# Author and independent agent run-through

30 September 2026 · teacher draft · no human classroom pilot, child engagement result, measured gain, assistive-technology certification or authority endorsement.

## Four desk paths

| Role | What was actually walked through | Finding / revision |
| --- | --- | --- |
| New educator | Prepare broad marks once; run ten daily scripts; read exact sources; keep a dated key; take independent fresh responses; choose next teaching. | Whole unit is AB, not AB plus finish. Actual child movement, adult performance, object model and score-only plan have separate records. First/then/over prompts and final handoffs are supplied. |
| Parent | Read tiny home choices without buying/printing/filming/uploading; discuss an own drawing opener; choose private keep. | Home is optional and fresh checks stay unrehearsed. A useful 2–4 minute choice needs no performance or submitted evidence. |
| Learner perspective | Choose two own ideas, read order/repeat/rest/ending through an available route, reject the author's purpose, optionally share or keep. | Actual participation was not tested. Adult-guided chunks and local access checks remain essential. Small-screen interaction was initially buried below a long legend; the activity now comes first with a native expandable mark key. |
| Potential funder/reviewer | Follow precise workbook rows, source/rights ledger, exact generated asset text, assessment limits and review records. | Original downloadable assets and actual accessible making opportunities are supplied. Full-year/all-state completeness and effectiveness remain held; no fictional example is promoted to a learner outcome. |

## Independent critique and fixes

The Visual Arts subject agent independently read the source cells and current official indexed QCAA alignment/conditions, six core documents and then routes/examples/home/read-alouds/crosswalk. It rendered and inspected all eight actual PDF pages, checked source/check shapes/orders, selectable words and page bounds. It exercised 18 planner combinations: three scores × two views × desktop1280/mobile390/mobile320, with exact order/actual shape/ARIA/options fit/bounds/keyboard/reset/reload and no-script routes at each width. No outside request or page error occurred in that check. Temporary contexts/browser closed.

Three concrete clarity fixes followed this review: Day37 now asks “What was this model made to help us try?”; standalone key headings explicitly show A/B; mobile paper-table caption now spans the full available width. The reviewer visually rechecked the corrected A/B PDF/text and both caption screens, with and without JavaScript, at320/390. Caption reads in two lines rather than seven to ten. Root subsequently moved the unchanged key into a native expandable section after the activity; root's actual browser use checked small-screen selection, Words/Pictures, the key, reset and reload after this layout change. Source orders and content remain unchanged.

## Reproduction and scope

The [read-only verifier](verify_pack.py) checks the exact local source hash/import/direct workbook cells, four code descriptions, ten scripts/timing/handoffs, 30 routes/20 worked choices/10 home choices, fresh isolation/key IDs, actual-record/source boundaries, eight one-page A4 pairs with exact PDF/SVG words and linear alternatives, font-based text bounds, all 17 deterministically generated print/text files, inline-script syntax/no network-storage-media API, complete paper orders and local links. [Manifest](manifest.json) records included-file SHA-256 digests; explicit `--write-manifest` is an authoring operation. Routine verification must preserve pack bytes/size/mtime.

Root default audit passed: 36 hashed files plus the manifest, 52 local links and all 17 print/text reproductions. Root inspected the actual eight-page rendered contact sheet. Sixteen isolated adverse cases were rejected: timing, daily handoff, route, worked choice, home day, fresh source leak, wrong official row, false creator boundary, planned audience substituted for actual field, altered score order, network call, wrong live-status setting, wrong A4 size, text outside safe bounds, mismatched PDF/SVG and manifest drift. The routine run and probes preserved the original bytes/size/mtime snapshot, including ignored local lint-cache files; no original asset was mutated to test these cases.

Root browser use at a 320-pixel viewport (305-pixel content width with scrollbar) showed equal client/scroll widths, correct five-moment With pauses Words view with no SVG pictures, correct four-moment Return ending, expandable native mark key and Reset restoring Two units/Pictures. Reload cleared score/view choices; the later source-layout change only moved the key after the activity. The paper caption matched full table width. Review tab closed, viewport restored and temporary local preview stopped. One missing automatic browser favicon request was a preview-only 404; the page's own controls/assets worked. No real learner or performer participated.

## Before actual use

An educator must check current local syllabus and timetable, learner comfort/access and appropriate practical safety, any outside-source rights/cultural authority, willing sharing/privacy and what actually happened. A graphic source's documented purpose supports one bounded source-response opportunity, not complete people/community coverage. Quiet or small movement can be real Dance when chosen and actually observed; the planner is a score comparison only. PDF text is selectable but untagged; alternatives are separate routes and not certification. Any published/assessed translation needs relevant speaker review. Candidate sound or filmed child assets are not included.