30 September 2026 · 0.1.0-draft · Days 161–170. Agent author review from teacher, parent/carer, student and funder/reviewer viewpoints, followed by the separate publishing review below. Actual source, generated PDF and local browser checks are reported below. No physical paper/card trial was performed by the agents; no human educator interview, child access test, classroom pilot, local material approval, real audience or learning-gain evidence claimed.
Teacher pass
Read ten distinct scripts, thirty switchable routes, twenty worked examples, ten home bridges, ten extensions, both held fresh AJ/AK and separate key. Each script totals 3+4+7+5+4+2 = 25 optional starting minutes. Timings/preparation estimate unmeasured; rest/adaptation/later short components allowed. Start here precedes source details. Actual question and welcomed looking/trying come first; adult source/status/safety references are not read aloud as a full audit script.
The sequence advances after optical panels: choose a paper-roof question, predict/notice a flat form, compare an adult-prepared creased form if wanted, plan a child-chosen follow-up, then parts/material/property connections, keep/change/recheck, a small shared record/pattern and an explanation. It adds a dry low-level material-form/card-placement investigation; no repeat optical/water test or drawing substituted as a real trial. Spontaneous child questions retained; no prescribed folded winner, exact mathematical pattern or strength contest.
Desk-ran protocol branches using hypothetical inputs, not physical results: unloaded gap discernible then lower bend; already touching before and still touching; unavailable/unclear view; slipped load; shifted support; mismatched paper source/gap/end overlap; no recorded prior prediction; one row only; two differing comparable rows; adult report/model only. Correct next moves are bounded descriptions, pause/adjust/accessible property/report, truthful difference/interruption, missing rather than wrong prediction, or no repeated-pattern claim. These branches do not establish any particular real paper behaviour.
One small-group row can hold the useful sample. Note each child's relevant question/prediction/response/support only as needed; no full duplicated logs or required trial count. A fold changes form, not automatically material identity. SFU03 partial composition/material/property evidence explicit; actual whole object/source/property must support the aspect claimed. Guided matching is useful, not a Year 1 fair-test requirement. Actual motion/strength/cause/rating/waterproofness not inferred from a neat artefact.
Remaining teacher risk: stable low setup, actual welcomed observation and approved load/card placement must be preflighted locally. Paper form/contact/overlap may vary. Different actual conditions must remain visible. Do not read the long adult caveats to children or turn short shared records into a paperwork exercise.
Parent/carer pass
Walked Start here → one invitation → optional home route. No new object/purchase, recipe, food stamp/allergy assumption, device, print, intricate folding, photo/upload or returned evidence. Familiar/broad is not proof of safety. All actual preparation/creases/support fixing are adult-owned before learners; a complete model/director/no-contact route remains available if actual setup isn't appropriate.
Only one broad very light card on a low fixed dry tray setup; no heavy loads/stacks, throwing/dropping/pressing, child blades, real shelter/weather tests, body/face/hands underneath, animals or liquids. Stop/requested pause matters. A child can choose a welcomed unloaded gentle bend as a different real property question, not substitute imagined visual sag evidence. Home conversations need no trial and keep held sources out of practice.
Remaining parent risk: actual item/source/condition, surface, supervision, access/privacy and local policy need a real adult check. No physical safety approval or home trial resulted from this desk review. A parent may pass on all optional home choices.
Student pass — agent simulation only
Tried the invitation path “What would you like to find out?” → own question → prior prediction/not sure → optional adult action → actual look/report → one useful idea. The child can keep/change/check again or pass; unexpected, unchanged and unclear outcomes are valid. One useful phrase/sign/AAC/director response may be enough. No child participated in this review.
Big A4 diagrams and blank own-question/after-record spaces support practical observation without a prefilled actual answer. Source AJ/AK are held original fictional notebook reports; their exact fixed facts and open safe next questions separated in the key. Adult reads chunks neutrally, first response before models, no clock/voice/motor/record-length grade. Source words may be accessed by attributed description; that is not invented child visual notice. An all-paper drawing or planned action does not certify composition or direct investigation.
Optional helper has three native controls and short question/before/placement/look-tell reminders. It returns attention to the actual question/try; changing it doesn't record an experiment. The after graphic deliberately supplies no result. Exact plan words and adult notes expand only if needed. Complete paper/sign/AAC/director/no-contact routes work with JavaScript off.
Remaining child risk: joy, comprehension, practical timing and actual communication/sensory access untested. An educator must check whether the proposed physical action/question is useful to the particular learner; no learning-style or diagnosis assumption used.
Funder/reviewer pass
All nine canonical Foundation Science rows checked against actual official workbook ZIP/XML E/J cells and import; seven selected exact descriptions/codes/rows in crosswalk. Fresh independent official workbook HTTP 200, 2,049,917 bytes and canonical SHA verified. Current official indexed QCAA one-page Prep alignment 230388 and complete three-page techniques 230153-02, October 2023 read. Direct entry/alignment/techniques HTTP 403; screenshot attempts returned no usable pixel blocks. Current QCAA PDF bytes/layout unverified. Receipts distinguish current indexed content, blocked requests and actual local evidence.
Original saved plans/fictional notebooks/graphics are actual computer-created educational works with creator/date/type/purpose. Represented physical setups/results are proposals or fiction, not photographed experiments, real observer/community testimony or child samples. No imported English roof report result used as ours. No cultural knowledge or authority invented. Original text/art CC BY 4.0, code Apache 2.0; curriculum/font/third-party/community-held rights retained. Current primary copyright terms checked; no endorsement or QCAA instrument claim.
Software/print source checks can establish file consistency, not classroom efficacy or safe suitability. Human educator/child/access review, actual approved observations/support and current school/jurisdiction reporting decisions remain due. Any real external person/community source needs its own provenance, rights and relevant authority before use; our blank source slot is not evidence.
Actual print and browser checks
- Rendered all eight actual one-page PDFs to pixels and inspected every page in the contact sheet; also viewed held AJ/AK, flat/crease and shared-template full-page pixels. Schematics depict the specific source relationships, including narrow versus wide marks, no-load AJ before, reported midpoint contact/card in AJ after and AK wide row, and creases inside a separately labelled top-view inset. No physical trial/result from a proposal. Blank actual after/shared/question areas remain unfilled.
- Print audit checks all exact SVG/PDF words/order, actual A4 page and word bounds, complete Markdown alternatives and source-specific geometry. PDFs searchable not tagged; Markdown is linear reading copy. Tactile analogues are models, not actual unseen paper results; no loaded-gap tactile test. Source descriptions and actual sense/report/model status are separate.
- CUA inventory at review returned no enabled apps/browsers. Used installed local Chromium/Playwright fallback; no in-app or human trial claimed. Inspected actual 1280 desktop/390 mobile default and 320/390 expanded placement, blank-after and noJS screens. Client/scroll 320/320 and 390/390, no horizontal overflow; three controls readable, complete noJS static/paper route. Review contexts/browser closed; no server needed for local files.
- 16 form/mark/step states × 1280, 390, 320 = 48 cases passed actual support positions/profile/inset/one-card/blank-after nodes, exact text/ARIA/status, bounds, whole-word/native-option fit, expanded notes, keyboard/select/reset/reload. All three widths noJS; zero script errors/outside requests/storage. Reset/reload return flat/closer/question, physical try NOT TRIED remains explicit.
- Authoring defects found/fixed: crease marks initially needed a labelled top-view inset to keep them visibly on a paper representation; reported contact profile extended to preserve drawn end overlap; one long slash-separated adult-note phrase overflowed 320px, replaced with ordinary comma-separated words. Re-rendered all pairs/exact alternatives and reran full browser audit. No local outcome or held source fact changed.
Reproduction and remaining gates
Verifier defaults read-only: actual source pins/wording, daily scripts, routes, home choices, extensions and check isolation, eight complete A4 pairs/exact alternatives, sixteen byte-identical temporary rebuilds, seven meaningful temporary-copy faults, local links and complete manifest. Faults: wrong selected row, answer leak, falsely matched AK gap, false AJ contact drawing, outside API, adult report falsely child vision and helper falsely actual trial. Browser audit read-only unless explicit --capture; --write-manifest explicitly refreshes hashes after authoring. Portable prerequisites in README; no private workstation paths.
Remaining gates: actual adult material/setup/access/safety approval; real child/teacher usability/access trials; truthful actual local questions/predictions/observations/support; current full QCAA and local curriculum/reporting review; privacy/storage arrangements; external source rights/authority if used; actual willing sharing/response if claimed. No physical trial, child success, cultural authority, human approval or learning gain follows from an agent/software desk pass. Own leaf only; no shared edits, staging, commit or push.
Separate publishing review · 30 September 2026
A second agent read the ten complete lessons, original thirty routes, worked examples, family choices, extensions, exact source accounts, assessment/key, source rights and curriculum crosswalk. The held AJ prediction/contact comparison and AK changed-form/changed-gap limit remain exact; their open next questions accept supported alternatives. The optional physical investigation is a low dry paper proposal, with no physical trial or local suitability established by this review.
The reviewer made two teaching improvements: replaced the repeated long route wording with thirty concise day-specific actions, and added twenty optional child-chosen interests in ten pairs. A spoken setting or sketch beside the tray is sufficient; it does not add props below a load, change the approved single-card action, prescribe a winner or supply held check facts. The original twenty worked inquiry examples remain. These are optional choices, not extra required lessons.
The read-only verifier passed all nine canonical Science workbook rows, seven selected exact descriptions, timing/route/source isolation checks, sixteen byte-identical temporary print rebuilds, seven meaningful temporary faults, 71 internal links and the original 45 resource hashes before editing. After these prose changes the manifest was deliberately renewed and the default verifier rerun. The fresh source accounts, separate key, eight SVG/PDF pairs, generator, digital controls and curriculum/source files were unchanged by the publishing edits.
The reviewer rendered all eight actual PDFs and inspected their page pixels together, then the full AK page. The paired narrow/wide support geometry, reported gap/contact, separately labelled crease inset and blank actual-record areas are clear. This is a pixel/word/layout review, not a physical print, experiment, tagged-PDF certification or human learner test.
Actual in-app browser use at a 320 × 760 viewport covered creased/farther/placement controls, actual diagram and word/ARIA changes, a blank after panel, expanded plain-word plan, reset, native keyboard selection and a fresh-page reload returning defaults. Client/scroll width was 305/305 with no horizontal overflow in the checked state. The initial local preview process stopped during a reload, so that navigation failed; after restarting the preview, a new healthy tab passed the reload and was closed. The old error tab could not be closed through the browser API because its error-page URL was blocked by that API's policy; viewport was reset and the preview stopped. No alternate control bypass was attempted. This bounded independent check does not repeat or replace the author's 48 browser cases.
Teacher and family desk paths retain a short actual question/try and one useful response. A paper/director/source route remains complete when a physical try is unsuitable. Learner interest, timing, welcomed senses, safe actual setup, human access and educational benefit still need real local review. No listening, classroom trial, student success, full-year/state certification or public deployment is claimed by this content checkpoint.