A8 — public surface | Publishing uses an explicit dist artifact; private books/translation are excluded. The unauthenticated AI proxy returns 403 without calling an upstream provider. | No deployment performed. |
| A9 — planning/build | Earlier untaught lessons are flagged and filterable. Date changes require a dependency check. Default catalogue build uses checked-in consolidated data rather than a temporary intake folder. Local Git history preserves the pre-repair snapshot and the repaired release. | Calendar and prerequisite decisions remain teacher reviewed. |

The Excel workbook adds “What I learned” to student logs and separate outcome-link/judgment fields to teacher evidence. Its existing diagnostic formulas and 13-sheet structure are preserved. Original Excel references remain unchanged and downloadable.

The published Evidence Map was inspected as a reference. Its course selectors, outcome matrix, inspector, I-can statements, rubric workflow and O/C/P evidence remain in this local application. This is an extension of that foundation, not a claim that the published Netlify site has been updated.

Read [the pilot checklist](../outputs/gold-standard-2026/Photoshop_Pilot_Checklist.html) for exact rehearsal limits and [the creative library](../outputs/gold-standard-2026/Creative_Photoshop_Library.html) for source selection. The selected English mask companion is kept in the excluded private folder and is not part of the public release.

## Verification

22 core/persistence test groups passed. The full browser suite passed all 56 course/route combinations, skill/diagnostic/import/export workflows and mobile layout checks. Targeted browser checks passed shared evidence, artifact revisions, failed writes, malformed restore, corrupt startup and source labels. Historical roster/backup round trips passed. The guide passed 35 local-link checks, offline search, keyboard disclosure, print and three viewport widths. Workbook formulas passed representative completeness, assistance, expiry, duplicate-ID and reassessment checks; revised sheets were visually inspected. These are software and artifact checks, not classroom validation.