# Release check · 13 September 2026

**Earlier release checks, followed by the 14 September repair addendum below.**

**Application and workbook:** local working release, with the limits below. **Photoshop teaching package:** teacher-led pilot pending complete native rehearsal and instructional captures. This report records checks performed by the implementing agent; it does not claim independent external review or classroom validation.

## Source reconciliation

- Inventoried 95 supplied file/archive entries at intake: 55 distinct byte sequences and 40 exact duplicates. No named local source was missing at initial intake.
- Preserved three distinct original Excel workbooks unchanged; their SHA-256 hashes match intake. Preserved and hash-verified 44 additional unique reference files from the intake cache (Markdown, JSON, CSV, HTML and the research DOCX). The three original ZIP paths are no longer available. Two TXT support files and one PATCH were inventoried but not extracted before those archives became unavailable; no claim of a complete binary archive backup is made. Later rebuilding used these preserved copies when the original Downloads paths were no longer available.
- Consolidated 332 dated lessons and 111 planned assessments. Applied ten Citrus lesson replacements by matching ID and four assessment revisions. Retained thirty date-neutral tickets and fourteen sneaker lessons separately.
- Read the four shared Google Sheets after sharing access became available. Inspected instructions, relevant skill references and bounded log/checkpoint headers. No class roster or student records from those references were imported.
- Reviewed primary Kagan articles/manual and curriculum/Adobe/IB sources linked in the teacher guide. Used selected sections of the supplied books as identified there; no claim that all related books or videos were reviewed.
- Detected the Multimedia Sound outcome transcription conflict. Flagged legacy entries and preserved student judgments. Proposed new skill links require school-adopted route/session confirmation.

## Application checks performed

Twelve Node test groups pass: missing criteria and whitespace evidence; support/access distinction; verification/rationale; latest eligible attempt and expiry; correction history; class sizes 0–40 with every selected learner once; reversible swaps; malformed backups; unsafe/duplicate planning imports; roster/course matching for student claims; CSV formula injection protection; and catalogue counts/unique IDs.

Isolated browser tests used synthetic students in a temporary Chromium profile. No tests altered the user’s working browser records. Checks passed for:

- Skill evidence and learning-plan save/reload, including rendering HTML-looking input as text.
- Blocking a readiness band when required evidence is missing; baseline/retest retention; the latest band and Reviewed status.
- Student claim import, repeated-import deduplication and separate teacher verification with no inherited self-rating.
- Group drafting, swaps, teacher review and a downloaded seating file containing every learner once and no readiness, private support notes or student IDs.
- Individual summary privacy, studio backup round trip, malformed-backup rejection and full app backup restoration. Empty display preferences may be initialised on restore.
- Planning import idempotence, retained teacher edits, rollback and byte-for-byte preservation of the existing `tos_*` keys during planning imports.
- All 56 combinations of four courses and fourteen app routes with no page errors.
- Standalone learner passport saving/reloading/exporting; fourteen document/asset links checked through the local server.
- Seven new app views and six teaching pages at 390-pixel width. One overflow in a printable writing line was repaired. New forms have visible labels and native keyboard-operable controls. Desktop pages and mobile screenshots were visually inspected. This is not a full screen-reader accessibility audit.

A focused synthetic-history test also verifies that re-importing a roster retains the earlier per-attempt criterion/level record, creates blank records for a new ID, and preserves both through a full backup round trip. New roster imports require stable IDs; an existing legacy-prefixed ID is retained when it matches the same original ID.

All 44 Resource Library links present at final download verification passed, including the source archive and complete release ZIP. The source archive has its own hash-verified index.

Additional maintenance repairs: removed automatic level-to-percent conversions; prevented a legacy evidence save without an observed level and a specific note; retained evidence when a real roster is re-imported with matching IDs; preserved raw legacy evidence details in full backups. Existing Gemini drafting was not invoked or tested.

## Excel checks performed

Created a 13-sheet workbook using the artifact spreadsheet library. Inspected rendered header/sample regions of every sheet, including the wide diagnostic/profile tables. Checked formula error cells and used temporary synthetic entries to exercise totals, missing criteria, assistance caps, roster/task matching, duplicate attempt IDs, review-date eligibility/expiry and independent reassessment on two dates. Test inputs were removed before export.

A second-row test exposed an artifact-library defect: `fillDown` rewrote quoted `MM12` and `COM11` course strings as relative cell addresses. The builder now translates only unquoted cell references, preserving absolute references and string literals. The revised second-row reassessment test passes. Test dates use the local calendar date, matching Excel’s TODAY behavior.

The workbook is prepared for 90 roster rows, 300 teacher evidence rows and 150 diagnostic attempts. Formulas use MAXIFS/MINIFS and target current Excel. Rows beyond the prepared ranges need extended formulas and validation. Validation helps entry; it is not access control. The source workbooks retain their original formulas and assumptions, including any prediction-model limitations in the source audit. They were not rewritten as validated prediction tools.

## Photoshop technical check and remaining release gates

Detected Adobe Photoshop 2026 version 27.10.0. The initial launch showed a scratch-disk space notice. A later attempt reached New Document through the keyboard; the dialog exposed an incomplete accessibility tree and pointer actions failed with AXError.notImplemented. Keyboard navigation did not provide reliable unit/size setup. The attempt was cancelled using Command-period, and the normal welcome screen was verified. No diagnostic document, PSD exemplar or complete task rehearsal was produced.

The two 800 × 800 PNG stimuli and 1200 × 1500 reference illustration are original geometric artwork. They are available with SVG sources and a manifest. They are not evidence of a Photoshop build. Actual A/B masking, editable text, changed-condition repair, PSD reopening and PNG output checks still require rehearsal on a working classroom installation. Genuine step captures must then be added. The fabrication prompt specifies those exact missing deliverables.

Instructional review repaired two assessment-design issues: both Photoshop variants now offer both hierarchy and mask-change opportunities; the DS source-use descriptor asks learners to weigh source contributions/limits that the synthetic case actually supplies. The DS answer key is a separate teacher file. The student distribution folder omits teacher-guide links and the key.

## Practical use and limits

Use the 1–4 number for a dated task family and the next teaching move. Preserve the criterion profile, help conditions and reassessment history. Trial moderation with several anonymised samples before relying on bands for grouping. Kagan informed participation/grouping; it did not validate this diagnostic. No pupil has been assessed or ranked by this work.

Confirm the actual school calendar, course route, applicable IB session guide and school reporting policy. Use current task evidence plus teacher review of access, relationships and classroom purpose. Readiness does not authorise safety-critical equipment use. Student self-ratings do not become marks. There is no automatic percentage conversion or student ranking display.

No deployment, school-grade import, Google Sheet mutation or third-party message was performed. Browser data and Excel remain separate. Use full backups and a consistent browser origin. The complete year’s resources still require the fabrication work identified by the preserved plans; their presence in the catalogue does not indicate a finished classroom package.

## Audit repair addendum — 14 September 2026

See ../../help/Audit_Repairs.md and ../../help/repair-test-results.json for the implemented persistence, validation, shared-evidence, curriculum and publication repairs. The current source catalogue and private German companion were added after the original resource audit. The updated workbook retains 13 sheets and adds learning reflection and separate outcome fields.

The earlier statement that no Photoshop files were produced is superseded: A/B scripted production builds and PSD reopening succeeded in Photoshop 27.10.0. The first files contain a typography encoding defect. A corrected script is supplied, but native Photoshop control then timed out and the corrected rerun remains outstanding. Read Photoshop_Pilot_Checklist.html before using those technical samples. Timed learner trials, complete manual step captures and inter-rater moderation remain outstanding. No accounts or live deployment were created.