Skill: write-like-james @ efd8fba · the build: write-lesson, write-mcqs, write-video-script, review-lesson (two blind reads and a re-read), review-figures, review-accuracy @ b0fa825

Topic 18 "How igneous and sedimentary rock form" — your read (2 October 2026)

Skim first

The lessons in teaching order, each with its one rule sentence; then the intro script; then the question types. The review page follows.

The lessons and their rules

Lesson	Atom	Rule sentence
L01 What igneous rock looks like	EAR4-007	Igneous rock is hard. Grains locked together, or smooth like glass. It never has layers.
L02 How igneous rock forms	EAR4-008	Melted rock cools and hardens. Fast cooling: tiny grains or glass. Slow cooling: big grains.
L03 What sedimentary rock looks like	EAR4-009	Grains or pieces pressed together. Often in flat layers. Sometimes with shells inside.
L04 How sedimentary rock forms	EAR4-010	Small pieces settle in layers. More layers press down. The pieces stick together, over a very long time.

The intro script (116 words, about 58 s)

Here's Jace at the edge of the Grand Canyon with his family. The far wall is striped: flat layers of red and tan rock, one on top of another. In his pocket is a black rock he picked up in Hawaii last summer. It has no stripes at all: it's dark and smooth, with grains too small to see. Two rocks, and they could not look more different. Were the canyon's stripes painted on? And was Jace's black rock ever hot? To find out, we will look at rocks made from melted rock, and rocks made from sand, mud and shells pressed together. Okay, so are you ready to find out how a rock is made?

The question types (from `TYPES_G4T18.md`)

Type 1 (easy, medium · MCQ, the family set; four drawn on family cards · 007 / 009): a rock described by its clues (grains, layers, also): which family is it in — igneous rock / sedimentary rock / a mineral / soil
Type 2 (easy, medium · MCQ · 007 / 009): which clue says a rock is (or is not) in a family; which second clue would settle it — four clues
Type 3 (easy, medium · MCQ, the grains set or the cooling set; two drawn close-ups · 008): from how the melted rock cooled, what the grains are like; from a close-up, how it cooled; where big-grained rock cooled — the grains set / the cooling set / four places
Type 4 (medium, hard · MCQ · 008 / 010): the correct reason for a stated feature (big grains, glass, layers, shells inside, limestone not sandstone) — four single-statement reasons
Type 5 (medium, hard · MCQ, the steps set or the rocks set · 010): which step comes first / next / last; which rock the named pieces make; where the pieces settle; what came before the pressing — the steps set / the rocks set / four places
Type 6 (hard · MCQ · 008 / 010): a named child's claim is wrong: what is the correct reason (the four catalogued beliefs, plus the one-settled-layer overreach); on the two always-there claims the ask is what the thing WAS before it was rock — four single-statement reasons, or four noun phrases
Type 7 (hard · MCQ · 008 / 010): from how a rock formed, which clue it will show, or which clue it will NOT have (two ideas chained) — four clues
Type 8 (hard · MCQ ×2 keyed on sediment; typed ×1 · 009): the settled pieces named from their description (by MCQ); the bands named in one word (typed) — sedimentary rock / sediment / soil / igneous rock; accept: layers, layer, a layer, the layers, flat layers and ten child spellings; stripes, bands, sheets and lines each get a wrong-term line
Type 9 (hard · MCQ keyed on the term · 007 / 009): the family named from its definition, no clue word in the stem — igneous rock / sedimentary rock / sediment / a mineral
Type 10 (hard · multi-select · 008 / 009): the TWO sedimentary clues; the TWO rocks that formed from melted rock — four clues / four rocks with what each is made of

Four lessons: what igneous rock looks like (hard, grains locked together or smooth like glass, never in layers); how it forms (melted rock cools and hardens; fast cooling gives tiny grains or glass, slow cooling deep underground gives big grains); what sedimentary rock looks like (grains or pieces pressed together, often in flat layers, sometimes with shells inside); and how it forms (small pieces settle in layers, get pressed together and stick, over a very long time). Every key agreed by two cold readers and a third on the rebuilt bank; the guidance-and-accuracy review, the figure review and the re-read's polish are applied. Pictures are placeholders until the picture pass.

The questions the author put to you (built as recommended)

1. Is "sediment" a third coined term in this topic?

The topic coins "igneous rock" and "sedimentary rock". The loose pieces that settle to the bottom of water need a name too, or the forming story says "the small pieces" every time.

Topic 20 and the Grade 5 STAAR use the word; it is recognised in an MCQ, never typed.

Built: explained first ("small pieces that settle to the bottom of water"), then named; recalled by MCQ.

Recommendation: yes.

Yes / explain it, no name

2. Does the topic say "lava" only, never "magma"?

Grade 3 taught "lava" for the melted rock that comes out of a volcano. The register's text for this atom carries only lava.

"Melted rock deep underground" carries the underground case in plain words, so nothing said is false; Grade 6 to 8 names magma.

Built: lava only; the cross-section's "lava" label sits on the flow.

Recommendation: lava only.

Lava only / add magma

3. Should the atom EAR4-007 say "no layers inside a piece of it" rather than "never has layers"?

The register's wording is "never has layers", and the page keeps it. Some igneous rock does show bands (lava that flowed in sheets; ash that fell in layers).

"No layers inside a piece of it" is true of every rock a child will hold, and it is the register's line to change, not this page's.

Built: the register's wording, on every line and item.

Recommendation: soften the atom.

Soften / keep

4. Does Topic 17 gain a scope line about glassy obsidian?

Topic 17 says a rock is made of grains of minerals. Obsidian has no grains a child can see, even up close.

This topic says "no grains you can see, even up close: smooth like glass", which fits both. Topic 17 is not edited.

Built: Topic 18's wording; Topic 17 untouched.

Recommendation: add the scope line.

Add / leave

5. Granite is drawn two ways across Topics 17 and 18: a scope line, or extend Topic 17's drawer?

Topic 17's grain box draws granite as speckles on pink. Topic 18's card draws a locked tiling of pink, white, gray and black, because the sort rests on "locked together".

A child who meets both sees two pictures of one rock. One depiction per thing is the standing rule.

Built: Topic 18's tiling; Topic 17 not edited.

Recommendation: extend Topic 17's drawer so granite is the one tiling in both.

Extend Topic 17's drawer / a scope line

The lessons

6. L01 What igneous rock looks like — Miss Lopez's tray: Ian's speckled granite and Isaac's glassy obsidian look nothing alike, and she says they are one family. Basalt and pumice join them; Jace's striped rock does not. The name comes last. page

7. L02 How igneous rock forms — Jace's black rock from a Hawaii lava flow was once a liquid. Melted rock pours out as lava, cools in the air and hardens. Fast cooling gives tiny grains or glass; slow cooling deep underground gives big grains. page

8. L03 What sedimentary rock looks like — Jace's striped rock is sandstone and Ian's new rock is limestone full of shells. Shale and conglomerate join them; granite does not. Settled pieces are sediment; the name comes last. page

9. L04 How sedimentary rock forms — sea shells inside a rock on a Texas hill, far from the sea. Miss Lopez's jar settles into flat layers. Pieces settle, more layers press down, and over a very long time the pieces stick. page

Applied without asking

Three reviews and a re-read changed the bank more than the lessons. Twelve bank items and two practice items were quiz scenes with a new letter or place, and in the hard tier the right answer was the longest option in nine items; every one now has a fresh rock, child or move, every hard key is one short clause, and no key is the longest option. The re-read then found five more bank items still on a lesson scene with a new place (Jace's lava rock, the jar, Ian's Texas hill, the sandstone cliff) and two claims near-copied from the quizzes; those are re-scened or ask a different thing. The two facts the quizzes tested that only the article said (the jar's layers stay loose; a hardened rock's grains stop getting bigger) are spoken in the L04 and L02 videos too. Figures no longer carry the key by colour on the fast-or-slow items, Jace's rock is drawn one way everywhere, five item cards are no longer twins of teaching cards, and the settled jar shows its pebbles. Wording: "pressed together in sand" for conglomerate (not "stuck"), "no grains you can see, even up close" for obsidian, "no longer hot from the lava" for the cool-rock claim, one TRUE claim added so the quiz claims are three of eight true, and the typed "layers" item corrects stripes, bands, sheets and lines each with its own line.

Also on the site

Topic intro and poll · summary and three questions · mixed practice · the PP100 bank · the type sheet

After you answer

Nothing on the page changes if you keep 1 and 2; 3, 4 and 5 are register and Topic 17 changes, not this page's. Either way the picture pass starts (14 photos, 2 cartoons), and the six-video batch goes on your word.

Build record (the structural decisions, the counts, v2 and v3)

Skill: course-build → write-lesson (L01 and L03 concept introductions in the fact form, each with one boundary near miss, the term coined last; L02 and L04 process lessons, L04 the capstone) → write-example-sequence, design-check-sequence, sentence-atomise, write-explanation, write-mcqs (quizzes, attached questions, practice, the PP100 bank and its type sheet); write-video-script (intro, four lesson scripts, summary; the elementary micro-lesson anatomy); review-figures (author self-check on fourteen renders, then the non-author pass below) @ b0fa825

Grade 4 Topic 18 — How igneous and sedimentary rock form — build record (v1, 2 October 2026)

What this is, in plain words

Four short lessons, one per atom, that teach the first two rock families the way James ruled on 25 September: what a family looks like FIRST, then how it forms, one family at a time. A child meets one running scene: Miss Lopez's class and its tray of rocks. First the child sees that speckled granite and glassy obsidian have something in common — both are hard, their grains are locked together or they are smooth like glass, and neither has layers — and only then meets the name igneous rock (L01). Next Jace's black rock from Hawaii "was once a liquid": melted rock pours out of a volcano as lava, cools in the air and hardens; lava in the open air cools fast (tiny grains or glass), melted rock deep underground cools slowly (big grains) — the one pair of cases the atom allows (L02). Then Jace's striped rock and Ian's rock full of shells have something in common too — pieces pressed together, often in flat layers — and the child meets sediment (the small pieces that settle) and sedimentary rock (L03). Last, the capstone: how did sea shells get inside a rock on a Texas hill? Miss Lopez's jar shows pieces settling in flat layers; in the sea, more layers press down, and over a very long time the pieces stick together into rock; the topic closes on the two-family table and an igneous-or-sedimentary sort (L04). Metamorphic rock and the three-family sort wait for Topic 19; weathering, erosion and deposition by name wait for Topic 20.

Counts. 4 lessons · 4 scripts (149, 150, 147, 150 spoken words) + intro (116 words, 3 question marks) + summary (96 words) · 4 attached questions + the intro's poll + 3 summary questions · 43 quiz items (9 + 11 + 11 + 12) · 13 practice items (12 MCQ, 1 written answer; three Grade 3 retrievals; every coined term recognised then recalled by MCQ) · PP100 bank 53 items (50 four-option MCQs, 1 typed, 2 multi-selects; EAR4-007 10, EAR4-008 15, EAR4-009 12, EAR4-010 16; easy 18, medium 17, hard 18; MCQ keys A 11 · B 14 · C 14 · D 11) · 59 figures (43 SVG by code, 14 photo placeholders, 2 cartoon placeholders) · checker 0 FAIL, 24 WARN (every WARN answered below) · check_g4_articles.py 0 FAIL (articles 299 / 281 / 298 / 306 words — the L04 WARN answered below) · names register exit 0 · page builds (/tmp/g4t18_test.html: 4 lessons, 13 practice, 53 pp100, 59 figures) · video batch built, check_brief.py clean except the INTRO's standing R79 hit, grep -ci strand 0 · NOT sent.

Structural decisions (in place of a blueprint)

Lesson	Title (the skill in student terms)	Headache scene	Example sequence	Check type	Video?
L01 EAR4-007	What igneous rock looks like	Ian picks a speckled rock and Isaac a smooth black one from the class tray; Miss Lopez says they are in the same family — what could they have in common?	the FACT form (James's Grade 3 Topic 3 test: no non-example differs from "hard, locked or glassy, no layers" by one feature only), one frame — "<Rock> is hard. Its grains are locked together / It is smooth like glass. It has no layers.": E granite → E obsidian → E basalt → E pumice → NE Jace's layered rock (hard and grainy, but layered: the one near miss) → the name coined last → rule → closing return to the two rocks	(a) fluency: seven is-it-igneous items (three on drawn family cards), one which-clue item, one term recognition item	video and article
L02 EAR4-008	How igneous rock forms	Jace's black rock from Hawaii; Miss Lopez says it was once a liquid; Isaac says rock is hard, so that cannot be right	process: recall (lava; a liquid cools into a solid) → melted rock pours out → cools in the air → hardens (step cards) → the rock cooled long ago, so it is as cool as the ground → the ONE pair: fast cooling in the open air (basalt, Yellowstone's obsidian) against slow cooling deep underground (granite), drawn as a labelled cross-section and a two-row table → rule → the simplification named	(c) what / why / misconception / transfer, two Grade 3 recalls first, a fast-or-slow fluency run in the middle	video and article
L03 EAR4-009	What sedimentary rock looks like	Jace's striped rock and Ian's new rock full of shells; Miss Lopez says these two are in one family too	the FACT form, L01's twin: E sandstone → E limestone → E shale → E conglomerate → NE granite → "sediment" explained as the small pieces that settle, then named → "sedimentary rock" coined last → rule → closing return. The painted-stripes belief is met by "the layers go right through the rock, so a broken edge shows them too"	(a) fluency: seven is-it-sedimentary items (three on drawn cards), one which-clue item, Ian's painted-stripes claim, two term items last	video and article
L04 EAR4-010	How sedimentary rock forms	Ian's limestone from a Texas hill, far from the sea, full of sea shells — how did they get inside?	process, the capstone: Miss Lopez's jar (shaken, then still: pebbles, sand, mud in flat layers) → in the sea, pieces settle in a layer → more layers press down → over a very long time the pieces stick → the pieces-to-rock table → the jar's limit said plainly → closing return (a shallow sea once covered that part of Texas) → the two-family table → rule	(c) what / why / misconception / transfer, one Grade 3 recall first, closed by four igneous-or-sedimentary sort items (two from looks, two from how it formed)	video and article

Why these calls: the register's cross note gives the order (what it looks like, THEN how it forms, one family at a time), so L01 never says how granite formed and L03 never says how sandstone formed; the two "looks like" lessons are fact-form runs because the only honest non-example is the other family's rock, which is a near miss (hard and grainy) differing by the locked / pressed grains and the layers; the two "how it forms" lessons open on a real question a child feels (how can a hard rock have been a liquid? how did sea shells get inside a rock on a hill?) and build the steps one at a time; L02 carries the one pair of cases (fast / slow) because the atom's behaviour names tiny-or-glassy against big grains and the pair is what makes "it cooled" more than a word; L04 is the capstone, so it alone carries the two-family table and the sort, and the sort stays two-way (the three-family sort is Topic 19's). "Magma" is not said: the register's text carries "lava from a volcano cooling into black rock" and nothing for the underground case, so the plain phrase "melted rock deep underground" carries it. Every lesson gets a video: L01 and L03 need the rocks seen side by side with their cards filling, L02 and L04 are processes a child should watch happen. The intro is one scene (Jace at the Grand Canyon rim with his Hawaii rock in his pocket): the striped wall and the smooth black rock are the topic's two families seen before they are named, and the two felt questions are the misconceptions the lessons close (painted stripes; a hot rock).

Grade 3 doorways, recalled by one item and one line each, never re-taught: Topic 24 (a volcano is an opening in the ground where melted rock from deep inside Earth comes out; the melted rock that comes out is called lava) — L02's article lines and L02-q1, PR-1; Topic 21 (when a liquid cools down enough, it turns into a solid) — L02's line and L02-q2, PR-2; Topic 26 (the bits of rock in soil come from bigger rocks slowly breaking apart) — L04's line and L04-q1, PR-3. Topic 17's register words are leaned on, not re-taught: "grains", "grains too small to see", "shiny", "gritty"; Topic 17's description card (COLOR / GRAINS / FEEL / SHINE) and this topic's family card (GRAINS / LAYERS / ALSO) are different things and never share a page.

The simplification, named (once, in L02): "We made one simplification here. We say 'a very long time' and give no number of years." L04 says "over a very long time" in its rule and gives no number either; its article does not repeat the note (it was cut to keep the capstone under 310 words).

Questions for James

Does "sediment" stand as a third coined term, or should the topic coin only "igneous rock" and "sedimentary rock"? Context: the brief names two coined terms and asks that "sediment" be explained as "the small pieces that settle" before it is named; L03 does exactly that in one line ("When small pieces settle to the bottom of water, we call them sediment"), which makes "sedimentary rock" mean something instead of being a sound. Recommendation: keep it — it is the word the family's name is built from, L04 reuses it with a one-line recap, and a Grade 4 child meets it everywhere in the Earth and Space topics ahead. Built with it (coined in L03, in terms and coined_at, recalled by MCQ because the checker's hard-to-spell list carries it).
Should "lava" be said at all, or should the whole topic stay with "melted rock"? Context: Grade 3 Topic 24 coined "lava" ("the melted rock that comes out is called lava"), so L02 recaps it in one line and uses it for the open-air case; the underground case stays "melted rock deep underground" because the register gives no word for it. Recommendation: keep lava as the Grade 3 word it is and never say magma in Grade 4. Built that way.

Design-check-sequence records

Skill: design-check-sequence @ b0fa825

Part: L01 (EAR4-007)

Type: (a) one-step categorisation — for a described or drawn rock, say whether it is igneous rock (yes / no); the answer is one of two labels.

Plan: Rock J card (y) → Rock K card (n) → Rock L card (y) → hard gray rock, tiny locked grains (y) → pale rock in thin flat layers (n) → brown rock of sand grains that rub off, in layers (n) → dark green glassy rock (y) → which of these does igneous rock never have (3 opt, 2) → the term item last (3 opt, 0; chosen, never typed). Two-option string A B A A B B A; three-option 2 0.

Belongs in practice/test instead: the four-way family set; the term recalled from its whole definition (PP type 9).

Skill: design-check-sequence @ b0fa825

Part: L02 (EAR4-008)

Type: (c) concept understanding — the lesson's payoff is a process (melted rock cools and hardens) with one pair of cases (fast / slow), and the register's literature names three beliefs.

Plan: recall lava (3 opt, 0) → recall a liquid cools into a solid (3 opt, 1) → the WHAT: lava in Idaho after it has cooled (3 opt, 2) → the WHY: the correct reason a lava-field rock's grains are tiny (3 opt, 1) → fast or slow on a drawn tiny-grain box (fast), a drawn big-grain box (slowly), a described glassy rock (fast) → Isaac's TRUE claim: as cool as the ground (yes) → Jace's claim: rocks were always here (no) → Ian's claim: grains kept growing like a plant (no) → TRANSFER: lava into the sea cools in minutes (tiny, or glass). Two-option string A B B A B B A (first option = fast / yes); three-option 0 1 2 1.

Belongs in practice/test instead: the four-option cooling and grains sets; the claims as reason items (PP type 6).

Skill: design-check-sequence @ b0fa825

Part: L03 (EAR4-009)

Type: (a) one-step categorisation — for a described or drawn rock, say whether it is sedimentary rock (yes / no).

Plan: Rock M card (y) → Rock P card (y) → Rock N card (n) → gray rock, tiny locked grains, no layers (n) → brown rock of pebbles stuck in sand, in layers (y) → pale rock with layers and shells (y) → black glassy rock (n) → which of these does sedimentary rock often have (3 opt, 1) → Ian's painted-stripes claim (no) → term: the family of pieces pressed together in layers (3 opt, 1) → term: the settled pieces named (3 opt, 0). Two-option string A A B B A A B B; three-option 1 1 0.

Belongs in practice/test instead: the family set; the terms from their whole definitions; "layers" typed (PP type 8).

Skill: design-check-sequence @ b0fa825

Part: L04 (EAR4-010)

Type: (c) concept understanding — the payoff is a three-step process (settle, press, stick) and the register's literature names three beliefs; the sort that closes the topic is a one-step categorisation appended as four fluency items.

Plan: recall where soil's bits of rock come from (3 opt, 0) → the WHAT on a drawn settled glass: which pieces settled first (3 opt, 2) → the first step (3 opt, 1) → the WHY: the correct reason a sandstone cliff has layers (3 opt, 0) → Isaac's claim: the rock grew its shells (no) → Jace's claim: the layers were always there (no) → Ian's TRUE claim: loose layers need many more layers and a very long time (yes) → TRANSFER: pebbles and sand on a river bed → which rock (3 opt, 2) → the sort: locked grains, no layers (igneous); shells and layers (sedimentary); settled and pressed (sedimentary); melted rock cooled (igneous). Two-option string B B A A B B A (first option = yes / igneous rock); three-option 0 2 1 0 2.

Belongs in practice/test instead: the steps set and the rocks set; the claims as reason items; formation → clue chains (PP type 7).

The key strings

Attached questions in video order (L01, L02, L03, L04, summary 1–3): A B B A A B B — B 4 of 7 (57 %).
Lesson quizzes, two-option items (first option = yes / fast / igneous rock): L01 A B A A B B A · L02 A B B A B B A · L03 A A B B A A B B · L04 B B A A B B A — no four in a row alternate; no position above 60 % within a lesson.
Three-option items: L01 2 0; L02 0 1 2 1; L03 1 1 0; L04 0 2 1 0 2; practice three-option 0 1 0 0 2 0 1 1 2, two-option A B B.
PP100 MCQ keys: A 11 · B 14 · C 14 · D 11 (50 MCQs); the typed item and the two multi-selects carry no letter.

The example sequences, written out

L01 (one frame; the rule is induced after, so the frames carry no "because"):

Granite is hard. Its grains are locked together, with no gaps. It has no layers.
Obsidian is hard. It has no grains: it is smooth like glass. It has no layers.
Basalt is hard. Its grains are too small to see, but they are locked together. It has no layers.
Pumice is glassy and full of tiny holes. It has no layers.
Jace's rock has flat layers, and you can rub sand grains off it. It is not in the family.
Rule: Igneous rock is hard, with grains locked together or smooth like glass, and it never has layers.

L03 (one frame; the non-example is L01's granite):

Sandstone is made of grains of sand pressed together. You can rub a grain off. It has flat layers.
Limestone has grains too small to see, with pieces of shell pressed in. It often has flat layers.
Shale is gray mud pressed into thin flat layers. It splits into flat sheets.
Conglomerate is round pebbles stuck together in sand, in flat layers.
Granite's grains are locked together; you cannot rub one off; it has no layers. It is not in the family.
Sediment: small pieces that settle to the bottom of water. Rule: Sedimentary rock is made of grains or pieces pressed together, often in flat layers, sometimes with shells inside.

L02 (the process, then the pair): melted rock deep inside Earth → out of the opening as lava → cools in the air → hardens into basalt (Jace's rock; cooled long ago, as cool as the ground) · fast in the open air → tiny grains or glass (basalt, obsidian) · slow deep underground → big grains (granite) · once hardened, the grains stay as they are. Rule: Igneous rock forms when melted rock cools and hardens. Fast cooling gives tiny grains or glass; slow cooling gives big grains.

L04 (the process, worked twice): the jar — shaken, still, pebbles then sand then mud in flat layers (sediment) · the sea — sand and mud settle in a layer; more layers press down; over a very long time the pieces stick (sand → sandstone, mud → shale, shells → limestone, pebbles and sand → conglomerate); the jar's layers stay loose because nothing has pressed them for a very long time · Ian's rock — a shallow sea, shells settling with tiny pieces, layer on layer, pressed into limestone, the sea gone and the rock a hill. Rule: Sedimentary rock forms when small pieces settle in layers, get pressed together and stick, over a very long time.

The rules applied

Rule (vault note)	Where it bites here
What a thing IS before what it DOES; one family at a time (James, 25 Sep; the atoms' cross note)	L01 looks → L02 forms → L03 looks → L04 forms; L01 and L03 never say how the rock formed
Explain, then label; coin as "when X, we call it Y"	igneous rock coined at the end of L01 after four examples and a near miss; sediment explained as "the small pieces that settle" then named; sedimentary rock coined last in L03
A coined term is used from then on, with a one-line recap on first reuse	L02 opens on the Grade 3 term lava with its recap line; L03 and L04 say igneous rock for granite with no scaffold phrase; L04 recaps sediment in one line ("Small pieces that settle like this are sediment")
Lesson text states the fact, never the topic that taught it	"A volcano is an opening in the ground where that melted rock comes out." / "When a liquid cools down enough, it turns into a solid." / "The sand and mud are bits of rock that broke off bigger rocks." — no "in Grade 3 you saw"
A non-example is a near miss, never an absurd case	L01's near miss is a hard, grainy rock with layers; L03's is granite; no absurd case anywhere
A headache scene could really happen there	a class tray of rocks; a boy's holiday rock; a rock from a Texas hill with shells in it; a teacher shaking a jar of sand, mud and water
A question in an article is answered in the next sentence, verdict first	L02: "Is she right?" → "Yes, she is right." (script); the article's "How can a hard rock have been a liquid?" is answered by the subtitle's first line
Every sentence a student reads has a subject and a verb (S38)	0 fragments; "And now here's basalt" and "And now here's shale" after the checker caught two clipped beats
Connectives carry cause; a turn opens with But	"You cannot rub a grain off, because the grains are locked together"; "But some melted rock never reaches the air"; "The rock cooled long ago, so today it is as cool as the ground around it"
The positive word	"Sea creatures died"; "the sea is gone"; "stayed loose"; "it never has layers" is the rule's own form (the atom's "never in layers")
No figurative verb a young reader takes literally; a rock never grows	"end up big", "stay as they are" — never "the grains grow"; the one "grows" is inside the claims ("the way a plant grows"), corrected in the why
Say the specific thing; pronouns point at one named thing	"Jace's rock", "Ian's rock", "the melted rock", "the pieces" repeated as nouns; "it" never carries two referents
Units as words; numerals on screen	no numbers in the student text beyond "one" (the simplification: no years)
Biggest / smallest for size only (S39)	0 hits; "bigger grains", "big grains" are size
Grade-level words, every odd word questioned (S36)	14 words listed and answered below; "underneath" → "below", "swirl" → "spread all through", "quarry" → "dug up" swapped before this record
No second everyday reading	"stick together" (not "set"); "spread all through the water" (not "swirl"); "lets it stand" (not "sits")
An attached question stands alone (S27)	each attached stem names its own scene (a hard glassy rock; melted rock out of a volcano; sand grains in flat layers; sand on a sea floor; lava on a hillside; pressed mud; melted rock deep underground)
Recognition before recall; typed only where the word is spellable (S33 / S34 / S35)	igneous rock, sedimentary rock and sediment are in RECALL_BY_MCQ (eleven and sixteen letters; the checker's hard-to-spell list carries sediment), so every recall is an MCQ among sisters; the bank's one typed key is "layers"
A quiz climbs recognise → apply → produce (S32)	fluency cards first, described cases, the which-clue item, the claim, the term items last in L01 and L03; recalls → what → why → fluency → claims → transfer → sort in L02 and L04
An option never bundles a verdict with its reason	every claim in a quiz or the practice is bare yes / no with the correction last in the why; every bank claim is "X is wrong … What is the correct reason?" with four single-statement reasons; one drafted bank item with reasoned options was recast
Fresh instances (S10; the bank-freshness grep)	0 S10 warnings; quiz rocks are letters J–P or described in words; the bank's rocks are letters R–V or fresh US places (Colorado, Oregon, Washington, Utah, Florida, Kentucky, Maine, Ohio, Idaho, Georgia, New York, Wisconsin, Kansas); the scene-noun grep below
A real thing survives the claim made about it	granite (locked grains, no layers, pink-white-gray-black), basalt (dark gray, grains too small to see, Hawaii's lava-flow rock), obsidian (black volcanic glass; Yellowstone's Obsidian Cliff), pumice (pale, full of holes, glassy), sandstone (pressed sand, layers, grains rub off; the Grand Canyon's walls carry sandstone layers), limestone (shell pieces; the Texas Hill Country's limestone with fossil shells; a shallow sea once covered that part of Texas), shale (mud in thin layers, splits into sheets), conglomerate (rounded pebbles in sand) — each checked
One depiction per thing, reused everywhere	one family card drawer, one close-up drawer with seven named kinds, one step-card drawer, one table drawer with a measured-width guard, one jar drawer (lidded in the lesson; lidless as the item's glass); Grade 3 Topic 17's chip and SVG helpers imported
Code draws what must be exact; David's pack renders what must look real	43 SVG figures (cards, close-ups, step cards, cross-section, tables, jar, rule cards); 14 photo placeholders for the eight real rocks, flowing lava, a granite hillside, the canyon wall and the Texas hill; 2 cartoon placeholders for the class scenes; no person or hand drawn as a glyph
An item figure's blind text is data, never the answer	every item card's blind text lists its three rows and the close-up as data; the close-ups list cell counts and sizes; the glass lists its layers bottom to top with their heights; grep of the bundle's `[Drawing: …]` lines for igneous / sedimentary / cooled fast / cooled slowly / settled first / melted / sediment → 0 hits
Summary opens "Here's everything we found out about …" and recaps points, not examples	the opener, then the four lexicon sentences with "First / Next / Then / And last", 96 words
Intro ≤ 120 words, ≤ 3 question marks, coins nothing, one scene, map line, ready line + poll	116 words, 3 question marks (the limit — answered below), no "is called", "To find out, we will look at rocks made from melted rock, and rocks made from sand, mud and shells pressed together.", "Okay, so are you ready to find out how a rock is made?"
Closing bullets a few words each; summary bullets ≤ 4	every lesson carries two or three short bullets; the summary carries four chips, one per lesson
Name the simplification you made, two sentences, taught terms only	L02's last two lines
US spelling (S14)	gray, color; 0 warnings; the figures' "grey" renamed "gray" by perl with `\b`, verified by grepping for the OLD word (0 hits)
A character's name belongs to one topic	Miss Lopez, Ian, Isaac, Jace — the Topic 18 row of `NAMES_POOL.md`; `build_tools_names.py Lopez Ian Isaac Jace` → all four free, exit 0

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

Belief	Source	Where it is confronted	Claim items
Rocks were always here; they do not form	Happs 1982 (children's ideas about rocks and minerals); Kusnick 2002, J. Geoscience Education 50 (children's ideas about rock formation)	L02's whole story and "Jace's rock was a liquid once"; L04's "each layer settled as loose pieces"	L02-q9 (Jace: no); L04-q6 (Jace: no); PP type 6 ×2 (the lava-field rock; the river-cliff layers)
Igneous rock is hot (it came from a volcano, so it stays hot)	Kusnick 2002	L02: "The rock cooled long ago, so today it is as cool as the ground around it"; the script's "so now it's as cool as the ground"	L02-q8 (Isaac, TRUE: as cool as the ground — yes); PR-8 (Isaac, hot to touch: no); PP type 6 (Isaac's Idaho lava-field rock)
Sedimentary layers are painted stripes on the outside	Ford 2005, Science Education 89 (children read layers as surface colour)	L03: "The layers go right through the rock, so a broken edge shows them too"; L04's reason items	L03-q9 (Ian: no); PP type 6 (Ian's painted cliff); PP type 4 (the New York shale cliff's reason)
A rock grows like a plant	Happs 1982 ("rocks grow")	L02: "Once the rock has hardened, its grains stay as they are"; L04: the shells settled, they were not grown	L02-q10 (Ian: no); L04-q5 (Isaac: no); PP type 6 (Ian's growing grains)
The classroom jar's layers will turn into rock	Kusnick 2002 (the demonstration read as the whole story)	L04: "Turning layers into rock takes many more layers on top, and a very long time, so the jar's layers stay loose"	L04-q7 (Ian, TRUE: yes); PP type 6 (Isaac's overnight tub)

Checker: 0 FAIL, 24 WARN (`python3 build_tools_check_topic.py grade4/topic18_build/content/G4T18.py`)

S8 (13) script beats of three or four sentences — kept, each beat is one screen state (the elementary scripts' shape, as in the demo and Topics 1 and 13); the four-sentence beats are the four scene-setting first beats and L04's sea beat.
S36 (11 words, after swaps), each answered: granite, obsidian, basalt, pumice, shale, conglomerate — the topic's own rocks, each shown as a photo placeholder and a family card and named in the register's examples (fine); glassy — the plain form of "smooth like glass", used where "glass" would misread as a window pane (fine); splits / split / splitting — the everyday verb for what shale does (fine); shallow — the only honest word for the Texas sea (fine); gritty — Topic 17's FEEL word for sandstone (fine); stripe — the misconception's own word, in the claim items only (fine); simplification — James's standing note wording (fine). Swapped before this record: "underneath" → "below", "swirl" → "spread all through", "quarry" → "dug up". The allowlist is Grade 3's (the tool note the demo record carries still stands).
Clean: S1–S7, S9–S14, S21, S22, S25, S27–S35, S37, S38, S39. The two earlier FAILs (two clipped script beats; two long fb) and the earlier WARNs (a stem carrying its key word "the gravel"; a negative stem over negative options on L01's attached question; two bank items word for word equal to practice items; a typed "sediment") were fixed, not silenced.
check_g4_articles.py: 0 FAIL; L04's article is 306 words (WARN above 300) — kept at that length because the capstone carries the jar scene, the sea story, the closing return and the two-family table, and the WARN is six words over Grade 3's approved median of 308; two lines (the second simplification note) were cut to get here.
check_brief.py: every lesson and the summary clean; the INTRO's standing R79 closing-bullets hit accepted (an intro has no rule slide), and its "3 question marks — at the limit" note answered: the two felt questions are the topic's two misconceptions and the third is the ready line.

Bank freshness: the scene-noun grep (re-performed before this record)

Scene nouns of every lesson, quiz, attached and summary item: tray · jar · Hawaii · Texas · canyon / Grand Canyon · Yellowstone · Kilauea · bucket · pocket · thumb · classroom · desk · lava flow · Austin · hill in Texas · Idaho lake. Grep over G4T18_pp100_bank.py and PRACTICE: tray 0 / 0 · jar 0 / 0 · Hawaii 0 / 0 · Texas 0 / 0 · canyon 0 / 0 · Yellowstone 0 / 0 · Kilauea 0 / 0 · bucket 0 / 0 · pocket 0 / 0 · thumb 1 (the phrase "rubs off on your thumb", not a scene) / 0 · classroom 1 (inside an hb note, not student text) / 0 · desk 0 / 0 · lava flow 0 / 0 (the practice claim was re-scened to "a lava field in Oregon") · Grand 0 / 0 · Austin 0 / 0. The quiz's one jar item was re-scened to "a tall glass" drawn lidless. The design-move inventory and the ⟂ draw constraints are on the type sheet.

Names run (`python3 build_tools_names.py Lopez Ian Isaac Jace` → exit 0)

Lopez free · Ian free · Isaac free · Jace free. (The register globs Grade 3; these names are Topic 18's row in grade4/NAMES_POOL.md.) No other character or claim-maker appears.

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

Lesson	Ideas	One idea per sentence	Cause → effect	Named actor	Verb carries	Plain word before term	≤ 8 lines per run	Fixes applied
L01	26	yes	yes (the clue, then the verdict)	Ian, Isaac, Jace, Miss Lopez, granite, obsidian	picks up, locks, rubs off, shines	locked grains / smooth like glass / no layers → igneous rock	yes (six subtitles, thirteen figures)	the closing two lines joined into one so the return is one sentence; "none of them has layers" kept as the rule's own form
L02	25	yes	yes (melts → pours out → cools → hardens; fast → tiny, slow → big)	Jace, Miss Lopez, Isaac, the melted rock, the lava	melts, pours out, cools, hardens, stays	melted rock → lava; the forming story → igneous rock	yes (five subtitles, six figures)	"The rock cooled long ago. Today it is…" joined with "so"; "grow" struck for "end up big"
L03	26	yes	yes (pressed grains rub off; layers go right through)	Jace, Ian, Miss Lopez, sandstone, limestone	presses, rubs off, splits, settles	small pieces that settle → sediment; pressed pieces in layers → sedimentary rock	yes (six subtitles, thirteen figures)	the Texas-hills line cut (L04 carries it); "Now consider shale. It is…" joined into one sentence
L04	27	yes	yes (settle → press → stick; shells settled → pressed → limestone)	Miss Lopez, Ian, the river, the pieces, the layers	shakes, settles, presses, sticks, covers	settled pieces → sediment (recap)	yes (six subtitles, nine figures)	"underneath" → "below"; "swirl" → "spread all through"; the second simplification note and three summary-of-the-lesson lines cut for length

Figures (59: 43 SVG by code, 14 photo and 2 cartoon placeholders; author's review-figures self-check on fourteen renders)

Depiction register: the family card (name or letter in the top row; GRAINS / LAYERS / ALSO; a close-up at the right; headache cards blank with their question; item cards lettered J–V outside A–D, no chip, no tint); the close-up (seven kinds: locked = a jittered tiling with no gaps, fine = the same at a tiny scale, glass = an even black fill with a shine streak, holes, pressed = round grains with gaps, pebbles = bigger round pebbles in sand, shells, mud = thin bands, layers = flat bands; the tiling guard shares every corner, the pressed guard asserts every gap); the step cards (three, numbered, each drawing the moment its caption names); the cross-section (labels on the teaching form only); the tables (columns from measured text, every cell asserted to fit); the jar (lidded in the lesson, lidless as the item's glass; layers clipped to the glass's rounded shape, labels outside the clip); rule cards. Renders opened at 800–1000 px: cross_section_labels (the lava label moved from the channel to the flow, twice), card_granite, card_limestone, card_pumice, card_shale, card_conglomerate (given its own pebbles close-up), steps_igneous (the floating close-up on card 3 replaced by the slab itself), steps_sedimentary (the layers close-up scaled to its card; the caption moved above the bands), table_two_families, table_fast_slow, table_pieces_to_rock, jar_settled (the layers clipped to the glass; the labels moved outside the clip after the first clip hid them), q_jar_w, q_card_k, q_grains_tiny, pp_card_r. Every photo and cartoon key has an IMAGE_BRIEFS entry for David's pack; no picture was rendered (the picture stage waits for James's answers above).

Judgement calls, recorded

"Magma" is not said. The register's text carries "lava from a volcano cooling into black rock" and nothing for the underground case, so the brief's condition is not met; "melted rock deep underground" carries it in plain words. Grade 6-8's EAR-011 names magma.
"Sediment" is a third coined term (James's call above), explained before it is named, as the brief asks.
All three terms are recalled by MCQ, never typed: the brief's rule (nine or more letters, or on the hard-to-spell list), and the checker's S35 list carries "sediment". The bank's one typed key is "layers". A Grade 4 child could probably spell sediment — flagged, not changed.
L01's one non-example is Jace's layered rock, named only in L03. L01 says "This rock is not in the family, because a rock in this family never has layers" and leaves its name for L03's headache, so the two looks-like lessons hand to each other without L01 teaching sedimentary rock a lesson early.
The intro's black rock is L02's headache rock. The intro never points at the first lesson (L01's tray, granite and obsidian are absent); Jace's pocket rock returns in L02 as continuity, not as a hand-over.
The jar demonstration's limit is said in plain words ("the jar's layers stay loose"), and Ian's TRUE claim and the bank's overnight-tub item test it, because Kusnick reports children reading the model as the whole story.
The fast / slow pair is the lesson's one comparison and sits in a two-row table (a comparison is a table), with the cross-section as the picture a child can see the two places in.
Real places checked: Hawaii's basalt lava flows (Kilauea is Grade 3's example and is not named here); Yellowstone's obsidian (Obsidian Cliff); the Grand Canyon's sandstone layers (among others — the lesson says "layers like these", not "all sandstone"); the Texas Hill Country's fossil-rich limestone laid down under a shallow Cretaceous sea (the lesson says "long ago, a shallow sea covered that part of Texas" and gives no age). Bank places (Colorado granite, Oregon obsidian, Idaho's lava fields and lakes, Washington's lava fields, Utah's red sandstone cliffs, Florida's shelly limestone, Kentucky's layered limestone road cuts, Maine's beach rocks, Ohio's layered river banks, Georgia's granite mountains, New York's shale cliffs, Wisconsin's limestone walls, Kansas's limestone hills) are each true of the rock named.
Pumice is a dry rock here (pale, glassy, full of tiny holes) and never meets water, because Topic 13 bars it from any floating scene.
S36's rock names are kept — they are the register's own examples and each is shown, carded and photographed; a child meets the word beside the thing.

Owed

Blind review of every key (the standing gate): the bundle is CUT — grade4/topic18_build/blind/2026-10-02/BLIND_BUNDLE_G4T18.md (112 MCQ, 1 typed, 1 written, 2 multi; keys sequestered in KEYS_SEQUESTERED_G4T18.json) — and waits for two cold readers; the non-author cold pass below is not the blind read.
James's answers to the two questions above; then the picture pass (14 photos, 2 cartoons) through David's pack.
Videos: the batch at handoff/G4T18/videos/batch_G4T18.json (intro, L01–L04, summary; READY.md beside it) is built and NOT sent — James's word is needed for Grade 4's first batch.
Topic 17 alignment: Topic 17 is being written in parallel; when its content lands, read its register words against this topic's uses ("grains too small to see", "shiny like glass", "gritty") and its figure register against the family card, and record any difference.

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

The reviewer answered all 116 bundled items cold and agreed with every key (quizzes, attached, summary, practice, bank; the typed "layers" accepted). Findings and actions:

Finding	Action
"Hot to touch today" (PR-8) is true of a dark rock in the sun	Rewritten: "still hot from the lava today"; the fb says none of the lava's heat is left. L02-q8 (the TRUE claim, "as cool as the ground around it") stands: the ground in the sun is warm too, so the comparison holds
"Which step comes last?" — pressing and sticking overlap (PR-7, PP type 5)	The bank item now asks "Which step turns the loose pieces into rock?"; PR-7 keeps "last" because the lexicon's order (settle, press, stick) is explicit and the fb walks it
PP type 2's shells item had reasoned options (verdict + reason)	Recast before this pass landed as a type 1 item on a Kentucky road cut, bare family options
PP type 2's Maine item: the key carried two clauses and no distractor did	Key cut to "It has no layers."
"It melted again." in the cooling set is a weak distractor	The set's fourth option is now "It was pressed by layers." (the sedimentary mix-up); "It never cooled." stays as the hot-rock belief
PR-9's starter "Long ago, sand settled…" gave the first step	Starter cut to "Long ago…"
PP type 9's igneous definition was L01-q9 word for word; PP type 6's growing-grains claim was L02-q10 word for word; PP type 2's never-clue item was L01-q8 word for word	All three re-instanced (the family named from "rocks that are hard … make up one family"; a New Hampshire mountain's grains "still getting bigger today"; "Which of these clues is never found on igneous rock?"); S28 now clean
The settled-glass item said "gravel" where the lesson says "pebbles"; its drawing was the teaching jar relabelled	One word, "pebbles"; the item glass is lidless with layers 14 / 34 / 10 against the jar's 22 / 26 / 16
Item cards J and S shared granite's tiling and colours while their rows said gray and white; card N shared basalt's tiling	A `locked_gray` close-up kind (grays and white) with its own seed for J and S; a new seed for N
Pumice card V showed no shine and some holes overlapped	A faint shine streak added; holes placed with a no-touch guard
`steps_igneous`: the "melted rock" label crossed the ribbon's end	Moved above the flow
`cross_section_labels`: "melted rock" floated in the brown ground; "big grains" sat on the still-liquid chamber	"melted rock" now sits on the orange pocket; the grain labels read "tiny grains or glass, once it has hardened" and "big grains, once it has hardened"
Rule cards: the third line sat on the box's bottom border	The card grew 12 px
Jars were left-heavy in a square thumbnail	The frame is cut to the glass plus its labels
"What is the correct reason…" and "X is wrong to say…" read as stock phrasing / a given verdict	Kept: this is the settled claim-key form (the stem fixes the verdict, the options are four single-statement reasons; `10 Rules/An option never bundles a verdict with its reason.md`)
The two "will NOT have" stems and the never-clue stem are negatives	Kept as single negatives over positive options — the legitimate negative-half read (Topic 11 bank precedent); no double negative anywhere
PP type 6 claims mirror the quiz claims' beliefs; PP type 8 mirrors L03-q11's move; PR-1 / PR-3 mirror L02-q1 / L04-q1	By design: the misconception rows ask for the belief in a lesson item AND a bank item (fresh scenes: a lava field, a river cliff, a tub); the hard tier must recall every coined term; the Grade 3 retrievals are one fact each. Recorded as same-move, different-scene on the type sheet
"Igneous rock never has layers" overclaims (flow banding, layered ash); "pressed vs locked" fails for tightly cemented sandstone	Kept: "never in layers" is the register's own wording for EAR4-007 (James's atom), and "locked" / "pressed" are the one pair of words the two-way sort rests on at Grade 4. Recorded here as the simplification the register makes; Topic 19's three-family sort is where "usually" would enter, and that is James's call
L02-q4 is answerable from "tiny" alone	It is the why item of a type (c) set: the stem states the verdict and asks the reason, which is the form

Re-run after the pass: checker 0 FAIL (the S28 WARN cleared), check_g4_articles.py 0 FAIL, the bundle re-cut (112 MCQ, 1 typed, 1 written, 2 multi), the batch rebuilt, check_brief.py clean except the INTRO's standing hit, the blind-line grep 0.

v2 — reviews applied (2 October 2026)

Skill: course-build → write-lesson (article lines) → write-video-script (the L01–L04 script beats) → write-mcqs (every item change) → write-explanation (fb / why / model lines) → sentence-atomise (every new sentence) @ b0fa825. Three records applied in order of authority: BLIND_READ_G4T18_2026-10-02.md, REVIEW_G4T18_2026-10-02.md, FIGURES_REVIEW_G4T18_2026-10-02.md, plus one row from Topic 19's figure review (SVG ids unique on a page). Nothing under grade3/ or another topic's folder was touched. Nothing committed.

In plain words. The keys were right; the bank was leaning on the lessons. Twelve bank items and two practice items were the quizzes' own scenes with a new letter or a new place, and in the hard tier the right answer was the longest option in nine items. Every one of those now has a fresh rock, child or move, every hard key is one short clause, and no key is the longest option. The two facts the quiz tested that only the article taught are now spoken in the videos too. The figures no longer hand over a key by colour, and Jace's rock is drawn one way everywhere.

Finding → change

Source	Finding	Change
Blind	PP-02 = L01-q1's Rock J re-lettered	Rock S is a dark red, gray and black locked tiling (`locked` with fills, seed 13); rows "dark red and gray"
Blind	PP-12 = L01-q2's Rock K	Rock T is pale yellow sand in rows (seed 15, yellow fills); K keeps tan and brown (seed 31) — both on the one sandstone depiction, neither a twin
Blind	PP-10 = L03-q5 as a card	Rock R: "pebbles of many sizes pressed in sand / flat layers / gray pebbles, pale sand" (seed 18, gray pebbles on pale sand)
Blind	PP-17's drawing = L02-q6's data	PP-17 is 12 gray cells of about 30 px (`big_gray_4`); L02-q6 is 30 cells of about 20 px (`big_gray`); ⟂ PP-16 ⟂ PP-17 on the type sheet
Blind, Review	PP-43–48 = the six quiz claims re-cased	each re-scened with a different child AND scene from its quiz twin: Jace's hot New Mexico desert rock; Isaac's always-there black California hillside (its igneous clues given); Jace's growing Wyoming grains; Isaac's painted Nevada road cliff; Ian's always-there Tennessee road wall; Jace's glass bowl after a week
Blind	three pumice items (PR-4, PP-04, PP-07)	PP-04's Rock V is dark gray with small holes; PP-07 is a Washington road-cut rock with tiny grains that will not rub off; PR-4 keeps pumice in words
Blind	PP-23 = L02-q11 with lake for sea	a thin lava flow over cold Alaskan ground, hard within an hour
Blind	PR-11 = PR-10 with bay for sea	a muddy puddle clearing after rain
Blind	rock names coined before items lean on them	confirmed in L01 / L03 articles; L03's SCRIPT now names conglomerate ("round pebbles pressed together in sand") so the video channel carries all four sedimentary rocks; L04's article line names all four pieces → rocks ("…and pebbles and sand make conglomerate"); which-pieces-make-which-rock items already present (L04-q8, PP-30–32)
Blind, Review	PP-40 typed: stem echoed "bands"; "sheets" uncorrected	stem says "flat parts that run right through it"; accept list widened to 15 (ten child spellings); wrong-term lines for stripes, bands, sheets ("Shale splits into sheets along its layers"), lines
Blind	settle → press → stick as three steps in L04 text	the article now says "So there are three steps: the pieces settle, more layers press down, and the pieces stick."; the script's beat 3 already speaks the three; ⟂ PP-27 ⟂ PP-28 ⟂ PP-29
Blind	PP-38 / PP-39 duplicates; L03-q10 / q11 answer each other	⟂ PP-38 ⟂ PP-39 on the type sheet; L03-q11 re-scened (a stream bed) with options soil / sediment / a mineral, so neither item prints the other's key
Blind	PP-19 / PP-20 lean on "never layers"	PP-19 rebuilt as the review wrote it (key "Its grains are locked together."; the three sedimentary clues as distractors); PP-20's key never leaned on it
Review (Essential)	nine hard keys the longest option	PP-34, 35, 44, 45, 46, 47, 48, 52, 53 cut to the reviewer's one-clause keys; distractors in PP-24, 25, 33, 43, 50, 52, 53 lengthened or the key trimmed so NO key is the longest option anywhere except inside a fixed set (the family set's "sedimentary rock", the grains set's "tiny grains, or glass", the steps set), where the same words sit in every item of the type
Review (Essential)	L04 script: the jar's limit article-only	beat 2 adds "They're still loose: making rock takes many more layers, and a very long time."; "Now that sea is gone." cut from beat 4 (kept in the article); trims elsewhere; 150 words
Review (Essential)	L02 script: the grains-stop line article-only	beat 4 adds "Once it's hard, the grains stay as they are."; beats 1–3 trimmed; 150 words
Review	"stuck" as the sedimentary clue	L03's conglomerate line, caption, card, photo note and brief, L03-q5, PR-5, PP-10, PP-53 say "pressed together in sand"; "stick / stuck" stays only as the third STEP of forming
Review	obsidian "no grains at all" vs Topic 17's definition	L01 forest, article and script, the obsidian card, Rock L, PP-03, PP-16, PP-25: "no grains you can see, even up close: smooth like glass"
Review	lava-field scene reused (L02-q4, PR-8, PP-25, PP-43, PP-44)	L02-q4: the slopes of an old volcano in California; PR-8 rebuilt as a TRUE claim (Isaac: the grains stopped getting bigger when the rock hardened — Happs 1982), which also gives the claim string a third TRUE (3 of 8); PP-25 an Arizona hillside; PP-43 / 44 as above; "lava field" grep over every content file: 0
Review	PR-13's stem named the key's head ("sediment")	"A rock formed when small pieces settled in layers, were pressed together and stuck, over a very long time."
Review	L04 subtitle staged the reader	"Pieces settle in a jar"
Review	L02-q2's why circular	"When a liquid cools down enough, it turns into a solid. / This liquid has cooled down enough. / So it is now a solid."
Review	claim string 2 of 8 TRUE	PR-8 is now TRUE (above)
Figures 1	five item close-ups pixel twins of teaching cards	K and T on the sandstone depiction with their own fills and seeds; L dark green; M seed 16 on a grayer cream; P seed 17 (six bands of varied thickness); R seed 18, gray pebbles on pale sand; teaching cards keep seed 3; renders compared side by side — no twin
Figures 2	colour carried the key on the fast-or-slow run	all four close-ups in basalt's two greys or a plain glass fill (`fine`, `big_gray`, `big_gray_4`, dark red-brown glass); blind texts keep counts and px sizes and drop "tiny" / "big"
Figures 3	Jace's rock drawn two ways	ONE depiction, `closeup("sandstone")`: round grains with gaps in rows, two tones swapping every two rows — card_striped, card_sandstone, card_sandstone_headache, K, T; the `pressed` kind retired
Figures 4	Rock L black, row says dark green	`glass` takes a fill; L is `GLASS_GREEN`
Figures 5	settled jar's pebble band flat	`jar_settled` and `q_jar_w` draw the pebbles as packed circles (r 3, staggered rows) on a pale sand ground
Figures 6	`steps_sedimentary` card 3 changed kinds	card 3 draws `SED_BANDS`, card 2's own colours; one `_arrow_down` helper so every settling / pressing arrow is one size
Figures 7	headache partners unplaced	`card_obsidian_headache` placed after `card_granite_headache` (L01); `card_limestone_headache` after `card_sandstone_headache` (L03); the L01 script's beat 1 names both
Figures	`pp_jar_x`, `granite_quarry_photo`	deleted (figure, brief and FIGURES entries); 58 figures
Figures (minor)	cross-section text touched a chip; "once it has hardened" lines floated	the OPEN AIR chip moved up; the tiny-grains line sits under the dark hardened flow, the big-grains line beside the pocket, both anchored
Figures (minor)	`steps_igneous` label on the stream	"melted rock" at the card's top right
Figures (minor)	granite "gray" with no gray cell	`GREY_M` in granite's tiling (card_granite, card_granite_l03, card_granite_headache)
Figures (minor)	`table_two_families_blank` missing	added (row names in place, every cell empty); the summary's first screen note names it
Figures (minor)	PHOT suffix "plain background" on scenes	`PHOT_SCENE` ("the real place in daylight, natural surroundings") on hawaii_rock, lava, canyon and texas_hill photos
Topic 19 review	jar clipPath id from `hash()` — two jars could share an id and the second be clipped to the first's rectangle	id = stage + glass geometry + a module counter (`jarclip_settled_30_20_100_160_2`); the built L04 article and quiz pages screenshotted (`/tmp/g4t18_figs/page_jars_sheet.png`): three jars, three ids, every jar renders with its own layers
Checker	PP-07's new stem had 3 sentences tagged easy (S23 FAIL)	joined to two sentences

Counts and key strings

Bank 53 items (50 MCQ, 1 typed, 2 multi); per atom EAR4-007 10 · EAR4-008 15 · EAR4-009 12 · EAR4-010 16; tags easy 18 → medium 17 → hard 18, non-decreasing with atom order; keys A 11 B 14 C 14 D 11. Bank key string (T = typed, M = multi): ADBBCDCACADBBADCBDACBCDACBACDACDBCBDBBDTMMABCBACCDCAB (unchanged positions). Video key string: A B B A A B B. Scripts 149 / 150 / 148 / 150 words; intro 116; summary 92. Articles 300 / 281 / 298 / 316 (L04 WARN above 300: the three-steps line and the conglomerate clause were added on the reviews' word; one Grade 3 recall line cut to offset).

Checker results

build_tools_check_topic.py → 0 FAIL, 24 WARN (S8 multi-sentence beats as before; S36's eleven words, the same list as v1 with "stripe" now in PP-46). check_g4_articles.py → 0 FAIL (L04 316-word WARN, above). build_tools_names.py Lopez Ian Isaac Jace → exit 0. Site built to /tmp/g4t18_v2.html (4 lessons, 13 practice, 53 pp100, 58 figures). build_tools_video_batch.py → 6 videos; check_brief.py clean except the INTRO's standing R79 line and its at-the-limit note — not sent. Blind bundle re-cut to blind/2026-10-02-v2/ (112 MCQ, 1 typed, 1 FRQ, 2 multi = 116); grep of its 15 [Drawing: …] lines for igneous | sedimentary | cooled | settled | melted | sediment | tiny | big → 0 hits. Every changed figure re-rendered to /tmp/g4t18_figs/ and opened (three contact sheets): no item card is a twin of a teaching card; the big-grain items carry no granite colour.

Questions for James — built as recommended

"Sediment" as a third coined term — built as the reviewer recommends: explained ("small pieces that settle to the bottom of water"), then named; recalled by MCQ.
"Lava" only, never "magma" — built as recommended: "melted rock deep underground" carries the underground case; the cross-section's "lava" label sits on the flow.

Two more for James (register and scope, not this page)

The atom EAR4-007's "never has layers" — suggest softening the ATOM to "no layers inside a piece of it" (the register, not this page; the page keeps the register's wording until the atom changes).
Whether Topic 17 gains a scope line about glassy obsidian ("a rock is made of grains of minerals" — obsidian's grains cannot be seen).
(Figures, cross-topic) Topic 17's grain_box("big") draws granite as speckles on pink; Topic 18 draws a locked tiling of pink, white, gray and black. A scope line here or an extension of Topic 17's drawer is James's call; Topic 17 is not edited.

Item ids whose stem, options, key, model, figure, blind text or accept list changed

Quiz: L01-q2 (figure), L01-q3 (figure, fb), L02-q2 (why), L02-q4 (stem, why), L02-q5 (figure, blind), L02-q6 (figure, blind), L03-q1 (figure), L03-q2 (figure), L03-q5 (stem, fb, why), L03-q11 (stem, options, key position, fb, why), L04-q2 (figure). Practice: PR-5 (stem, fb, why), PR-8 (stem, key, fb, why, source), PR-11 (stem, fb, why), PR-13 (stem, fb). Bank: PP-02 (figure, blind), PP-03 (stem), PP-04 (figure, blind), PP-07 (stem, why), PP-10 (figure, blind, why), PP-12 (figure, blind), PP-16 (figure, blind, why), PP-17 (figure, blind), PP-19 (options, why), PP-23 (stem, why), PP-24 (options), PP-25 (stem, options, why), PP-33 (options), PP-34 (options), PP-35 (options), PP-40 (stem, accept, wrong_fb, why), PP-43 (stem, options, why), PP-44 (stem, options, why), PP-45 (stem, options, why), PP-46 (stem, options, why), PP-47 (stem, options, why), PP-48 (stem, options, why), PP-50 (options), PP-52 (options), PP-53 (stem, options). Keys compared against v1's sequestered file after the re-cut: 15 entries differ — the 13 key TEXTS trimmed or rebuilt above (PP-19, 25, 34, 35, 40's accept list, 44–48, 50, 52, 53), PR-8's verdict (no → yes, the new TRUE claim), and L03-q11's shuffle letter (the key word "sediment" is unchanged; its authored position moved from option 1 to option 2 by design). Every other key is byte for byte v1's.

Sentence-atomise, the new sentences

Line	Ideas	Actor	Verb carries	Link stated
"Obsidian has no grains you can see, even up close: it is smooth like glass."	1 (one clue, its look)	obsidian	has / is	colon: the look explains the clue
"So there are three steps: the pieces settle, more layers press down, and the pieces stick."	a list of three, as a list	the pieces, the layers	settle, press, stick	"So" closes the run
"Sand makes sandstone, mud makes shale, pieces of shell make limestone, and pebbles and sand make conglomerate."	a four-part list	the pieces	make	—
"They're still loose: making rock takes many more layers, and a very long time." (script)	1 + its reason	the pieces; making rock	are loose; takes	colon: reason
"Once it's hard, the grains stay as they are." (script)	1	the grains	stay	"Once"
"A stream slows down, and the sand it was carrying sinks and settles on the stream bed."	cause → effect	the stream, the sand	slows, sinks, settles	"and" (sequence)
"A muddy puddle clears after the rain stops, and a thin layer of mud lies on its bottom."	cause → effect	the puddle, the mud	clears, lies	"after", "and"

v3 — polish (2 October 2026, from the v2 re-read)

Skill: course-build → write-mcqs (every item change) → write-explanation (the why lines) → sentence-atomise (each new sentence) @ efd8fba. Source: BLIND_READ_G4T18_2026-10-02.md, the "v2 re-read" section (Reader C, every key agreed). Nothing committed.

In plain words. The keys were right. Five bank items still sat on a lesson or quiz scene with a new place or child, two bank claims were near-copies of quiz claims, one bank key and one quiz key were the longest option, and one quiz claim compared a rock to the ground. Every one is changed; no key letter moved.

Flag	Change
PP-46 was L03-q9's painted-stripes cliff renamed	Isaac's striped pebble on a Michigan lake shore; the why adds "a broken edge shows them too"
PP-33's key the longest option and the only one with the lesson phrase "in flat layers"	key "at the bottom of water"; the why's last line cut to match
L04-q4's key the only long, specific option	"Sand settled one layer at a time."
PP-43 was Jace's lava rock with a new place	a black rock on a shelf in a Minnesota museum that was lava long ago; the why ends "no longer hot from the lava"
PP-48 was the jar / bucket with a new vessel	a flood leaves a flat layer of mud on an Iowa field; a year on, the mud is still loose
PP-52 was Ian's Texas hill with a new state	an Indiana courthouse built from blocks of limestone
PP-44 near L02-q9 and L02-q4	the move varied: the clues are given, the claim is given, and the student says what the rock WAS before it was hard rock (melted rock, among loose sand / soft mud / pieces of shell)
PP-47 near L04-q6	the move varied: the student says what each layer WAS before it was rock (loose pieces that settled, among melted rock from a volcano / a cut made by road builders / a line of paint)
L02-q8 "as cool as the ground around it"	"no longer hot from the lava" in the stem, both fb and the why; the article and script lines stand (the review ruled them)
PP-40: accept "stripes" and "bands"?	NOT accepted: the typed item's `wrong_fb` already carries a wrong-term line for each (stripes and bands are what the layers look like from outside), and those lines stay

Freshness grep before the re-scenes (every content file): museum 0, pebble-on-a-shore 0, Michigan 0, Minnesota 0, Iowa 0, Indiana 0, Montana 0, field 0, flood 0, courthouse 0. Scene kinds now in the bank for type 6: a museum shelf, a hillside, a mountain, a lake-shore pebble, a highway rock wall, a flooded field — none is a lesson or quiz scene kind (classroom, Hawaii lava flow, Texas hill, jar, bucket, canyon wall, sandstone cliff, old volcano's slopes).

Option lengths, the changed items (key never the longest): PP-33 39 / 22 / 16 / 22 (key 22); PP-43 38 / 35 / 42 / 38 (key 38); PP-44 10 / 11 / 8 / 15 (key 11); PP-46 48 / 44 / 42 / 41 (key 44); PP-47 25 / 26 / 27 / 15 (key 25); PP-48 36 / 40 / 44 / 51 (key 44); PP-52 40 / 30 / 43 / 41 (key 40).

Sentence-atomise, the new sentences.

Line	Ideas	Actor	Verb carries	Link stated
"The lava cooled and hardened long ago, and none of its heat is left."	cause → effect	the lava, its heat	cooled, hardened, is left	"and"
"So the rock is no longer hot from the lava."	1	the rock	is	"So"
"Igneous rock forms when melted rock cools and hardens." / "So the rock was melted rock before it was hard rock, and it was not always there."	rule → case	igneous rock; the rock	forms; was	"So"
"Each layer settled as loose pieces, and more layers pressed it into rock."	two steps in order	each layer, more layers	settled, pressed	"and" (sequence)
"So the stripes go right through the pebble, and a broken edge shows them too."	1 + its visible test	the stripes, a broken edge	go through, shows	"So", "and"
"The mud has settled in a layer, but no more layers have pressed down on it for a very long time."	the step done, the step missing	the mud, more layers	has settled, have pressed	"but"

Checks and the bundle

build_tools_check_topic.py → 0 FAIL, 25 WARN (the S8 beats and S36 words as in v2; "stripe" stays in PP-46; "museum" is new in PP-43). check_g4_articles.py → 0 FAIL (L04 316-word WARN as before). check_svg_ids.py on figures.py → PASS (58 figures, 0 shared ids). build_tools_names.py Lopez Ian Isaac Jace → exit 0. Site built to /tmp/g4t18_v3.html (4 lessons, 13 practice, 53 pp100, 58 figures). Blind bundle cut to blind/2026-10-02-v3/ (112 MCQ, 1 typed, 1 FRQ, 2 multi = 116); its [Drawing: …] lines grepped for igneous | sedimentary | cooled | settled | melted | sediment | tiny | big → 0 hits. Keys compared with v2's sequestered file: 5 key TEXTS differ (L04-q4, PP-33, PP-44, PP-47, PP-48 — the trims and the varied moves above); every key LETTER is v2's. Bank key string unchanged: ADBBCDCACADBBADCBDACBCDACBACDACDBCBDBBDTMMABCBACCDCAB; keys A 11 B 14 C 14 D 11; tags easy 18 → medium 17 → hard 18.

Item ids changed in v3

Quiz: L02-q8 (stem, fb, why), L04-q4 (key text). Bank: PP-33 (key text, why), PP-43 (stem, options, why), PP-44 (stem, options, hb, why), PP-46 (stem, options, why), PP-47 (stem, options, hb, why), PP-48 (stem, options, hb, why), PP-52 (stem, options, why). Type sheet rows 4 and 6 and the type 6 exemplar updated in TYPES_G4T18.md.

Guidance, questions and accuracy review (2 October 2026)

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

Topic 18 — guidance, questions and accuracy review, 2 October 2026

The topic is sound and teaches in the right order (looks, then forming, one family at a time); the science is honest at Grade 4 depth and every key is right. Not ready to ship as it stands for two reasons a child would feel: in the hard tier the correct reason is the longest option in nine items and the only two-clause option in four, so a guesser who picks the longest wins the mastery tier; and two quiz items (L02-q10, L04-q7) rest on sentences only the article says. Everything else is wording. Re-performed: scripts 149 / 150 / 147 / 150; intro 116; summary 96; articles 299 / 281 / 298 / 306; bank 53 (50 MCQ + 1 typed + 2 multi), per atom 10 / 15 / 12 / 16, keys A 11 B 14 C 14 D 11; blind lines 0 answer words; names clean; the five beliefs each met. One false record claim: the freshness grep searched "lava flow" and missed "lava field" (L02-q4, PR-8, PP-25, PP-43, PP-44).

The author's two questions, the reviewer's view: (1) "sediment" as a third coin — yes (explain-then-label with the name unpacked; Topic 20 and STAAR G5 use it; MCQ recall costs no spelling). (2) "lava" only, never "magma" — yes ("melted rock deep underground" IS what magma means, so nothing is false; the cross-section's "lava" label sits on the flow).

Leave (and two for James): "never has layers" is the register's wording — suggest James soften the ATOM to "no layers inside a piece of it", not this page; "pressed together and stick" has no false mechanism; the Texas hill's "the sea is gone" is an honest dodge of uplift; "deep underground" without magma is honest. Obsidian "has no grains at all" collides with Topic 17's "a rock is made of grains of minerals" — the fix pass takes the Topic 18 wording ("no grains you can see, even up close: smooth like glass"), and whether Topic 17 gains a scope line is James's.

where	finding	rule broken	severity	the rewrite
PP-44–48, 52, 53, 34, 35	key longest by 20+ characters; PP-44–47 the only two-clause option	key tells; check 47	Essential	PP-44 "It hardened from melted rock." · PP-45 "The melted rock cooled slowly." · PP-46 "Each stripe is a layer of sand that settled." · PP-47 "Each layer settled as loose pieces." · PP-48 "They need pressing for a very long time." · PP-52 "The settled pieces were shell, not sand." · PP-53 "They settled in a layer under water." · PP-34 "Mud settled one layer at a time." · PP-35 "The shells settled among the other pieces."
L04 script beat 2	the jar's limit article-only; L04-q7, PP-48 test it	each channel teaches the whole part	Essential	add "They're still loose: making rock takes many more layers, and a very long time."; cut "Now that sea is gone."
L02 script beat 4	grains-stop line article-only; L02-q10 tests it	same	Essential	add "Once it's hard, the grains stay as they are."; trim as the reviewer wrote
PP-19	key does not sort; silly distractors	real distractors; check 74	Worth fixing	key "Its grains are locked together."; distractors "It has flat layers." / "A grain rubs off on your thumb." / "It has pieces of shell inside."
L03 conglomerate line, caption, card; L03-q5; PR-5 fb	"stuck" as the sedimentary clue, against Topic 17's "grains stuck together"	one word for one thing	Worth fixing	"round pebbles pressed together in sand, in flat layers"; fb "Pieces pressed together in flat layers are the clues of sedimentary rock."
L01 obsidian line; PP-3, PP-25	"no grains at all" vs Topic 17's definition	category definition	Worth fixing	"Obsidian has no grains you can see, even up close: it is smooth like glass."
PP-43, PP-44; L02-q4; PR-8	lava-field scene reused	fresh instance; check 50	Worth fixing	old volcano's slopes, as the reviewer wrote
PR-13 stem	"sediment" names the key's head	stem never names its answer	Worth fixing	"A rock formed when small pieces settled in layers, were pressed together and stuck, over a very long time. What is that rock called?"
L04 subtitle	"Watch pieces settle"	an article never stages	Worth fixing	"Pieces settle in a jar"
L02-q2 why	circular	a why carries the deciding steps	Worth fixing	"When a liquid cools down enough, it turns into a solid.<br>This liquid has cooled down enough.<br>So it is now a solid."
PP-40 typed	"sheets" uncorrected	wrong-term line	Worth fixing	wrong_fb "sheets": "Shale splits into sheets along its layers. Each band is a layer."
claim string 2 of 8 true	thin	keys spread	Worth fixing	one more TRUE claim

Figure review (2 October 2026)

Skill: course-build → review-figures (non-author pass; filed by the coordinator from the reviewer's hand-back) @ b0fa825 — figures at g3-5-science commit 2402ddf

Topic 18 — figure review, 2 October 2026

The cards are clean and uniform, and the two grain pictures (a locked tiling with no gaps against round grains with gaps) read at a glance; the settled jar reads instantly. But sandstone looks like two different rocks in L01 and L03, five quiz cards carry the teaching card's exact close-up, and colour carries the key on the fast-or-slow items. Mechanical pass clean (59 of 59 rendered; no text outside a frame, no overlap).

Five item close-ups are pixel twins of teaching cards (K = card_striped, L = card_obsidian, M = card_limestone, P = card_shale, R = card_conglomerate). Seed-driven variation (streak position, band count and thickness); pass seeds for M and R.
Colour tracks the key on the fast-or-slow run: tiny = basalt greys, big = granite colours, glass = obsidian black. Draw the big-grain items in a neutral grey or an untaught palette so grain size is the only difference; blind texts drop "tiny" / "big" and keep the px sizes.
Jace's rock drawn two ways: card_striped (flat bands, no grains) in L01 and card_sandstone (round grains, no bands) in L03 are the same rock; items K and T split the same way. One depiction: grains in rows of two tones (grains and layers at once) for Jace's rock, sandstone, K and T.
Rock L's row says "dark green, shiny"; the close-up is black. A fill parameter on the glass kind.
The settled jar's pebble band is a flat grey band; the shaken jar draws pebbles as circles. Packed circles on jar_settled and q_jar_w.
steps_sedimentary card 2 → 3 changes kinds (sand / dark mud → tan / orange-brown). The after box keeps the before box's colours.
Headache pairs half placed: place card_obsidian_headache after card_granite_headache (L01) and card_limestone_headache after card_sandstone_headache (L03).

Unplaced: card_obsidian_headache, card_limestone_headache — place (7); pp_jar_x — delete (twin of jar_shaken, unused); granite_quarry_photo — delete (its only slot would show deep-cooled rock as a hillside with no account of how it got there — Topic 20's).

Minor: cross_section_labels text touches a chip and the two "once it has hardened" lines float (anchor each beside its thing); steps_igneous label on the stream; settling arrow sizes; granite row says "gray" but the tiling has no grey cell; table_two_families_blank missing for the summary's staged variant; scene briefs end with the PHOT suffix "plain background" (split the suffix). Cross-topic: Topic 17's grain_box("big") draws granite as speckles on pink while Topic 18 draws a locked tiling — a scope line, or extend Topic 17's drawer (James's call, recorded); Grade 3 Topic 26 owns a shaken / settled jar — reuse the glass or record the difference.

key	kind	used by	verdict	what is wrong	the fix
card_granite_headache, card_sandstone_headache	svg	L01 / L03	OK	pair incomplete	place the partner
card_obsidian_headache, card_limestone_headache	svg	none / script only	PLACE	—	(7)
card_granite, card_granite_l03	svg	L01, L03, summary	minor	no grey cell	GREY_M
card_obsidian, card_basalt, card_pumice, card_limestone, card_shale, card_conglomerate	svg	L01 / L03	OK	—	—
card_striped, card_sandstone	svg	L01 / L03	FIX	two depictions of one rock	(3)
q_card_j, q_card_n, pp_card_s, pp_card_v	svg item	items	OK	—	—
q_card_k, q_card_l, q_card_m, q_card_p, pp_card_r	svg item	items	FIX	twins of teaching cards; L black not green	(1), (3), (4)
pp_card_t	svg item	bank	minor	"red" vs orange; split depiction	(3)
q_grains_tiny, pp_grains_glass	svg item	items	minor	blind words; obsidian black cues fast	px only; (2)
q_grains_big, pp_grains_big	svg item	L02 q6, bank	FIX	granite colours cue slow	(2)
steps_igneous	svg	L02	minor	label on the stream	move
steps_sedimentary	svg	L04, summary	FIX	kinds change; arrow sizes	(6)
cross_section_headache, cross_section_labels	svg	L02	OK / minor	text touches chip; floating lines	move; anchor
table_fast_slow, table_pieces_to_rock, table_two_families	svg	L02 / L04	OK / minor	blank variant missing	add key
jar_shaken	svg	L04	OK	—	—
jar_settled, q_jar_w	svg	L04	FIX	pebble band shows no pebbles	(5)
pp_jar_x	svg	none	DELETE	twin	—
rule_igneous_looks / forms, rule_sedimentary_looks / forms	svg	rule steps	OK	—	—
tray_cartoon, jar_cartoon	cartoon	L01 / L04	OK	—	—
the nine rock photos	photo	L01 / L03	OK	—	—
hawaii_rock_photo, lava_photo, canyon_photo, texas_hill_photo	photo	intro–L04	minor	"plain background" on a scene	split suffix
granite_quarry_photo	photo	none	DELETE	paradox if placed	—

Three best: the cross-section pair, jar_shaken → jar_settled, steps_igneous card 2. Three weakest: q_card_k, q_grains_big / pp_grains_big, jar_settled.

Blind read and v2 re-read (2 October 2026)

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

Every key agreed, bar one lettering slip: Reader B wrote D for PP-17 while reasoning "30 big cells = slow", which is the key (C, "It cooled slowly"); Reader A agreed with the key outright. 116 items each; both multi-selects (PP-41 B, D; PP-42 A, D) matched; the typed PP-40 "layers" matched; the FRQ model in the key's words. No item with two defensible options. All six drawings give data only. The intro poll is not bundled by design.

The biggest issue is reuse, not keys. Reader A: PP-02 is L01-01's Rock J card re-lettered; PP-12 is L01-02's Rock K; PP-10 is L03-05 as a card; PP-17's drawing is L02-06's data; PP-43 to PP-48 are the six quiz claim scenes re-cased as "correct reason" items; three pumice items (PR-04, PP-04, PP-07); PP-23 = L02-11 with lake for sea; PR-11 = PR-10 with bay for sea. The fresh-instance rule would not pass these.

Flags carried into the fix pass

Item	Who	Flag	Ruling
PP-02, 12, 10, 17, 43–48, 23; PR-04 / PP-04 / PP-07; PR-11	A	lesson and quiz scenes renamed into the bank and practice	re-scene every one (fresh rock, fresh child, fresh move where the move is the quiz's)
rock names (granite, obsidian, conglomerate, sandstone, shale, limestone)	A, B	no bundle item defines them; L04-08, PP-30, 31, 32, 52 depend on shale, limestone, conglomerate	the lessons name them (67 mentions in the lesson file) — the guidance review confirms each is coined with its sediment before an item leans on it; add a which-sediment-makes-which-rock item if not
PP-40 typed	A, B	the stem says "bands" twice; L03 taught "stripes" for the same thing	reword the stem so it does not echo; accept "layers" and child spellings; rule once on "bands" / "stripes" (reject, with a sister-word why)
L04-03, PR-07, PP-27 / 28 / 29	A, B	settle → press → stick must be three distinct steps in the lesson text	confirm the lesson states the three; ⟂ PP-27 ⟂ PP-28 ⟂ PP-29
PP-19, PP-20	A	lean on "never layers"	"most likely" on PP-19; the guidance review rules on "never"
PP-38 ⟂ PP-39; PP-16 ⟂ PP-17; PR-10 / PR-11 duplicates; L03-10 / L03-11 answer each other	B	duplicates and give-aways	⟂ on the type sheet; re-scene PR-11; vary L03-11
"a mineral" as a distractor in 12 items	A	Topic 17's word, not coined here	fine (Topic 17 coins it); vary the distractor set

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 (116 items; MCQ disagreements 0; both multi-selects matched; the typed "layers" on the list; the FRQ model in the key's words). No item with two defensible options on the science. Flags to the polish pass: PP-46 is L03-09's painted-stripes cliff renamed (re-scene); PP-33's key is the longest and the only option carrying the lesson phrase "in flat layers" (cut the tail to "at the bottom of water"); L04-04's key is the only long, specific sentence (trim); PP-43, PP-48, PP-52 are the lesson's own cases with a new place or child (re-scene the three); PP-44 and PP-47 near L02-09 / L02-04 and L04-06 (vary the move); L02-08 "as cool as the ground around it" → "no longer hot from the lava". The rock names are coined in the lessons (67 mentions; confirmed by the guidance review). Reader's file: READER_C.md.

Topic 18 "How igneous and sedimentary rock form" — your read (2 October 2026)

Skim first

The lessons and their rules

The intro script (116 words, about 58 s)

The question types (from TYPES_G4T18.md)

The questions the author put to you (built as recommended)

The lessons

Applied without asking

Also on the site

After you answer

Grade 4 Topic 18 — How igneous and sedimentary rock form — build record (v1, 2 October 2026)

What this is, in plain words

Structural decisions (in place of a blueprint)

Questions for James

Design-check-sequence records

The key strings

The example sequences, written out

The rules applied

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

Checker: 0 FAIL, 24 WARN (python3 build_tools_check_topic.py grade4/topic18_build/content/G4T18.py)

Bank freshness: the scene-noun grep (re-performed before this record)

Names run (python3 build_tools_names.py Lopez Ian Isaac Jace → exit 0)

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

Figures (59: 43 SVG by code, 14 photo and 2 cartoon placeholders; author's review-figures self-check on fourteen renders)

Judgement calls, recorded

Owed

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

v2 — reviews applied (2 October 2026)

Finding → change

Counts and key strings

Checker results

Questions for James — built as recommended

Two more for James (register and scope, not this page)

Item ids whose stem, options, key, model, figure, blind text or accept list changed

Sentence-atomise, the new sentences

v3 — polish (2 October 2026, from the v2 re-read)

Checks and the bundle

Item ids changed in v3

Topic 18 — guidance, questions and accuracy review, 2 October 2026

Topic 18 — figure review, 2 October 2026

Topic 18 — How igneous and sedimentary rock form — blind read, 2 October 2026

Flags carried into the fix pass

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

The question types (from `TYPES_G4T18.md`)

Checker: 0 FAIL, 24 WARN (`python3 build_tools_check_topic.py grade4/topic18_build/content/G4T18.py`)

Names run (`python3 build_tools_names.py Lopez Ian Isaac Jace` → exit 0)

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