# Greenhouse-effect pilot audit · 13 September 2026

The pilot presents ten SL/HL units as complete study notes with 19 labelled diagrams, visible equation steps and optional action-driven experiments. It contains ten worked examples, 30 focused attempts, three combined exam attempts and six held-out final questions. Every unit includes an authentic source placement; original teaching and practice items are labelled separately. No predictions of future papers are claimed.

## Scientific diagrams

The 19 lesson diagrams use a white scientific textbook presentation with coloured radiation, labelled areas and explicit model assumptions. Sixteen inspected illustrations are combined with three exact native figures for oblique illumination, emissivity scale and planetary power balance. Full-size image links support closer inspection on small screens; every figure also has a written description and caption.

The written lessons use an independent teaching layer: short explanations introduce the physics and define terms before a separate Exam focus card connects the ideas to questions. Archived narration remains preserved and is not presented as a recording of this revised text.

Each asset has a teaching purpose, review notes and a checked content hash in figures.json. Delivered illustrations use quality-96 WebP copies at the original dimensions, visually checked for fine lines and label readability; the approved original PNGs remain preserved. These web copies use lossy compression and are not pixel-identical to the originals. Generation success is not approval: drafts with incomplete rays, incorrect angles or a mismatched bar ratio were excluded. Exact checks verify the twofold sloping footprint at 60 degrees to the normal, the 0.80 emission-bar ratio, ray/sphere intersections, and the radius endpoint. Narrative energy-flow arrow widths are explicitly schematic.

## Syllabus evidence

Checked the IB-authored first-assessment-2025 guide, updated November 2024, printed pages 40–41, and its update log on printed pages 79–81. The update log covers May/August/November 2023 and March/November 2024 and lists no B.2 amendment. Shared SL/HL coverage is confirmed. B2.U1–U10 are local references to the ten successive understanding bullets, not official IB subpart codes. Two further local references identify the equilibrium-temperature and surface–atmosphere guidance.

Source: https://ibpastpaper.com/SAIBSA/physics-guide-en_847a6832-850a-484f-a343-0de30855c17e.pdf

The original February 2023 school-hosted guide supplied in the plan was the starting reference. The updated IB-authored PDF was inspected through the web's extracted text. Its screenshot/download endpoints failed; no local PDF checksum or visual page inspection of that updated copy is claimed. No authenticated Programme Resource Centre access was available, so unpublished or later member-only notices remain an explicit verification limit.

## Inventory and editorial demand review

All 169 originally tagged question entries were read, including the 20 previously checked parts across 14 entries. Searching other extracted parts and their shared contexts found 16 additional candidate entries: 185 appearances in 141 normalised whole-question text groups. Part-level decisions include concepts, local syllabus references, assessed actions, existing prerequisites/setup, suitability, marking evidence and release status. The ledger retains all appearances and 119 potential duplicate-part groups; exact text equality is evidence, not proof that differing diagrams are interchangeable.

58 of the 185 candidate entries have no selected current B.2 demand. Examples include gravitational fields, muon relativity, optical resolution, a thermistor circuit inside a horticultural greenhouse, and distractors mentioning greenhouse gases. Other entries contain useful B.2 parts alongside unrelated mechanics, nuclear or electrical parts. The original archive's broad tags were preserved; this learning route uses its own explicit selections.

The context search recovered local roof-temperature and pipe-emission candidates. Inspection also recovered untagged relevant parts inside already tagged questions, including the solar-area explanation and equilibrium temperature. The 104 papers without complete parsed status remain an extraction gap; this audit does not certify full archive recall. Image-only text or unusual wording can still be missed. Broad legacy claims about volcanoes, energy technology, thermal expansion or ultraviolet transmission are flagged for qualifications or excluded from the pilot.

## Source release checks

11 unique authentic parts are used in 13 placements. All 19 displayed excerpt/setup/answer images were visually inspected, plus the four multiple-choice key pages. Source identity, question number, answer rows, diagram labels and relevant givens were checked. Three previously provisional parts received new, explicitly bounded question and answer crops; the original source catalog and PDFs are preserved. Source checksums, page coordinates and new asset checksums are in the authoring evidence files.

The 2025 surface-account focus supplies 238 W m⁻² absorbed solar radiation and 351.1 W m⁻² surface emission, giving 113.1 W m⁻² downward infrared. Its source scheme also accepts a prior rounded 240 W m⁻²; the combined checker accepts the corresponding approximately 111 W m⁻² result. The 2014 temperature example explicitly labels the constant/rounding adaptation. Numerical acceptance checks value and units, never reasoning marks.

## Original practice and assessment separation

The coverage matrix maps every original item to concepts, assessed action, representation, unknown, stage and family. Original practice fills the additional guided/independent variants required by this pilot; it is not evidence that no authentic source exists. Numerical answers were independently recomputed and selected rearrangements reverse-checked by conservation of energy. Written models/rubrics were reviewed for alternatives and the stated physical assumptions. Final assessment holds out inverse-albedo, multi-transfer planetary surface, molecular-to-global explanation, equilibrium ratio, partial atmospheric absorption and controlled feedback combinations. Authentic duplicate families are not used in that final set.

## Narration and interface validation

Narration uses Microsoft Edge en-GB-RyanNeural at a calm reduced rate, replacing the originally planned OpenAI service at the user's request. MP3 files and scripts are local, hash-linked and regenerated only when changed. The site needs no API credential or live AI service. The current reading presentation hides the narration player and continuous animations. The 40 recordings, scripts and existing storyboards are preserved in source for later production; unused audio files are omitted from the reading release; their presence does not imply a student playback control. Technical audio verification checks MP3 files, duration, script hashes and word boundaries. Independent local speech-recognition results, when completed, are retained separately; these are not a claim of human listening approval.

The current 71 Node checks cover filtering/provenance, complete lessons, numeric parsing, rounding, wrong units, misconceptions, original answers, audio hashes, source presence, audit completeness and family separation. The initial pilot validator passed for 3,602 active crops and all 917 original PDF checksums; later topical releases have their own additional asset evidence. It exposed 28 stale paths following earlier PDF renames; only catalog paths were repaired after byte-hash verification. See source-path-repairs.json.

Browser QA checks are recorded in validation.md. The session stores learner attempts in memory only; navigating preserves answers, resetting or reloading clears them. Written self-marking remains a learner judgment. Human listening review of all narration is not claimed.

## Reusable files

content.py / content_second_half.py: authored teaching and original items.
enrich.py: lesson placements and combined practice, separate from source questions.
build.py: versioned static catalog and source references.
reading.py / reading-figures.js / reading-experiments.js: reading structure, reviewed diagrams and optional experiments.
figures.json / figure_assets.py: teaching-purpose ledger, reviewed hashes and independent display catalog.
scientific_geometry.mjs: exact angles, ray intersections and calibrated emission bars.
figure-production/: generation prompts and rejected-draft inspection evidence.
source_assets.py: reproducible bounded original excerpts.
audit_inventory.py / audit_decisions.py: retrieval and explicit editorial decisions.
coverage-matrix.json: teaching coverage and original-question authoring matrix.
narrate.py / check_audio.py: incremental Ryan production and local recognition QA.
source-review.md / validation.md: review evidence and limits.