Topic 24 "Where a plant's body comes from" — your read (8 October 2026)

The science is right and in the right words, the reviewer found: the keystone claim with "mostly"; the plant never eats; the control group shown kept for comparison before it is named; the willow numbers recomputed from the classic report (200 lb = 90,718 g; 2 oz = 57 g; 5 lb ≈ 2 kg; 169 lb ≈ 76 kg); "carbon dioxide" recalled with no mechanism; Grade 4's six sentences found verbatim in the Grade 4 modules; the two beliefs read in the register files; every subtraction in the topic recomputed and right. The ten Grade 4 faults were looked for one by one: no bald jargon (one coined term, in the frame; "fertilizer" described first), no invented belief, every verdict recaps the claim it answers, "So" opens only deductions, no lettered cases, no vague frame, contrasts in one frame, the convenience name never tested, no quoted child except on the register's two beliefs.

Fixed: the L03 rule card (its first line ran off both edges); the soil bar's label, alt and caption ("too thin to see" over a visible sliver); the summary still (76 kg read with the small tree on the pan); a double-negative caption; L01-q3's stem carrying q2's key; "got the most of the change" twice; PP-41's absurd distractor and PP-37's second-defensible one; PP-50 / PP-51 re-scened on their own pear-tree table; "pot test" said in passing before the rule; L01, L02 and L04 as two parts with a check between.

Left: the working-line form (the vault's unit-on-the-line against the course's every-sentence-has-a-verb — both met, recorded as a collision); "Bean plants grow about 20 centimeters in two weeks anyway" (read as Yara's plants; the rule line fixes the scope); the leaf's midrib under its label; "backs the claim"; the L04b first key's list shape (the key-tell checker reads 0 of 111). Phone-width rendering was not verifiable from this machine's headless browser (it clamps near 500 px); the builder's CSS reflows at 400 px and every figure scales.

55 items in content/G5T24_pp100_bank.py: 47 four-option MCQs, 7 typed (two "control group", one "water", four produced differences: 7,000 g; 12 g; 24 kg; 18 g) and one two-of-four select. Per atom: LIF5-010 13 · SR5-004 13 · SR5-005 12 · LIF5-011 17. Easy 20, medium 19, hard 16, rising with the atom's place. MCQ keys A 14 · B 13 · C 11 · D 9. Ten question types, one exemplar each on the type sheet. Every hard item carries its "hard because" line; the two claims are "X is wrong. What is the correct reason?" with single-statement reasons; every scene is fresh against the lessons, quizzes and practice, and now against the bank itself (the lemon table read twice as MCQs, the pear table twice as numerics, each pair a draw constraint).

check_topic 0 FAIL, S46 B = 0 C = 0 D = 0 · check_terms hits {} · check_g5_articles 0 FAIL (articles 353 / 405 / 293 / 319 / 293, each long one in two parts; scripts 148 / 150 / 147 / 148 / 149; intro 112; summary 149) · key tells 0 of 111 · check_svg_ids PASS (43 figures) · check_lessons 0 FAIL, S118 0 FAIL · names: one child per name · blind bundle re-cut (116 held; drawing lines free of verdict words) · batch rebuilt, check_brief.py every video ✓ (the standing summary note only). Not sent; not committed.

In plain words

The topic teaches the band's keystone the right way round and in the right words. A plant's body is built mostly from the air and from water, not from the soil; the oak's flat ground is seen before the sugar line is recalled; Grade 4's sentences come back word for word; "carbon dioxide" is recalled with no mechanism; the bag that says "plant food" is met by name and the plant never eats; the control group is shown kept for comparison before it is named; the willow test's numbers are the classic ones in metric and every subtraction in the topic comes out; the two beliefs the children voice are the register's two, met in the lessons and tested with their sources cited. Nothing here would make a biologist reach for a pen.

One thing had to change before anything else: the L03 rule card's first line ran past both edges of its card on the render and on the built page. Ten smaller things were worth fixing and are fixed: a drawn soil bar labelled "too thin to see" that a child can see; the summary still's scale reading 76 kilograms with the small tree on its pan; a caption with a double negative; a quiz stem that carried the previous item's key; an awkward "the most of the change" in two feedback lines; a bank distractor no child would pick and a bank distractor that was the key in other words; two hard typed items that re-read the lemon table two MCQs already read; the plain name "pot test" used in the rule and twenty items before any line had said it; and the three long articles (L01, L02, L04) now in two parts with a check between, James's 2 October form. The author's own calls (the L04 / L04b split, the willow numbers, the typed term moved out of L02's quiz, the working-line form) all stand.

Every gate is back at its stated result; the blind bundle is re-cut at blind/2026-10-08-reviewed/ (116 items, every [Drawing: …] line free of verdict words); the batch is rebuilt and clean. Nothing was sent or committed.

The author's recorded calls, judged

Findings — one row each

Verdicts: Must-fix / Worth fixing / Simplification-leave / Not an error. "Applied" means the rewrite is in the content files and every gate was re-run.

Role B — the guidance checklist, each check named (clear unless a row above says otherwise)

• B1 structure. One atom per lesson (L04 and L04b both claim LIF5-011, a routine in two parts); forest–trees–forest in every lesson (headache scene, narrator's question, the detail, the closing return "Here is … again"); prerequisites taught before use (Grade 4's sugar line, fair-test sentence and mass lines recalled by the recall bank BEFORE the video that needs them, and said in the lesson in Grade 4's words); "control group" said only after L02 coins it (S46 A = 0); concrete before abstract (the flat ground before the sugar reasoning; both groups measured before "what happens anyway"); every scene noun introduced before it moves (no staging imperative; "Suppose" and "Now consider" open each new case). Clear.
• B2 sentences. Every idea line ≤ 22 words, one sentence, a noun subject and a verb (S38 fragments = 0); causes before effects; the explicit link stated ("Only water went into the pot, and air surrounded the leaves. So the tree's body came from the water and the air."); contrasts in one frame ("If soil built the tree, the soil would lose a lot of mass. If water and air built the tree, the soil would lose almost none."). Rows 4 and 6 are the only wording findings.
• B3 words. Plain word before the term everywhere ("a gas from the air called carbon dioxide"; "fertilizer, a powder people sprinkle on soil"; "the group that gets none of the change … is called the control group"); one word for one thing (never "weigh" beside a mass; "mass" throughout); units as words in prose, symbols on drawings; no figure referred to that is not on the screen. Row 10 (the bare "pot test") is the one finding.
• B4 checks. L02's fluency run (two which-group, three yes / no, the two near misses the lesson pinned, one drawn plan, the term recognised last); design-check-sequence types match the sets (L01 and L03 type (c) what → why → claim → transfer; L02 type (a); L04 and L04b type (b)); every check stands alone; each boundary case's sorting sentence is on the page ("A little of the change is still the change. The control group gets none of it." for L02-q4 and q6; "Without the control group, nobody could tell…" for L03-q7; "If soil built the tree, the soil would lose a lot of mass." for L04b-q7); recall MCQs before the checks that lean on earlier grades (RECALL_BANK, six items, each before its lesson's video). Row 5 is the one finding.
• B5 explanations. fb ≤ 30 words, one fact aimed at the option; why ≤ 60 words, one link per line; calculations as working lines (row 13 stands). Rows 6, 7, 8 are the findings.
• B6 figures. Every one of the 43 renders opened at 2× and on its built page; the table below. The opening figures (the oak cartoon; Yara's two rows with no verdict; the scientist with the sapling) carry none of their lesson's answer; the headache figure for L02 is the teaching figure (pots_two_groups) with its title on and no "control group" label.
• B7 register. No "in this lesson" / "you can now"; the glossary's one terms string carries no provenance; the intro coins nothing (S15 clear), asks two questions, ends on a topic-specific ready line and the poll; the summary opens on the exact line, recaps five rule sentences (145 protected words) and no example, ends "Okay, now your turn." Clear.

Review-questions — the item pass (every quiz, practice and bank item read with its key)

Key-tell checker before item 1: KEY TELLS: 0 of 111 MCQs flagged. Checks 1–83 run on every item; rows below name the hits; every other item had no fault.

Key letters (check 18): quizzes L01 B C A B · n · A C A — L02 B A · y n y · C · B · B — L03 A B C · n · A B C A — L04 B B A C B C A A — L04b C C · 38 · 40 · n · A A C; no four alternate. Bank A 14 · B 13 · C 11 · D 9 (A = 30 %). Attached B A A A B A B B. Design-check-sequence type per part (check 13): L01 (c) ✓, L02 (a) ✓, L03 (c) ✓, L04 (b) ✓, L04b (b) → bare ✓; each set matched its plan.

Review-figures — every render opened (43), then the built pages

Pages opened: topic / intro, every lesson's article route and quiz route (L01, L02, L03, L04, L04b), the L01 and L04b script routes, summary, practice, PP100 — at 1100 px, and after the fixes the L01, L02 and L04 articles and the PP100 route again. The page forces one colour scheme (no prefers-color-scheme rule), so one shot per route. check_svg_ids.py: 43 figures, 0 ids, PASS.

Review-accuracy — the subject-expert rows (the brief's named checks first)

Deliberately omitted (right to leave out at grade 5): the mechanism inside the leaf; the minerals' names and roles; respiration's loss of mass; the fallen leaves of the willow's four autumns; van Helmont's name and his own wrong conclusion (water alone).

Gates after the fixes

• build_tools_check_topic.py → 0 FAIL, 37 WARN; S46-COUNTS G5T24 B=0 C=0 D=0 (A = 0: no S12 FAIL); S30-COUNTS typed=4 fail=0 warn=0; S37 strand=0; S38 fragments=0; S44 negative_stems=0; S39 hits=0. The WARNs are the author's answered set (S8 three-sentence beats, S36 unusual words, S29 unit symbols on drawings) less one S8 beat (L04 beat 2 is now two sentences) plus two S36 words the fixes introduced and kept: "sliver" (the gain-loss figure's own label, now matched by its caption) and "pear" (the new table's tree).
• build_tools_check_terms.py → hits: {}.
• grade5/check_g5_articles.py → 0 FAIL; article WARNs at L01 353, L02 405, L04 319 (each now two parts with a check between: 191 + 162, 224 + 181, 179 + 140); scripts 148 / 150 / 147 / 148 / 149; intro 112; summary 149.
• check_key_tells.py → KEY TELLS: 0 of 111 MCQs flagged (a: 0, b: 0, c: 0).
• check_svg_ids.py → PASS (figures read 43, ids found 0).
• check_lessons.py --adapter tools/check_adapter_g5.json → 0 FAIL, 76 WARN; S118 0 FAIL.
• build_tools_names.py Xavier Yara Yusuf Xu → all INFO (cross-grade recurrence) or free; one child per name in the topic.
• build_tools_blind_bundle.py … blind/2026-10-08-reviewed → bundled 104 mcq, 11 text (7 numeric), 1 multi; adversarial rows 4; held 116. Every [Drawing: …] line grepped for "far too little / did nothing / mostly from / is the control group / none of the change": 0 hits.
• build_tools_video_batch.py → 7 videos to handoff/G5T24/videos/batch_G5T24.json; check_brief.py → every video ✓, the standing summary/protected note only (145 words); grep -ci strand → 0. Not sent.

Article and script counts, before → after (the course count: idea, rule and subtitle lines)

Items changed (ids; the keys are in KEYS_SEQUESTERED_G5T24.json, never here beside the ids' new wording)

L01-03 (stem), L02-06 (one fb), PR-07 (one fb), PP-37 (one distractor), PP-41 (one distractor), PP-50 and PP-51 (new scene, new figure, new answers). No key letter moved; no option order changed. The two cold readers read the whole reviewed bundle, so no separate re-read list is needed; blind/2026-10-08-reviewed/CHANGED_IDS.md carries the ids for the adjudicator's convenience.

For James

None. Every change above is one his rulings already settle (the two-parts-with-a-check form, figure–text agreement, a stem never carrying an earlier key, fresh instances in a bank, a distractor a child would pick). Nothing here would change what a student learns by his answer the other way.

What this is, in plain words

Four lessons (the last in two parts) on the band's most counter-intuitive fact: a plant's body is built mostly from the air and from water, not from the soil. Xavier stands under a schoolyard oak that grew from one acorn; the ground around it is flat, so no soil went missing. Grade 4's line is recalled — the leaf makes sugar from water and carbon dioxide — and the one new step is added: the plant builds its body from that sugar (L01). Then the two Scientific Reasoning lessons the pot test needs: Yara sprinkles fertilizer on ten bean plants and none on ten others; the group that gets none of the change is named the control group (L02, coined for Florida SC.5.N.1.4, taught as an example sequence); two weeks later both groups have grown 20 centimeters, so the fertilizer did nothing — the control group shows what happens anyway, and Yusuf's "the other pots were just spares" is the register's documented belief, judged after the evidence (L03). Finally the classic willow test backs the claim: dry soil and a small tree on the scale, only water for five years, then the scale again — the tree gained 74 kilograms and the soil lost 57 grams, about one small apple, so the lost soil is far too little to build the tree (L04 the set-up, L04b the numbers; the capstone). The only new word is "control group"; "carbon dioxide" and "fair test" are Grade 4's, recalled in Grade 4's sentences.

Counts

• 5 lesson entries (L01, L02, L03, L04, L04b); articles 333 / 394 / 285 / 277 / 276 words; scripts 147 / 149 / 147 / 150 / 149 spoken words; intro 112; summary 149 (five rule sentences, 145 protected words).
• Quizzes 8 + 8 + 8 + 8 + 8 = 40 items (36 MCQ, 2 numeric; L04b's two numeric items sit after two MCQs); 5 attached questions + 3 summary questions (keys in video order B A A A B A B B, A 4 of 8); practice 13 (3 Grade 4 retrievals, 1 numeric, the term recognised then typed last); recall bank 6.
• PP100 bank 55: 47 four-option MCQs, 7 typed (2 text "control group", 1 text "water", 4 numeric), 1 multi-select. Per atom: LIF5-010 13 (8 easy / 3 medium / 2 hard), SR5-004 13 (6 / 5 / 2), SR5-005 12 (3 / 5 / 4), LIF5-011 17 (3 / 6 / 8). Tiers 20 / 19 / 16. MCQ keys A 14 · B 13 · C 11 · D 9. Ten types (records/TYPES_G5T24.md).
• Figures 42: 33 svg by code, 2 photo placeholders, 7 cartoon placeholders; every photo / cartoon key has an IMAGE_BRIEF.

Structural decisions (in place of a blueprint)

Lesson order and titles. The architecture row's order, LIF5-010 → SR5-004 → SR5-005 → LIF5-011, with the capstone split into two parts on the brief's word ("a long routine lesson … LIF5-011 … is split into short parts with a check between"):

1. L01 A plant's body is built from the air and from water (LIF5-010, Fact).
2. L02 The group that gets none of the change (SR5-004, Categorical: an example sequence, the term coined explain-then-label).
3. L03 Why the control group matters (SR5-005, Fact: what the control group shows).
4. L04 The pot test: how it is set up and L04b The pot test: what the numbers show (LIF5-011, Routine, two parts, a quiz between).

The running scenes. One class (Miss Xu's, named in text only). Xavier's oak (intro and L01); Yara's twenty bean plants, ten with fertilizer and ten with none (L02 and L03), and her seed trays (L02's third case, L03's second test); the classic willow test (L04, L04b: a scientist four hundred years ago, unnamed in student text; the numbers are van Helmont's in metric — dry soil 90,718 grams → 90,661 grams, a loss of 57 grams; tree 2 kilograms → 76 kilograms, a gain of 74 kilograms). One dataset per stage: beans +20 cm both groups; seeds 3 days warm, 6 days shelf.

Headache scenes and who asks. L01: Xavier under the oak; the narrator names the wonder ("It seems almost impossible: one small acorn became a trunk too wide to hug") and asks ("So where did all that wood come from? Let's look."); Xavier voices the register's belief in the body ("The wood is soil, packed together, that came up through the roots.") because it IS the lesson's point and LIF5 SD-2 documents it; the verdict after the evidence recaps and contradicts his words. L02: Yara's two groups; the narrator asks ("One of Yara's two groups has a special job in the test. Scientists have a name for that group. Which group is it?"). L03: Yara measures; Yusuf voices the register's SR5 SD-1 belief ("The other ten pots were just spares, in case a plant died."); the narrator asks what the fertilizer did; the verdict recaps and contradicts after the control group is measured. L04: the small tree, the pot and the heap of soil; the narrator asks ("How can a scale show where a tree's new body came from?"). L04b: five years on; the narrator asks the prediction ("What will the scale read: much less than 90,718 grams, or almost the same?"). Two of five openers quote a child, each on a documented belief; three are narrator-only. Openers vary: a tree, two rows of pots, a measuring ruler, a pot and a scale, a prediction.

Cast. Xavier (the oak, the LIF5 claim), Yara (every plant test), Yusuf (the SR5 claim). Each name after the first does a job the first could not: Yara is the experimenter across two lessons (one character carries the scene); Yusuf voices a belief Yara, who set up the control group, would not hold. Names from Topic 24's row of grade5/NAMES_POOL.md.

Example sequence (L02), written out. Frame "Is this group the control group? It is, if it got none of the change." The rule is coined BEFORE the run (explain then label on Yara's real test), so each frame trails its because. Shape n p p p n: (1) Yara's second group, half a spoon of fertilizer on each pot → no, it got some of the change; (2) the same group, no fertilizer at all → yes; (3) Yara's seed tray on the classroom shelf, the warm cupboard being the change → yes; (4) the children who learned a word list with no music → yes; (5) the same children with quiet music → no. Both crossings keep the object (Yara's second group at 1→2; the children at 4→5); between positives only the object changes, the change held at none. Sorting sentence for the boundary: "A little of the change is still the change. The control group gets none of it." The script carries the half-spoon non-example and the no-music positive; the article carries the whole run (the block's where says so). S118 passes.

L01, L03, L04 and L04b carry no sequences block: L01 and L03 are facts shown on their cases, L04 / L04b a routine run once on the classic test and practised on fresh pot tests in the quiz.

Check types. L01 type (c): what / why / the documented claim (Yara's carrot) / transfer (a floating pond plant; a mowed lawn) and the boundary that the water does come from the soil. L02 type (a) fluency on described tests (which group / is this group the control group), the near misses the lesson pinned (a group that got less; every tray got the change), one drawn plan, the term recognised last (typed recall in the practice set and the bank's hard tier). L03 type (c): what (same result → did nothing; different result → made a difference), why, the documented claim (Xavier's goldfish tank), transfer (no control group → cannot tell), a drawn results table with the subtraction, the term in its new role last. L04 type (b) scaffolded: the steps one at a time, the two predictions the logic makes, a bare "which plan is a pot test". L04b type (b) scaffolded → bare on fresh pot tests: two MCQs (why only water; which two things), then the gain and the loss produced (numeric), the judgement, a drawn table to its conclusion, the counterfactual, the argument sentence.

Which lessons could do without a video. None: L01's oak and flat ground, the inputs diagram building arrow by arrow; L02's rows of pots and the verdict strip filling; L03's ruler and the table filling row by row; L04's lid sliding on and five years of rain; L04b's scale readouts and the two bars drawn to one scale. Every entry is "video and article".

Words. "control group" is coined in L02 in the frame ("In an experiment, the group that gets none of the change is called the control group. It is kept for comparison.") and said in L03, the practice and the bank; recognised before it is typed; typed because both words are everyday and spellable at grade 5 (child spellings accepted), so RECALL_BY_MCQ is empty. "carbon dioxide" is Grade 4 Topic 34's term, recapped on first reuse ("a gas from the air called carbon dioxide") and never re-coined; "fair test" is Grade 4 Topic 2's, recalled in the recall bank. Grade 4's sentences reused word for word: "A plant does not eat: it makes its own food, and it makes that food in its leaves." / "The roots take in water from the soil, not food." / "The leaf uses the energy the light carries to make sugar from the water and the carbon dioxide." / "The sugar is the plant's food." (G4T34 L01); "Everything except the one thing you change must stay the same." (G4T02 L03); "No stuff was added … and no stuff was taken away." (G4T15, adapted by "to the pot except water"). A plant never "eats"; the soil is never its "food"; "weigh" is never said (the scale reads; find the mass); "fertilizer" is described before it is named ("a powder people sprinkle on soil"); "plant food" appears only as the bag's label, met by name as the belief's origin. "mostly" stays in the rule because the plant does take tiny amounts from the soil, said once in L01.

Questions for James

None. Every structural call here is one his earlier rulings already settle (the term, the opener forms, the split of the routine, the documented beliefs); nothing in it would change what a student learns by his answer the other way.

Per-lesson table

Design-check-sequence records

`

Skill: design-check-sequence @ cca3af6

Part: L01 — Type (c): the payoff is a fact with a why (the ground does not sink; the body is built from the sugar) and the register names a belief.

Plan: 3 what (3-opt), 1 why (3-opt), 1 claim (bare yes / no, LIF5 SD-2), 3 transfer (a pond plant; the water's source; a mowed lawn). Keys B C A B · n · A C A.

Belongs elsewhere: the four-option where-from items (bank type 1); the claim's reason (type 3); the multi (type 3).

Part: L02 — Type (a): naming the control group in a described test is a one-step sort on one feature (got none of the change?).

Plan: 2 which-group (2-opt), 3 yes / no, 1 none-of-them (3-opt), 1 drawn plan (2-opt), the term recognised last (3-opt). Keys B A y n y C B B.

Belongs elsewhere: four-group picks (bank type 4); both / neither (type 5); typed term (type 6, and the practice set).

Part: L03 — Type (c): the payoff is what the control group shows, and the register names the "spare" belief.

Plan: 2 what (3-opt), 1 why (3-opt), 1 claim (bare yes / no, SR5 SD-1 belief 2), 3 transfer (prose; a drawn table with the subtraction; no control group), the term in its new role last. Keys A B C n A B C A.

Belongs elsewhere: four-option result reads (type 7, 8); claim reason (type 9); chained items (type 9).

Part: L04 — Type (b): the set-up is an ordered routine; each step asked alone, then the logic's two predictions, then a bare plan judged.

Plan: 4 steps (3-opt), 2 predictions (3-opt), 1 bare plan (3-opt), 1 why-first (3-opt). Keys B B A C B C A A.

Belongs elsewhere: the one-step picks as four-option items (bank type 10 easy).

Part: L04b — Type (b): the argument is a routine on two readings; MCQs first, then the differences produced, then the argument on fresh tests.

Plan: 2 apply (3-opt), 2 numeric, 1 judge (2-opt), 1 drawn table (3-opt), 1 counterfactual (3-opt), 1 argument (3-opt). Keys C C · 38 · 40 · n A A C.

Belongs elsewhere: produced differences from drawn tables (type 10 hard); the counterfactual as four options (type 10).

`

Prior facts each check leans on, and where the recall sits: a plant makes its food in its leaves and the gas is carbon dioxide (LIF4-010) → recall bank before L01; everything else stays the same (SR4-003) and the fair test (SR4-005) → before L02; the mass of a mixture (MAT4-042) → before L04; breaking into pieces keeps the mass (MAT4-040) → before L04b. L03 leans on L02 and L04b on L04: the lesson immediately before, so no recall item (the 8 Oct ruling).

The key strings

Quiz keys per lesson: L01 B C A B · n · A C A — L02 B A · y n y · C · B · B — L03 A B C · n · A B C A — L04 B B A C B C A A — L04b C C · 38 · 40 · n · A A C. No position above 60 % within a lesson's same-size items; the vault's S116 (no four alternating keys across the topic's run) is clean after L04 q5 and L04b q1, q2, q6, q7 were reordered. Attached (video order): B A A A B A B B. Bank MCQ letters: A 14 · B 13 · C 11 · D 9. Claim verdicts across the topic's quizzes: no, no — two claim items on the register's two documented beliefs; the bank's two claims are "X is wrong. What is the correct reason?" items.

The rules applied (the brief's rulings 1–13 and the vault notes), with where each bites

Documented misconceptions used, and the beliefs deliberately not stated

Not stated: SR5 SD-1 beliefs 1, 3 and 4 (vague predictions; repeating is for mistakes; any science is an experiment) belong to Topics 3 and 19; LIF5 SD-2's energy-loop belief belongs to Topic 25. No belief is stated in lesson prose and then denied; each quoted claim is judged after the evidence, in its own words.

Checker results, every WARN answered

• build_tools_check_topic.py → 0 FAIL, 37 WARN; S46-COUNTS G5T24 B=0 C=0 D=0 (A = 0: no S12 FAIL); S30-COUNTS typed=4 fail=0 warn=0; S37 strand=0; S38 fragments=0; S44 negative_stems=0; S39 hits=0; S29 units_content=1 units_script=1 units_figures=8.
• S36 unusual words (27): fertilizer (described before it is named; the register's own example word), spray, scientist, sprout / sprouts / sprouted, spares / spare (Yusuf's own word in the documented belief), claim (the atom's verb: "back the claim"), shampoo, syrup, shinier, surrounded / surrounds, bamboo, basil, birch, olive, elm, mint, drier, mist / misted, app, normal, collect, wilting, argument (one bank stem; the atom's behaviour is to argue), sand — all kept: each is a plain US word a ten-year-old knows or is described where it first appears.
• S8 nine beats of three sentences: each is one picture state (the opening scene; a frame question with its verdict; a reading with its line; a recap opener before a two-sentence rule). Kept; every script is at or under 150 spoken words.
• S29 "2 kg" / "90,718 g" in screen cues and figure text: numerals with the unit symbol on drawings is James's rule (2 Oct); whether the on-screen unit is the symbol or the word is his open call — the symbol is used here, the prose says the word.
• build_tools_check_terms.py → hits {} (no A, B, C or D line).
• grade5/check_g5_articles.py → 0 FAIL; L01 353 and L02 405 words by its count (WARN above 300; see judgement call 2), L04 312.
• check_key_tells.py → KEY TELLS: 0 of 111 MCQs flagged (a: 0, b: 0, c: 0); exempt sets: yes / no, the multi's four things.
• check_svg_ids.py → PASS (figures read 42, ids found 0).
• check_lessons.py --adapter tools/check_adapter_g5.json → 0 FAIL; S118 0 FAIL. WARN lines (78): S11 "fb[key] is not empty" (the Grades 3-5 convention gives the right answer a confirming fb; the adapter reads the HS convention); S22 "a later stem carries an earlier key 'control group'" (15 — every L03 stem is about the control group, the term L02-q8 recognised; the adapter reads the topic's quizzes as one set) and 'nothing' / 'only water' (the same reading); S8 as above; S30 article-less forms (the accept lists carry them; the adapter's pattern differs). Two earlier FAILs were fixed, not waived: S135 (a worked line's result carries its unit — every "= N" line now reads "… = 38, so the tree gained 38 kilograms."; the local S38 refuses a bare "= 38 kilograms" as a sentence with no verb, so the two checkers meet on this form, recorded as a collision) and S116 (alternating keys across L04 / L04b, reordered).
• build_tools_names.py Xavier Yara Yusuf Xu → Xavier (INFO: Grade 3 Topic 30), Yara and Yusuf (INFO: Grade 3 Topic 4), Xu free; all from Topic 24's row; within the topic each name is one child.
• build_tools_build_topic_site.py → builds (5 lessons, 13 practice, 55 pp100, 42 figures). Screenshots looked at: the L01, L02 and L04b script routes, the L02 and L04b article routes, the L03 quiz route, the practice route (/tmp/g5t24_png/site_*.png).
• build_tools_blind_bundle.py … blind/2026-10-08 → bundled 104 mcq, 11 text (7 numeric), 1 multi; adversarial rows 4; held 116.
• build_tools_video_batch.py → 7 videos written to handoff/G5T24/videos/batch_G5T24.json; check_brief.py → every video ✓; the standing summary/protected note only (145 words: five rule sentences, one per lesson part). grep -ci strand → 0. The batch is NOT sent (Topic 24 is in James's Life review sample: his taps on the scripts page are the cue).

The recall bank (RECALL_BANK, 6 items)

Bank freshness, re-performed

Scene nouns of every lesson, quiz, attached, summary, practice and recall-bank item: oak, acorn, schoolyard, bag of plant food, bean plants, fertilizer, warm-cupboard seeds, children learning words with music, willow, scientist, small apple; sunflower, pine trunk, carrot, floating pond plant, mowed lawn, corn, maple; salt on ice, cucumber plants, dog food, grass with extra water, night-light, warm-water seeds, heated barn, toothpaste; tomato spray, puppies' chew toy, goldfish food, guppy flakes, runners' sports drink, pepper spray, shampoo; birch pot test, apple-tree table, 70-kilogram soil, tomato plant pot test, sunflower seedling plan; fence paint, peach tree; spinach, paper towels, flour in milk, bamboo, jar with pebbles, helmet pad, fan on basil, window cleaner, singing to cows, rose bush, 8,000-gram soil; bean windowsill, bouncing balls, kinds of soil, salt water, sawn log. Grepped against the bank's stems, options, feedback and whys (the script in this build's shell history): 0 hits after four re-scenes (a bush → a fig tree; a new drink → a new soil mix; lettuce under a shade cloth → cabbages; the gardener's fertilizer → sand; a lamp over pepper plants → mint; Yusuf's oak → an elm; a fish-food pair → rabbit food; singing to seeds → red pots and radishes; boat paint → boat wax). "fertilizer" survives in the bank only as an option word (PP-20) and in L04 / L04b items as the thing a pot test must not add.

Blind-text grep

The bundle's 13 [Drawing: …] lines were grepped for the verdict words (far too little, did nothing, mostly from, nothing, is the control group, none of the change): 0 hits. Every drawn table lists its cells as data ("two rows: the lemon tree, 4 kg before, 52 kg three years later; …"); a results table's row label "the control group (usual flakes)" names the group as the stem does, on items that ask what the change did, never which group is the control group; the four-pots item lists the labels 0, 1, 2, 3 spoons.

Figures (author self-check on the renders; the non-author review-figures pass is owed)

All 42 figures rendered through headless Chrome (/tmp/g5t24_png/); opened: plant_inputs (three arrows in, one out, every label by code), flat_ground (the trunk base on a level line, roots below, the two chips), pots_two_groups and pp_pots_four (rows of pots, labels under each), verdict_control (the question above five marks), scale_soil_before / scale_tree_after (the readout in numerals; the caption label was nudged clear of the box after the first render), lidded_pot, before_after_table, gain_loss_strip (the soil's bar a two-unit sliver beside the full-width tree bar), summary_still, and the built article routes for L02 and L04b at phone width. Kinds: 33 svg by code; 2 photo placeholders (the bag that says plant food; a small apple on a kitchen scale); 7 cartoon placeholders (Xavier under the oak ×2, Yara with her pots, the children with and without a speaker, Yara measuring with Yusuf, the scientist with the sapling, the scientist beside the grown tree). The scale is drawn as a modern flat-pan scale with a printed readout on a four-hundred-year-old test: a deliberate anachronism so the reading is exact and seen to change (the instrument rule), said nowhere in the student text. Units on drawings are symbols (g, kg, cm); prose says the word.

Names run

python3 build_tools_names.py Xavier Yara Yusuf Xu → Xavier INFO (Grade 3 Topic 30), Yara INFO (Grade 3 Topic 4), Yusuf INFO (Grade 3 Topic 4), Xu free. Cross-grade recurrence is information only (James, 6 Oct 2026); within Topic 24 each name is one child.

Sentence-atomise self-check (collapsed to the rows that changed)

Judgement calls and rule collisions

1. The capstone split (L04 / L04b). The brief names LIF5-011 as a long routine to split with a check between; the set-up (five steps and the logic the scale will answer) and the reading (two subtractions, the comparison, the argument) are each a script under 150 words. Both parts claim LIF5-011; the lexicon carries one rule sentence per part and the summary says both.
2. Article length. L02 at about 400 words carries the coin, a five-case sequence with both crossings, the sorting sentence and the closing return; L01 at about 340 carries the belief, the flat ground, Grade 4's three recalled sentences, the new step and the bag. Both are split by a subtitle or a figure every three to five idea lines and read in short runs on the built page. Splitting L02 would separate the coin from its run; recorded for the non-author read, not asked.
3. "Carbon dioxide" is said, not merely allowed once. The task brief allowed the word once as the test's word; Grade 4 Topic 34 coined it, so it is a recalled term here (recap on first reuse in L01, said in the inputs diagram and two L01 items, one recall-bank item). The mechanism stays unsaid.
4. The pot test told on the classic willow, with a modern scale drawn. The register names the willow experiment and its grams-against-kilograms shape; the numbers are van Helmont's in metric (soil 90,718 → 90,661 grams; tree 2 → 76 kilograms), the scientist is unnamed in student text ("a scientist, four hundred years ago"), and the scale is drawn with a printed readout so the number is seen to change. The 1 kilogram = 1,000 grams line is stated as a fact (Grade 4 mathematics), not taught.
5. Two checkers on one line. The vault's S135 wants a worked result's unit on its own line; the course's S38 refuses "76 − 2 = 74 kilograms" as a sentence with no verb. Every working line now reads "76 − 2 = 74, so the tree gained 74 kilograms." — the line is still shown under the sentence, the unit sits beside the result, and both sweeps pass. Recorded as a collision for the coordinator.
6. The control-group term in L04. S46 first warned that L04 and L04b never say the coined term. One honest line was added to L04 ("The pot test has no second group, so it has no control group. Instead, the soil before is compared with the same soil after."), which answers the question a student who has just met the term will ask; L04b stays silent on it.
7. "Control group" typed, not RECALL_BY_MCQ. Both words are everyday and spellable at grade 5; the accept list carries the article forms and the child spellings; near-miss grading covers one-letter slips. The typed item sits last in the practice set and in the bank's hard tier, after recognition items in L02, L03 and the practice set; it was removed from L02's quiz so that no quiz item follows a typed term item (the vault's S26 reads the topic's quizzes as one run).
8. Four quiz keys and the attached set rebalanced after the sweeps: L04 q5, L04b q1 / q2 / q6 / q7 and L02's attached question changed option order (recomputed strings above), never a key flipped.
9. "mostly" in the rule. A plant does take tiny amounts of other things from the soil (L01 says so once, at the bag); the rule keeps the register's "mostly" so the claim survives the chemistry a later grade adds.
10. Summary protected words at 145. The standing note; five rule sentences (L04's routine is two), no example re-told, no convenience term.

Owed

• The non-author reads: review-questions over every item bank; review-figures over the 42 renders and the built pages; review-accuracy (the willow numbers and their metric rounding; the claim that bean plants grow about 20 centimeters in two weeks; seeds sprouting sooner when warm; the pond plant, air plant and cactus transfers).
• The blind read of blind/2026-10-08/BLIND_BUNDLE_G5T24.md (keys sequestered beside it); zero unverified keys ship.
• Pictures: 9 photo / cartoon placeholders through David's pack at the picture stage — no question for James on this topic, so the pass can run once the text is reviewed.
• Videos: the batch at handoff/G5T24/videos/ waits for James's taps on the Life sample (Topics 24 and 25); nothing sent.
• The site deploy and the scripts page row.

v2 — the non-author review applied (8 October 2026)

Skill: review-lesson (Role B and the Role C fix pass) → review-questions, review-figures, review-accuracy @ cca3af6. The full record is REVIEW_G5T24_2026-10-08.md; this section lists what moved in the content.

• Must-fix (1): rule_anyway redrawn on three lines — its first line ran past both edges of the card on the render and on the built L03 page.
• Worth fixing (10, all applied): gain_loss_strip label and alt say "only a sliver" (the bar is visible); summary_still puts the grown tree on the 76-kg pan; the pots_half caption reads "Both groups got some of the change."; L01-q3's stem no longer carries q2's key; "got the most of the change" → "got even more of the change" (L02-q6 fb, PR-7 fb); PP-41's absurd distractor → "The wood is built from water alone, with nothing from the air."; PP-37's second-defensible distractor → "Sand soaks up the water before the roots can take it in."; PP-50 / PP-51 re-scened on their own table (pp_pear_table: 3 → 27 kg, 55,000 → 54,982 g; answers 24 and 18; type sheet's ⟂ list updated); "pot test" said in passing in L04 ("People call his test the pot test.", declared in TOPIC convenience_terms) and in script beat 1, with beat 2 trimmed (script 148); L01, L02 and L04 articles are two parts with a check between (James, G4T38, 2 Oct 2026) — L01 after "The sugar is the plant's food." (q2, q7), L02 after Yara's two cases (q1, q3), L04 after the three set-up steps (q1, q4); the items stay in the quizzes; L02's sequences case locations re-indexed.
• Left as recorded: the L04 / L04b split, the willow numbers (recomputed from the classic report), the typed term moved out of L02's quiz, the working-line form (the S135 / S38 collision), "mostly" in the rule, the modern scale. The docstrings' "no 'lift'" claim is corrected to "only in its literal sense" (L04b lifts the tree out of the pot).
• Counts now: figures 43 (34 svg, 2 photo, 7 cartoon); articles 353 (191 + 162) / 405 (224 + 181) / 293 / 319 (179 + 140) / 293; scripts 148 / 150 / 147 / 148 / 149; bank 55 unchanged in shape (MCQ keys A 14 · B 13 · C 11 · D 9); no key letter moved.
• Gates: check_topic 0 FAIL, S46 B = 0 C = 0 D = 0; check_terms hits {}; check_g5_articles 0 FAIL; key tells 0 of 111; check_svg_ids PASS (43); check_lessons 0 FAIL, S118 0 FAIL; blind bundle re-cut at blind/2026-10-08-reviewed/ (116 held; [Drawing: …] lines free of verdict words; CHANGED_IDS.md beside it, ids only); batch rebuilt, check_brief.py every video ✓ (the standing summary note only). Not sent; nothing committed.
• Owed now: the blind read of blind/2026-10-08-reviewed/BLIND_BUNDLE_G5T24.md by two cold readers (keys sequestered beside it); then pictures (9 placeholders) and James's taps on the Life sample before the batch goes.