Skill: course-build → write-lesson (L01 a fact true of every piece of one material, same-material pairs, the rule last; L02 a two-class sort by test results, coins floater, sinker and dense last, plus the heavy-for-its-size mechanism on one measured scene; L03 a two-step routine, the capstone, worked on one tray) → write-example-sequence, design-check-sequence, sentence-atomise, write-explanation, write-mcqs; write-video-script (the Matter opening video, the intro, three lesson scripts, the summary; the elementary micro-lesson anatomy); review-figures and review-questions (author self-check, then a non-author agent's cold pass, below) @ b0fa825

Grade 4 Topic 13 — How to tell materials apart — build record (v1, 2 October 2026)

What this is, in plain words

Three short lessons, one per atom, and the opening video for the whole of Grade 4 Matter. A child meets one classroom: Miss Castillo's, with a tub of water on the table and a tray of odds and ends, and three children, Elliott, Emery and Emilia. First the child learns that a small piece of a material does exactly what a big piece does in water: a steel bolt sinks, so a tiny steel paperclip sinks too, however small and light; a wooden block floats, so a wood chip floats too. Then the child learns why that is: floating or sinking belongs to the MATERIAL, not the object — two spoons of one shape, one wood and one steel, go different ways — and the class sorts its materials into floaters and sinkers; a wooden block and a steel block of the same size on Grade 3's pan balance show that steel is heavy for its size, and "dense" is named once as the word for that. Last, the capstone routine: Miss Castillo asks for the one sample that sinks AND that a magnet pulls; Elliott grabs the stone because it sinks; the class tests the first property, crosses out the floaters, tests the magnet on the three left, and one sample remains, the steel bolt. Built with no blueprint, under the Grade 4 standing brief; the structural decisions are below.

Counts. 3 lessons · 3 scripts (150, 149, 148 spoken words) + the Matter opening video (85 words) + intro (118 words, 3 question marks) + summary (109 words) · 3 attached questions + the two polls + 3 summary questions · 32 quiz items (10 + 12 + 10; 1 typed) · 14 practice items (12 MCQ, 1 typed, 1 written answer; three Grade 3 retrieval items first; dense recalled by MCQ, sinker typed last) · PP100 bank 51 items (47 four-option MCQs, 2 typed, 2 multi-selects; 16 / 18 / 17 per atom; 17 easy, 20 medium, 14 hard) · 53 figures (47 SVG by code, 4 photo placeholders, 2 cartoon placeholders) · checker 0 FAIL, 16 WARN (answered below) · grade4/check_g4_articles.py 0 FAIL (articles 292 / 299 / 295 words; every script ≤ 150; intro 118; summary 109 with the set opener) · names run exit 0 · page builds (/tmp/g4t13_test.html, the Matter card added by content/add_strand_intro.py) · video batch built (handoff/G4T13/videos/batch_G4T13.json, 5 videos, NOT sent) and check_brief.py clean of closing findings on L01, L02, L03 and the summary (the intro's R79 "no closing bullets" hit is the standing accepted hit on every topic intro — an intro teaches nothing and has no rule slide; the batch tool's READY.md records it).

Structural decisions (in place of a blueprint)

Why these calls: the atom titles give the order and the capstone; L01 keeps to same-material pairs because its performance is the PREDICTION (a small piece from a big piece), and the moment the material changes is L02's whole point, so the two spoons open L02 rather than closing L01; the misconception "small and light things float" lives in L01's headache (the paperclip) because Grade 3 Topic 19 already confronted "big things sink" with the log and the nail, and the paperclip is the piece a child most expects to float; L02 coins "floater" and "sinker" (the atom's own words) because the sort needs two labels that belong to a MATERIAL — an object floats, a material is a floater — and because the PP100's typed hard tier needs a term to recall; "dense" is named once, as the label for heavy-for-its-size, never typed (James's rule), with the comparison to water spoken plainly ("more dense than water") and no formula, no number; the heavy-for-its-size evidence is drawn on Grade 3 Topic 18's pan balance with two blocks of the SAME size because "the pan that hangs lower holds the object with more mass" is a line the child already has; L03 takes Grade 3's two tests as its two properties because both are recalled, never re-taught, and because sinks-AND-magnetic is the pair real materials can satisfy (steel) while the floats-AND-magnetic pair has no single-material answer — that pair appears only as the toy-box "none of the samples" item and the attached cork question, where the honest answer is no. Every lesson gets a video: each is one scene a child can watch, and L01's and L02's misconceptions need the piece seen dropping.

Hollow objects: avoided entirely, and the simplification named once. No cup, can, bottle, boat or toy car appears in any tank, item or figure; every object is solid all the way through and of one material. L02's article ends with the two-sentence note "We made one simplification here. Every object in this topic is solid all the way through, with no air inside. A steel boat has air inside, so boats wait for a later grade." (taught terms only). Plastics are barred too, because plastics differ (some float, some sink) and the course has not sorted them.

Questions for James

• Do "floater" and "sinker" stand as the lesson's two coined labels, or should the sort say only "floats / sinks" and coin nothing? Context: the atom title names them; the vault's "coin no word the everyday word covers" could read "floats / sinks" as enough. Recommendation: keep them — they carry the property-of-the-MATERIAL point (an object floats; a material is a floater), they are the PP100's typed hard tier, and a Grade 4 child spells both. Built with them (coined in L02, used from then on; recognition before recall everywhere).
• Is "dense" (named once, as an adjective: "a material that is heavy for its size is called dense … more dense than water") the one naming you wanted, or should the noun "density" be spoken as well? Context: your rule names "density"; the register's sharpening (water as the reference) is Grade 5's MAT5-002. Recommendation: "dense" only in Grade 4; "density" the noun waits for Grade 5. Built that way; TOPIC["synonyms"] carries density so any accept list would honour it.
• Should L02 keep the two-sentence simplification about air inside a boat, or is avoiding hollow objects enough on its own? Recommendation: keep it — a child who has seen a steel ship will otherwise meet "every piece of steel sinks" as a contradiction. Built with the note.

The per-lesson table

Article length (the brief: 170–300 words, a subtitle or figure every eight idea lines or fewer): 292 / 299 / 295 words of idea, rule and subtitle text. L02 was 384 at first draft and was trimmed to 299 by folding the two "every piece of wood / steel" lines into one, cutting the restated forest line (the rule states it), and cutting the spoons' two lowering lines to one; nothing taught was lost (the sentence-atomise step 13 read).

design-check-sequence records

`

Skill: design-check-sequence @ b0fa825

Part: L01 (MAT4-001)

Type: (a) one-step prediction — given what one piece of a material does, say what another piece does (one of two labels).

Plan: recall MAT3-007 (how to find out: lower it in and let go, 3 opt, 0) → glass marble → bead, drawn BEFORE card (0) → oak log → twig (1) →

candle stub → tall candle, drawn (1) → steel nail → screw (0) → brick → chip, drawn (0) → whole potato → slice (0) → Emery's "a tiny steel washer

will float because it is small and light" (no) → Emilia's "the same wood cut into a cube will sink" (no) → transfer: a big piece sinks, a cut-off

small piece (3 opt, 2). Two-option string 0 1 1 0 0 0 1 1 (no four alternate; 4 of 8 at the second position); three-option 0 2.

Belongs in practice/test instead: the four-option prediction forms (PP types 1–3); the reversed direction (small → big) with a reason (PP type 3).

Drawn BEFORE cards: the known piece in its tank, the new piece held above the second tank; the figures carry no verdict chip.

`

`

Skill: design-check-sequence @ b0fa825

Part: L02 (MAT4-002)

Type: (a) one-step categorisation — from the results for several pieces of one material, floater or sinker (two labels); the heavy-for-its-size

item and the term items are the climb's top.

Plan: recall MAT3-004 (blocks A and B, same size, drawn balance: which has more mass, 0) → rubber, drawn results table (1) → copper in words (1)

→ oak, drawn table (0) → cedar in words (0) → apply: an untested cork stopper (1) → copper block v wooden block, same size: which material is heavy

for its size (3 opt, 1) → Emilia's "a heavy log sinks because it has more mass than a bolt" (no) → Elliott's TRUE "a tiny steel chip is as dense as

a steel bar" (yes) → term: heavy for its size → dense (4 opt among sister property words, 0) → term: every piece sinks → a sinker (3 opt, 0) →

term: every piece floats → floater, TYPED (after the MCQ that offered it). Two-option string 0 1 1 0 0 1 1 0; three-option 1 0; four-option 0.

Belongs in practice/test instead: the four-option sort (PP type 4); the odd-one-out (PP type 5); the reason items (PP type 7).

`

`

Skill: design-check-sequence @ b0fa825

Part: L03 (MAT4-003)

Type: (b) multi-step routine — two decisions in order (which samples fail the first property; which of those left passes the second), with the

worked example fully written on the page (the tray of six), so the scaffold sits on a SECOND case and the bare items on fresh ones.

Plan: recall MAT3-008 (a magnet pulls the nail: magnetic, 3 opt, 0) → scaffold on the pencil-case table: which sample fails Sinks? (3 opt, 1) →

which of the three sinkers does the magnet pull (3 opt, 2) → bare: the garden table (4 opt, 1) → the kitchen table, floats AND soft (4 opt, 0) →

the beach table (4 opt, 1) → the toy-box table, floats AND magnetic: none (4 opt, 3) → which test first when the order is given (2 opt, 1) → what

Elliott does after the first test (3 opt, 0) → Elliott's "she has found it after one test" (no). Three-option 0 1 2 0; four-option 1 0 1 3; two-option 1 1.

Belongs in practice/test instead: the judgement items about the routine (skip the second test; the order; two left; a failed sample; the count

tested) and the multi-select with two matches (PP types 8–9).

Fresh-case read: the scaffold's pencil case and every bare table were read against the article's tray (block, bolt, marble, cork, stone, candle)

and against each other; no sample set repeats.

`

The key strings

• Attached questions in video order (L01, L02, L03, S1, S2, S3): A B B A A B — 3 of 6 at each position (the batch tool prints the same string).
• L01 two-option 0 1 1 0 0 0 1 1; three-option 0 2. L02 two-option 0 1 1 0 0 1 1 0; three-option 1 0; four-option 0. L03 three-option 0 1 2 0; four-option 1 0 1 3; two-option 1 1.
• Practice: two-option 1 0 0 1 1 0 0 1 1 (the iron-nail and clay items swapped after the cold read found five alternating keys); three-option 0 1; four-option 1.
• PP100 MCQ keys: A 10 · B 14 · C 11 · D 12 (47 single-answer MCQs; the fixed prediction and sort sets are reordered per item so no letter tracks a verdict).
• Claims: L01 two (both no); L02 two (one yes, one no — the TRUE claim is "a tiny steel chip is as dense as a steel bar"); L03 one (no); practice two (no, no) and one TRUE-by-form recall (together). No never above 70 % of a lesson's claims where a lesson has two.

The example sequences, written out

L01 (fact form, one frame). "The steel bolt sinks. The steel paperclip sinks too. Both are steel." → "The wooden block floats. The wood chip floats too. Both are wood." → "The tall candle floats, with most of it under the water. The stub floats the same way. Both are wax." → the table (three rows, two columns of readings) → rule. Varied: the material (steel → wood → wax) and the direction of the answer (sinks → floats → floats); held constant: the frame, the tank, the big-then-small order. No non-example (the fact form); the near miss a child believes (small and light → floats) is the headache and the L01-q8 claim.

L02 (boundary pair, then one frame). Boundary first: "The wooden spoon floats, and the steel spoon sinks. Only the material was different." Then: "Every piece of wood the class tested floats, and every piece of steel sinks." → the six-row table → "A material whose every piece floats is called a floater. Wood, wax and cork are floaters." / "A material whose every piece sinks is called a sinker. Steel, glass and stone are sinkers." Sorting sentence for the misconception: "A big log has a lot of mass, but wood is light for its size, so the log floats."

L03 (routine, worked once). Step 1 on all six (three crossed out, the question "Does it sink?" printed above every verdict chip); step 2 on the three left (two crossed out, "Does the magnet pull it?" above the marks); one left; the yes / no table; the rule in three sentences; the order line.

The rules applied

Misconception rows (the catalogue is silent on floating; the literature named in the brief, and what was done)

Checker: 0 FAIL, 16 WARN (python3 build_tools_check_topic.py grade4/topic13_build/content/G4T13.py)

grade4/check_g4_articles.py: 0 FAIL (articles 292 / 299 / 295; scripts 150 / 149 / 148; intro 118; summary 109; opener ok).

Names run (python3 build_tools_names.py Castillo Elliott Emery Emilia → exit 0, all four "free")

Only the Topic 13 row of grade4/NAMES_POOL.md is used: Miss Castillo; Elliott, Emery, Emilia. Every claim-maker is one of the three children. Neighbouring Grade 4 Matter topics (14–16) are not yet built; Grade 3 Topic 19 (Fidel, Tomi) and Topic 18 names are not reused.

Sentence-atomise self-check (one row per lesson; every idea line one sentence and ≤ 22 words, S1 / S38 clean)

Figures (53; author's review-figures self-check on fifteen renders, then the non-author's pass below)

Opened at 1.5× (headless Chrome): tank_pair_wood, tank_pair_steel, tank_pair_spoons, tank_pair_wax, table_two, table_l02, cards_step1, cards_step2, rule_same_material, rule_two_props, rule_floater_sinker, balance_wood_steel, tray_samples, q_glass_before, sort_l02, pp_two_bath (in a 600-px page). Fixes from the renders: the rule cards are now sized from the rule's measured line length (460 px floor), after L03's rule overflowed its frame at the half-size miniature's width; tank_steel_headache added (the bolt sunk, the paperclip held above) so the L01 headache figure has its answer layer off. Every floater sits at the waterline with its registered part under (the candles mostly under, the chip and the spoon half under); every sinker rests on the floor; every chip inside its frame; every table cell inside its column (Topic 19's guard asserts it); item figures carry no verdict chip, tint, tick or cross, and their alt text says what is drawn without the answer (the blind comment carries it). The image briefs for the four photo placeholders and the two cartoons are in figures.py (IMAGE_BRIEFS); the picture stage waits for James's answers to the questions above.

Judgement calls, recorded

1. L01 keeps to same-material pairs; the different-material pair opens L02. The register's cross note gives L01 the prediction and L02 the property; putting the spoons in L01 would teach L02's point a lesson early.
2. "floater" and "sinker" are coined (James's call above) and carried by L03, the practice, the bank and the summary.
3. "dense" named once, adjective only, never typed; "density" held in synonyms for any accept list.
4. No hollow object and no plastic anywhere; the simplification named in two sentences in L02.
5. Real behaviours checked against Grade 3 Topic 19's drawn verdicts (one dataset per thing): the apple, the leaf, the orange, the cork and the candle float; the coin, the marble, the potato, the key, the brick, the stone and the nail sink. New materials sorted here: rubber (an eraser, a band, a ball) sinks; copper, aluminum, glass, clay sink; pine, oak, cedar, cork, beeswax float. A stainless-steel spoon is never put under a magnet (many are not magnetic); the magnet items use plain steel (nail, bolt, paperclip, washer, pin, hook, key, nut).
6. L03's second property is Grade 3's magnet test so both tests are recalled; "soft" and "hard" (Grade 3 Topic 16's property words) give the bare items a second property pair.
7. No bank item returns to a lesson scene after the cold read: the type-7 reason item that had reused L02's two spoons now asks about a wooden bead and a glass bead of one shape and size; the heavy-log reason, the same-size-blocks item, the cedar and clay classify items and the typed definitions were re-instanced too.
8. Four S36 words kept (bolt, cedar, stub, bamboo) with reasons; "slab" is the one for James.
9. Materials whose behaviour varies are barred, after the cold read: rubber (natural-rubber balls and bands float, filled erasers sink) joins plastic on the bar list; the chalk results table replaces it in L02's quiz, clay replaces it in the bank. A sponge (full of air) and an acorn (a sound acorn sinks) were removed from the item tables; a "steel bottle cap" row that floated was removed (it broke L01's own rule and is a cupped shape). "Steel" is dropped from every two-property table row and option where it would hand the key over (a child who knows steel is magnetic and sinks would never read the table): the rows read nail, screw, key, pin, hook, washer, nut, chain, tent peg; the material word stays only where two steel rows differ (the toy box, where no sample matches).
10. The tall candle is drawn lying on its side at the waterline (a long thin candle floats horizontally), after the cold read caught it drawn upright.
11. The intro sits at three question marks (the checker's limit); the two felt questions are the topic's payoff (what decides; does one piece tell you about every piece).

Owed

• Blind review of every key (the standing gate): the bundle is CUT — grade4/topic13_build/blind/2026-10-02/BLIND_BUNDLE_G4T13.md (96 MCQ, 4 typed, 1 written, 2 multi; keys sequestered in KEYS_SEQUESTERED_G4T13.json) — and waits for two cold readers; the non-author cold pass below is not the blind read.
• Item-figure blind text is data, never the answer (the coordinator's line of 2 Oct, after Topics 7–9 failed it): every item figure's blind= lists what is drawn — the object and its level in each tank ("the glass marble rests on the floor of the tank, under the water; the glass bead is held above the second tank"), the cells of a table ("stick: sinks; lump: sinks; small piece: sinks"; "pebble: Sinks? yes, Magnet pulls it? no; …"), the tilt of a beam ("the left pan, with block A, hangs lower than the right pan") — and never a verdict word. Grep of the bundle's [Drawing: …] lines for floater / sinker / dense / matches both / more mass / yes-yes / will sink / will float: 0 hits (grep -i "drawing:" BLIND_BUNDLE_G4T13.md | grep -i -c "floater\|sinker\|dense\|matches both\|more mass\|yes-yes\|will sink\|will float" → 0).
• James's answers to the three questions above; then the picture pass (4 photos, 2 cartoons) through David's pack.
• Videos: the batch at handoff/G4T13/videos/batch_G4T13.json (intro, L01–L03, summary; the Matter opening video is a sixth brief to add on James's word, as Topic 1's Physics opener was) is built and NOT sent.
• Non-author review-figures / review-questions pass: done (below) and applied; the blind read is still owed.

The non-author cold pass (review-questions + review-figures, a separate agent, 2 Oct 2026) — findings and what was done

The reviewer answered every item cold and agreed with every key (quizzes, attached, summary, practice, bank). Findings and actions:

After the pass: checker 0 FAIL, 16 WARN (the table above); articles 292 / 299 / 295; scripts 150 / 149 / 148; page builds; batch rebuilt, key string A B B A A B, check_brief clean of closing findings on L01–L03 and the summary.

v2 — reviews applied (2 October 2026)

Skill: course-build → write-lesson (the article and script lines) → write-video-script (the Matter opening video; L02's three balance beats) → write-mcqs (every item change) → write-explanation (fb, why and model lines) → sentence-atomise (every new sentence) @ b0fa825

In plain words. Three reviews read the finished topic cold: two blind key readers (every key agreed), a guidance-and-science reviewer, and a figure reviewer. Nothing they found changes what the lessons teach or in what order; the fixes make the opening video teach nothing (it had leaked Topic 15's answer), make three term items have exactly one true option, put the missing links into L02 (Miss Castillo puts the blocks on the pans; water is stuff too, so "heavier for its size than water" is said before the bank asks about it; "density" named once beside "dense"), make the L02 video say what its chips show, use one word for the water ("tub"), re-material the bank items that were the lessons' own scenes with new names, and tidy six figures. The bank still holds 51 items, at least five per atom, with difficulty rising by atom and the keys spread 12 / 12 / 12 / 12.

Finding → change

Renders opened at 1000 px (/tmp/g4t13_figs/, /tmp/g4t13_render.py): pp_slate_before, q_brick_before, q_balance_two_blocks, pp_balance_blocks, cards_step1, sort_l02, rule_floater_sinker, tray_samples, balance_wood_steel_headache, table_two_headache, pp_two_craft, pr_oak_before (and pp_results_bamboo, pp_results_steel, q_two_kitchen, q_two_toybox, tank_steel_headache re-rendered). Every one reads as its caption says: the tile and the chip one grey, the brick and the chip one red, both balance blocks one fill, the six cards in two rows, the tray in two rows, the headache balance standing on a drawn table with its chips inside the frame, the headache table empty.

Questions for James — built as recommended

• floater / sinker kept as the two coined labels (the register's own atom title; the reviewer agrees).
• dense AND density: "A material that is heavy for its size is called dense." and "How heavy a material is for its size is called its density." — both said once in L02, neither typed (RECALL_BY_MCQ); CLAUDE.md's rule as written and the built default. The bank and quizzes test only "dense".
• The boat line kept, in student words, no meta-talk: "Every object here is solid all the way through, with no air inside. A steel boat is full of air, so it is a different case."
• For the register (MAT4-002), a scope line to add: "wood" means the woods the class meets (pine, oak, cedar, maple, walnut, birch, cork); some tropical woods (ebony) sink, so the register's "every piece of wood floats" is read with that scope. Not a text change here.

Counts and key strings (v2)

• Scripts: L01 149 · L02 150 · L03 148 words; the Matter opening video 88 (≤ 90); intro 118; summary 109. Articles 292 / 329 / 314 words (L02 and L03 WARN above 300 after the review's added lines — the bridge, the coin's second sentence, the headache table's caption, the order line; Grade 3's approved articles ran to 505).
• Quiz keys: L01 two-option 0 1 1 0 0 0 1 1, three-option 0 2 · L02 two-option 0 1 1 0 0 1 1 0, three-option 1 0, four-option 0 · L03 two-option 1 0, three-option 0 1 2 0, four-option 1 0 1 3. Practice two-option 1 0 0 1 1 0 0 1 1 1, three-option 0, four-option 1. Attached A B B A A B.
• Claims: L01 two (no, no); L02 two (no, yes); L03 one (yes); practice two (no, no). 7 no : 2 yes.
• PP100: 51 items (48 four-option MCQs, 2 typed, 1 multi); 16 / 18 / 17 per atom; 16 easy, 21 medium, 14 hard; MAT4-001 easy + medium only, MAT4-003 medium + hard only (S31 clean); MCQ keys A 12 · B 12 · C 12 · D 12; types 1–10 hold 9 / 1 / 6 / 6 / 3 / 2 / 3 / 6 / 10 / 5.

Checker results (v2)

• build_tools_check_topic.py → 0 FAIL, 19 WARN: S8 (several short sentences per beat — the elementary micro-lesson, as in v1); S13 L02-q7 (names both materials, as in v1); S36 words: bolt, cedar, stub, slab, bamboo (answered in v1), domino (a toy every child knows), birch (a US tree, four letters), craft ("craft box" is a classroom word), heavier ×2 (the comparative of heavy, in the bridge line the review asked for — fine).
• grade4/check_g4_articles.py → 0 FAIL (two article-length WARNs, above).
• build_tools_names.py Castillo Elliott Emery Emilia → exit 0, all four free.
• build_tools_build_topic_site.py … /tmp/g4t13_v2.html builds (3 lessons, 14 practice, 51 pp100, 50 figures); add_strand_intro.py adds the Matter card; "tank" on the page = 0.
• build_tools_video_batch.py rebuilt handoff/G4T13/videos/batch_G4T13.json (5 videos, key string A B B A A B, not sent); check_brief.py clean on L01, L02, L03 and the summary; the intro's R79 "no closing bullets" and the three-question-marks note are the standing accepted hits on every topic intro.
• Blind bundle re-cut to grade4/topic13_build/blind/2026-10-02-v2/ (103 items: 97 MCQ, 4 typed, 1 written, 1 multi). grep -i "drawing:" BLIND_BUNDLE_G4T13.md | grep -i -c -E "floater|sinker|floats|sinks|heavier|dense|matches" → 0. Item-figure blind text now lists a results table's cells as what each piece did ("stick: sank") and a two-property table's columns in the cards' own question form ("Does it sink? and Does the magnet pull it?; the cells, in that order — pebble: yes, no; …").
• Rename check (the vault rule): grep for the OLD figure keys (pp_two_bath, pp_steel_before, pp_stone_before, tub_block_chip_cartoon, tank_pair_steel_above) in content and handoff → 0; grep for "tank" in the page and the batch → 0.

Items whose stem, options, key, model, figure, blind text or accept list changed (the coordinator re-reads exactly these blind)

• L01: L01-02, L01-04, L01-06 (blind text "tub"; L01-06's figure fill).
• L02: L02-01 (figure fill), L02-02, L02-04 (blind text form), L02-03 (stem: rod), L02-10 and L02-11 (swapped: q10 is now the sinker MCQ, q11 the dense MCQ).
• L03: L03-01 (options), L03-02, L03-03, L03-04, L03-06 (blind text form), L03-05 (rows, options, figure), L03-07 (rows, options, figure), L03-10 (stem, key — now yes).
• Practice: PR-04 (blind text), PR-08 (fb), PR-09 (blind text form), PR-11 (model), PR-12 (now the glass-sinker MCQ, two options), PR-13 (now the copper-dense MCQ, stem).
• PP100 (every item after PP-04 is renumbered; v2 numbers): PP-02, PP-03 (stem, figure), PP-04, PP-05 (option order), PP-09, PP-11 and PP-12 (figure rows, why), PP-13, PP-14 (option order), PP-15, PP-18, PP-20 (new), PP-21 (option order), PP-23, PP-26 (option order, figure fill), PP-27, PP-28, PP-30, PP-31, PP-33, PP-43, PP-44 (blind text form), PP-32 (scene, options, figure), PP-37 (new), PP-40, PP-42 (single MCQ), PP-45, PP-48, PP-50, PP-51. Unchanged in content, renumbered only: PP-01, 06, 07, 08, 10, 16, 17, 19, 22, 24, 25, 29, 34, 35, 36, 38, 39, 41, 46, 47, 49.

Sentence-atomise, the new lines (one row each)

Skill: course-build → review-lesson (Role B, B2) → review-questions → review-accuracy (non-author pass; filed by the coordinator from the reviewer's hand-back) @ b0fa825

Topic 13 — guidance, questions and accuracy review, 2 October 2026

The lessons teach the right thing in the right order and the science is sound; every key agreed cold; the blind-text grep is genuinely zero. Two record claims fail on re-performance: the Matter opening video is 101 words (cap 90) and its map line states Topic 15's rule; the bank-freshness claim is false on about nine items (lesson and quiz scenes re-materialled), and the type sheet carries no ⟂ list. Three term items have a second true option ("Which word says what steel is? — magnetic / strong / smooth / dense": all four). The L02 video and article are not one lesson (the video never says "light for its size", the MORE MASS chip is unspoken, the blocks jump onto the pans with no line, one "tub" becomes two "tanks"). The claim string is 8 "no" to 1 "yes".

The reviewer's view on the author's three questions: keep "floater" / "sinker" (the register's own atom title). "Dense" alone does not meet CLAUDE.md's rule as written ("name density when teaching heavy-for-its-size"), and TEKS / STAAR say "relative density": add one sentence after the coin, never typed, no formula — "How heavy a material is for its size is called its density." — and bridge steel-versus-wood on the balance to steel-versus-water in the rule with one line. Keep the boat simplification but cut the meta-talk ("Every object here is solid all the way through, with no air inside. A steel boat is full of air, so it is a different case.").

Science rows: an iron filing can rest on the surface film (say "bar"); a bamboo pole and a copper pipe are hollow, which the topic bars (chopstick; a wire, a rod and a nail); "every piece of wood floats" is false of ebony — a register scope line, not a text change.

Skill: course-build → review-figures (non-author pass on the 53 figures, rendered with headless Chrome at 1000 px, every item card opened, extents measured) @ b0fa825

Topic 13 — figure review, 2 October 2026 (filed by the coordinator from the reviewer's hand-back)

Sound where it matters: every tank tells the truth (floaters at the waterline with the register's share under, sinkers on the floor, same-material pairs alike at any size); the pan balance tips the right way; "same size" blocks are the same size; verdicts under their questions; Grade 3's drawers called, not redrawn; no alt or blind text names a key; all 16 item tables colour-free. Seven fixes, none a redraw.

1. A brick and its chip are two colours (q_brick_before: the brick red from Grade 3 Topic 19's brick_small, the chip sandy from its BRICK constant) under a stem that says "a chip of the same brick". brick_chip takes brick_small's fill; the Grade 3 inconsistency is queued for that topic's next opening.
2. Colour tracks the key on both balance items (q_balance_two_blocks, pp_balance_blocks: the grey block, read as steel, is the heavier). Both blocks in one neutral fill; the letters carry the identity.
3. cards_step1 is 618 wide and two chips run border to border. per_row=3, w=150.
4. sort_l02 is 600 wide and about 70 % empty. W 540, boxes about 140 tall.
5. tray_samples is 620 wide. Two rows of three at 480, or object-word chips.
6. balance_wood_steel_headache runs one unit out of frame and the blocks "wait on the table" with no table drawn.
7. Three keys placed nowhere: tank_pair_steel_above (delete), table_two_headache (place before the finished table), tub_block_chip_cartoon (place at "Now wood" or delete — a cartoon brief is owed for nothing otherwise).

Note: Topic 13 adds keys to Grade 3's OBJ / H / UNDER dicts at import (additive, in memory) — a line for the record.

Best: tank_pair_spoons, cards_step2, table_two. Weakest: q_brick_before, the two balance items, cards_step1. Renders: /tmp/g4t13_figs/*.png.

Skill: course-build → review-lesson (Role A blind key read, two readers; Role C adjudication by the coordinator) @ b0fa825

Topic 13 — How to tell materials apart — blind read, 2 October 2026

Every key agreed by two blind readers; no stem or drawing stated its own answer. Bundle: 103 items (the author had already grepped its drawing lines clean). Adjudication: MCQ disagreements 0 for both readers; typed off-list 0. Multi-selects PP-42 (B, D) and PP-43 (A, B) match both readers; PR-11's written answers carry the model's points.

Flags carried into the fix pass

Readers' files: blind/2026-10-02/READER_A.md, READER_B.md.

v2 re-read, 2 October 2026 (Reader C on blind/2026-10-02-v2, the whole bundle)

Every key agreed (103 items; MCQ disagreements 0; the multi-select PP-43 A, B matched; typed answers on the lists; the written PR-11 in the key's words). No two-defensible items; the three "which word means heavy for its size" items each have exactly one true option. The reader's two lesson-text checks are met: L02's article and script coin "dense" and say "How heavy a material is for its size is called its density." before the quiz, and L02 teaches "steel is heavier for its size than water, so steel sinks" (the water bridge), so PP-41 and PP-42 test a taught step. Soft notes, not applied: L02-11's distractor "magnetic" arrives one lesson before L03 coins it (a Grade 3 word); L03-02 and PP-51 are the heaviest stems; accept "floating material" / "sinking material" at the next touch. Reader's file: READER_C.md.