Topic 25 "Weather and climate: reading the records" — your read (2 October 2026)

Two readers answered 121 items cold and agreed every key. Their flags, all applied in v2: a scale item with two true options ("0 to 100" also fits every bar; the stem now asks for the numbers that reach 13 and leave the least empty space); three bank items that lean on two lines L02 had not said (both lines added to L02: anything shorter than year after year is weather; a sentence with no time word says what usually happens); a quiz key that printed the definition a later item asks for; a stem that stated the previous item's answer; ten bank stems that were quiz stems with a fourth option (re-scened); "pattern", "mild" and "weather records" used before they were explained; a rain graph with no unit; wider accept lists on the typed numbers.

Reader C then read the whole v2 bundle, 135 items, and agreed every key; the v1 number faults (a table saying 12 where the graph drew 11) are gone. One item would not ship: PP-42's weather key said "for ten years", the clue the lesson teaches as a climate signal (fixed). Two figures to confirm on the render (confirmed: Town M and Town N both say "Sunny days"; the two no-label graphs carry no text naming what is missing). Four soft notes, all applied in v3: PP-20's "one whole winter" is now one week; L01-02's "pattern" distractor is "season"; "millimeters" is now in PP-32's stem; Lorena is now Bianca.

A sound topic, the reviewer said: the coin is Grade 3's "typical weather" sentence with "season" widened to "place"; the graph words are Grade 3's exactly; the made-up records are plausible and never called real; every misconception is met; one atom per item. Then the faults: three bank whys told the student the wrong option was right; one stem named its own key ("a weather reporter" over the key weather); one item had two true options; the video call said 68 degrees while the January table said 62; two videos skipped a step their items test (one month of records is still weather; the year-by-year subtraction); three articles did two jobs. All fixed in v2; the two long articles became four lessons. On the author's four questions the reviewer said: split SR4-012 out; keep "mild" and explain it; go back to one colour per quantity; keep the gridline split and say a line's worth once.

The graphs are clean and honest: every bar lands on a numbered line, the Duluth graph grows step by step with the lines ruled before the first bar, and Lincoln's wrong 0-to-2 graph is plainly wrong beside the 0-to-12 redraw. Five fixes, all made in v2: three rule cards whose text ran through their borders (a wrapper now measures the longest line); the tan graph always had fewer (one colour per quantity now); the headache figure carried its verdict chip on first showing (now shown bare, the chip after the text names the fault); crowded side numbers on four read graphs (numbered every 2 now); two-place text too small at phone width (larger, checked in a 360-pixel frame). Ten unplaced figures were placed or deleted (86 figures now).

52 items in content/G4T25_pp100_bank.py: 47 four-option MCQs, one two-of-four select and four typed numbers (a difference in lines; a difference in days; a sum then a difference; a count of years). Per atom: EAR4-070 8 · EAR4-071 12 · SR4-010 8 · SR4-011 8 · SR4-012 7 · EAR4-072 9. Easy 15, medium 25, hard 12, sorted easy to hard and rising with the atom's place in the topic. MCQ keys A 11 · B 12 · C 13 · D 11. Ten question types, one exemplar each on the type sheet: which word names the usual weather; which of four sentences tells the climate (or the weather); what a one-week record is about; which words go up the side; how many lines higher (typed); which town had more (two drawn graphs); month by month, the lead changes; the result that does not fit (drawn); the climate sentence from five years (drawn); judge a claim that one low year unmakes the climate (drawn). Every scene is fresh against the lessons, the quizzes and the practice (the scene-noun grep is in the build record).

Skills in full: course-build → write-lesson (L01 a concept introduction that coins "climate"; L02 an example-sequence classification — weather or climate?; L03 a graph-construct routine; L04 a graph-compare routine; L05 the capstone routine with SR4-012 inside it) → write-example-sequence (L02), design-check-sequence, sentence-atomise, write-explanation, write-mcqs (attached questions, quizzes, practice, the bank and its type sheet); write-video-script (the intro, five lesson scripts, the summary; the elementary micro-lesson anatomy) — all at vault commit 3a2c39d.

Grade 4 Topic 25 — Weather and climate — build record (v1, 2 October 2026)

What this is, in plain words

Five short lessons in one class: Miss Jensen's, in Austin, Texas, with Lincoln, Logan and Lorena, and a partner class in Duluth, Minnesota. On a January video call, snow falls behind the Duluth class while Austin sits at 68 degrees; Lincoln says Duluth is a cold place, Logan says that is just today's weather — and both are right, about two different things. That is the whole topic. The child learns that the weather is one day's sky and air, and that the weather a place usually has, year after year, is its climate (Grade 3's "typical weather", widened from a season to a place and given its name). Then the child sorts sentences by asking one question — one day or a few days, or year after year? — and meets the two traps (a cold week is still weather; one month of one year's records is still weather). Then the investigation: the Duluth class's five Januaries become a bar graph, built step by step the way Olivia taught Grade 3 (number every line, rule it right across before any bar), with both labels; Austin's graph is drawn on the wrong scale first, so one snowy day looks as tall as Duluth's bars — the fault is named and the graph redrawn on the same numbers; and last, the climate is read off the records (find the weather that came every year), with one freezing Austin afternoon checked and then set aside, because one odd day does not change the climate.

Decisions made without a blueprint — one line each, with the reason

• Five lessons for six atoms: SR4-012 (the result that does not fit) is a STEP inside the capstone EAR4-072, because the capstone's own misconception probe ("it froze today, so Austin has a cold climate") IS the odd-result step; a separate lesson would teach the same point twice. The bank tags SR4-012 items on their own (7 items), so the atom is still diagnosed. James, 25 Sep 2026: SR atoms are taught inside the investigation that needs them.
• SR4-010 and SR4-011 keep their own lessons but are written as steps of the climate investigation (graph the Duluth records; compare with Austin), never as abstract graph lessons with toy cars — each is one routine with one target behaviour.
• Atom order kept (070 → 071 → 010 → 011 → 012 → 072): every prerequisite edge points backwards; the coin (L01) precedes the sort (L02), and both precede the data work.
• One running scene, two school weather tables (Austin and Duluth, 2022–2026), made-up class records: the numbers are plausible for the two towns (Duluth 8–11 snowy school days a January; Austin 0–1; an Austin January afternoon in the sixties; one cold-snap day at 28) and are named as made-up class records, never as station data. Chosen so the two places differ plainly, as the brief asked (a Texas town and a Minnesota town).
• "Snowy days each January" as the five-year quantity, counts out of about twenty school days: a class can only record school days (Grade 3 Topic 23 counted rainy school days the same way); counts graph on a 0–12 scale with every line numbered, which is where Olivia's technique bites; July data was rejected because school is out.
• Temperature appears as the daily record (L05) and on the two drawn thermometers (L01), in degrees Fahrenheit: the brief asked for Topic 6's convention; readings 22, 28 and 68 sit on the drawn scale (20 to 110).
• The intro's scene is NOT the video call: a T-shirt morning in Austin and the evening news from Duluth pose the same question without naming L01's example (James, 26 Sep: an intro never hands over to the first lesson's example).
• The headache that could really happen: a partner-class video call; a teacher writing six sentences on the board; a child wanting the five Januaries in one picture; a child drawing a second graph on its own small scale; a teacher asking for one climate sentence. No invented sign or game.
• L04's wrong drawing is called wrong: Lincoln's first try (Austin on 0–2 beside Duluth on 0–12) is drawn, named as the fault ("the numbers up the side are different"), and redrawn — the scale-mismatch misconception is the lesson's headache, not a footnote.
• Two colours in the comparison figures (Duluth pale blue, Austin tan): the brief allows two where the comparison needs them; the register fixes which place takes which, and item pairs follow (first town the quantity's colour, second town tan).
• Graph words stay Grade 3's ("up the side", "along the bottom"), no "axis": the AP rule R188 (x-axis / y-axis) is for AP; Grade 3 coined these words and Topic 25 recalls them word for word, adding the two LABELS as the Grade 4 face ("A bar graph also needs two labels…", joined with "also").
• "climate" and "weather" in RECALL_BY_MCQ: the brief's instruction; both are recognised among sister words (weather / climate / season / temperature), never typed. The hard tier's typed items are therefore all numeric (a difference, a sum-then-difference, a count).
• Austin's climate sentence names "mild winters" and "hot summers": "mild" is the plain word for a winter that is neither cold nor hot; S36 flags it (answered below).
• The Grade 3 fossil-style "recall by one line and one item" is applied to five Grade 3 sentences (two places, typical weather, pattern, read a bar, make a graph) and two Grade 3 skills in the practice (read a thermometer, find a value in a table).

Questions for James (the build does not wait)

1. Should SR4-012 ("the result that doesn't fit") have its own lesson rather than living inside the capstone? Context: the architecture lists it as its own step; here it is L05's middle part (the 28-degree afternoon), with seven bank items of its own. Recommendation: keep it inside — the odd day is the capstone's natural misconception probe, and a sixth lesson would repeat the point. Built: five lessons.
2. Is "mild" the right word for Austin's winters, or do you want "warm winters" / "winters that are not cold"? Context: S36 flags it eleven times; a nine-year-old in Texas hears "mild" on the weather report. Recommendation: keep "mild"; add it to the course's word additions. Built: "mild winters with almost no snow".
3. Two colours for two places in the comparison graphs (Duluth pale blue, Austin tan) — or one colour per quantity as Grade 3's register says? Context: the brief allows two where the comparison needs them; the eye separates the two sets faster with two. Recommendation: two, as built, with the register line fixing which place takes which.
4. Every gridline numbered on the graphs children build (0 to 12), but only every 10 on the temperature graphs they read (0 to 80, a line every 2): is that split right, or should the read graphs also number every line? Recommendation: keep the split — Olivia's technique is for construction; Grade 3 L01 taught reading between numbered lines. Built as the split.

Counts

Five lessons · 49 quiz items (10 / 10 / 10 / 9 / 10) · 5 attached lesson questions + 3 summary questions · 12 practice items (3 Grade 3 retrievals, 1 typed numeric, 1 written) · PP100 bank 52 items (47 four-option MCQs, 1 multi-select, 4 typed numeric) · per atom EAR4-070 8 · EAR4-071 12 · SR4-010 8 · SR4-011 8 · SR4-012 7 · EAR4-072 9 · easy 15 / medium 25 / hard 12 · 88 figures (78 SVG by code, 5 photo placeholders, 5 cartoon placeholders) · 7 videos in the batch (intro 117 words, lessons 145 / 149 / 148 / 147 / 146, summary 171).

Articles (student words, headings included): L01 336 · L02 342 · L03 390 · L04 274 · L05 375. Three sit above the 300-word WARN line: L03 is a five-step routine with a figure per step (nine figures, a subtitle every ≤ 6 lines); L02 is a six-card example sequence (one card per case, a subtitle every ≤ 7 lines); L05 carries the worked case, the second case and the odd day, each under its own subtitle with its figure (a figure or subtitle every ≤ 5 lines); L01 was trimmed from 375 by cutting three restatements. No stretch of idea lines runs past eight without a subtitle or a figure.

The per-lesson table

Every lesson gets a video: each carries a drawing that changes in place (thermometers, cards with verdict chips, a graph building bar by bar, two graphs rescaled, a bar chipped DOES NOT FIT), which is what the video does better than the page.

#### design-check-sequence records

• L01 — type (c) what / why / misconception / transfer. Recall (Grade 3, two places one day) → which word (weather) → which word (climate) → why many years → claim: climate = weather today (No) → apply (a forecast → weather) → apply (a book line → climate) → transfer (what tells a climate) → the word pattern → term recognition last. Each quiz item names its script beat: beats 4 (weather def), 5–6 (the table, year after year), 7 (the coin), 8 (the two climates).
• L02 — type (a) fluency on the fixed set [the weather, the climate]. Eight fresh sentences, each a different clue (every year; this morning; tomorrow; most years; year after year; last Tuesday; all this week — the near miss; most winters), then the hot-place claim (No), then the clue-word item last. The frame question is spoken in beat 2 and every verdict in beats 3–7.
• L03 — type (b) scaffolded → bare. Grade 3 recall (read a bar) → three scaffolds on the Reno table (side words; bottom words; before the first bar) → how high the bar goes → drawn wrong bar → claim (tallest bar = snowiest day, No) → bare: tallest month on a fresh graph; what is missing on a fresh graph → why the bottom label. Beats 2–6 carry the steps.
• L04 — type (b) scaffolded → bare. Recall (L03 side label) → two drawn comparisons (more / fewer) → typed bar-by-bar difference (12 − 1 = 11) → the mismatch claim (No) → what to check first → every-month item → which shows less → the more-bars claim (No). Beats 2–5.
• L05 — type (b) with the claims. Recall (the term) → drawn five-year graph in two steps (what every May had; the climate phrase) → the routine's step → drawn daily record (which does not fit) → what to do first → claim: one cold afternoon = cold climate (No) → TRUE claim: one snowy day in a dry town does not change the climate (Yes) → fresh drawn capstone (Bangor) → written climate sentence. Beats 2–3 (the routine), 4–6 (the odd day).

The key strings

• Attached questions in video order (L01–L05, then the three summary questions): B A A C C A A B (A 4 of 8 = 50 %).
• L01: two-option A B A B; three-option A B A B A; four-option B. L02: two-option B A A B B A A B B (no four in a row alternate; 4 and 5); four-option B. L03: three-option C A B A B C B A A; two-option B. L04: three-option A A B A B B; two-option B B; numeric 11. L05: three-option B A B C C A; two-option B A; four-option C. Practice: three-option B B C B C A A B B; two-option A; numeric 10.
• Bank (47 four-option MCQs): A 12 · B 12 · C 12 · D 11. Multi keys [0, 2]. Numeric accepts: 9, 4, 30, 5 (bare number, word, and the unit forms).

The example sequence (L02), written out

Frame, identical on every card: One day or a few days — or year after year? E1 "It rained in Austin yesterday." → yesterday, one day → WEATHER. E2 "Austin has hot summers most years." → most years → CLIMATE. E3 "A storm is coming to Duluth this afternoon." → part of one day → WEATHER. E4 "Duluth's winters are long and snowy." → no time word; "winters are" says every winter → CLIMATE. NE1 (near miss) "Austin has been cold all this week." → a week is a few days, not year after year → WEATHER. NE2 (near miss) "Last January, Duluth had 11 snowy days." → one month of one year → WEATHER. Only the sentence changes from card to card; the verdict chips are the lexicon's two words; the clue words are listed after the run; the rule comes last. Lorena's belief ("a week must be climate because it lasted more than one day") is confronted by name between NE1 and the clue words.

Worked cases: L03 builds Duluth (9, 11, 8, 10, 9) in five steps with a figure each; L04 draws Austin on the wrong scale, then the right one, then one bar-by-bar line (11 − 1 = 10); L05 reads Duluth (every January 8 or more → cold and snowy), then Austin (1 or fewer → mild, almost no snow), then the 28-degree afternoon among seven afternoons of 58 or warmer.

The rules applied

Misconception rows (the physics catalogue is silent on weather; the brief's four beliefs, with the literature named)

Checker: 0 FAIL, 4 WARN (python3 build_tools_check_topic.py grade4/topic25_build/content/G4T25.py)

Fixed on the first run (10 FAIL → 0): one 24-word idea line split; L02's script 163 → 149 words; the L02 closing bullets' colon fragments; four why lines in PP-45 / PP-46 written as sentences; two three-sentence stems merged (PP-12, PP-29); and the first run's S36 words frosty / forecast / width / mentions / sunlit / replace swapped for cold / temperature / how wide / says / hot glass / used instead.

python3 grade4/check_g4_articles.py grade4/topic25_build/content/G4T25.py → 0 FAIL (articles 336 / 342 / 390 / 274 / 375 — the four WARNs answered under Counts; scripts 145 / 149 / 148 / 147 / 146; intro 117 with 2 question marks; summary 171, opener ok).

Names run (python3 build_tools_names.py Jensen Lincoln Logan Lorena → exit 0, all four free)

Miss Jensen; Lincoln, Logan, Lorena — the Topic 25 row of grade4/NAMES_POOL.md. No other name appears in a student string; the Duluth class and its teacher are unnamed; items use "a student".

Bank freshness (re-performed; the scene-noun grep)

Scene nouns of every lesson, quiz, attached and practice item: Austin, Duluth, Texas, Minnesota, the video call, the partner class, the board, Lorena's graph, Lincoln's two graphs, January 15, the cold snap; quiz / practice towns Denver, Miami, Nashville, Phoenix, Fairbanks, Chicago, San Francisco, Atlanta, Seattle, Las Vegas, Dallas, Boston, Buffalo, Houston, Omaha, Reno, Lexington, Tulsa, Portland, Amarillo, Savannah, Fargo, Des Moines, Towns J K M N, Yuma, Memphis, Bangor, Boise, El Paso, Sacramento, Spokane, San Diego, Billings, Charleston. Grep of the bank's stems and options: Austin 0 · Duluth 0 · Minnesota 0 · video 0 · board 0 · Lincoln 0 · Logan 0 · Lorena 1 (the one bank claim-maker, PP-49) · every quiz / practice town 0 after the re-scene pass (Buffalo → Syracuse; Spokane → Flagstaff; Savannah → Charlotte; Phoenix → Bakersfield; Boise → Helena; Towns J/K → V/W; Towns L/M → G/H) · Texas 1 (Corpus Christi, Texas — the state, not the lesson's town; kept). Bank towns: Cleveland, Syracuse, Tampa, Wichita, Juneau, Little Rock, Richmond, Milwaukee, Flagstaff, Charlotte, Madison, Bakersfield, Honolulu, Helena, Towns P R V W G H, Albuquerque, Baltimore, Jackson, Raleigh, Olympia, Las Cruces, Laredo, Tucson, Anchorage, Corpus Christi, Bismarck, Pensacola.

Blind bundle

grade4/topic25_build/blind/2026-10-02/BLIND_BUNDLE_G4T25.md (112 MCQ, 6 numeric, 2 written, 1 multi; keys sequestered in KEYS_SEQUESTERED_G4T25.json). Every [Drawing: …] line is data ("bar graph; the axis is numbered every 1 from 0 to 14 with a gridline every 1; Dec: top on the 10 line; …"; "card with the sentence: Chicago, Illinois, gets cold, windy winters every year."); grep of the drawing lines for climate / weather / hotter / colder / wetter / drier / usual → 0 hits. The sentence cards' blind text is the sentence itself (the item's data); the sort is the student's.

Video batch (built, checked, NOT sent)

python3 build_tools_video_batch.py grade4/topic25_build/content/G4T25.py → handoff/G4T25/videos/batch_G4T25.json (7 videos: intro 117 words, L01–L05 145 / 149 / 148 / 147 / 146, summary 171) and READY.md; check_brief.py: L01–L05 and the summary ✓, the intro's one standing hit (no closing bullets — an intro has no rule slide). grep -ci strand → 0. Nothing has been sent.

Sentence-atomise self-check (one row per lesson; every paragraph read; S1 / S38 clean)

Figures (88; author's review-figures self-check on 25 renders, a non-author's pass owed)

Rendered with headless Chrome at 1000 px and opened: build_side, build_bottom, build_grid, build_bar1, build_done_tallest, compare_mismatch, compare_same_verdict, austin_jan_graph_odd, therm_call, therm_28, duluth_table, austin_jan_table, q_table_reno, card_2, card_4, sort_table, board_six, rule_make_graph, q_tulsa_nolabel, pp_g5, pp_odd2, weather_climate_card, q_cmp_mismatch_jk, plus the five built article pages (L01–L05) and the PP100 route as full-page screenshots. Fixed on the render: (1) the DOES NOT FIT and THE MOST SNOWY DAYS chips sat on the bottom-label row and overprinted "Day in January 2026" / "Year" — moved to just above their bar (above=); (2) the records table's title ran outside its frame (Topic 18's table sizes from its columns) — a title-aware wrapper widens the frame to the measured title; (3) the step-1 "title only" figure was a 340 px empty frame — now a 70 px strip. The Jackson rain dataset failed the drawer's gridline assert (values not on a line) and was redrawn with a line every 1, numbered every 5. Every bar sits on a gridline (asserted at import); every two-place figure's two scales are read side by side; the thermometer readings 22, 28 and 68 are on the drawn scale.

Depiction register in figures.py's docstring (the Grade 4 bar graph with two labels; the two-place graph; the records table; the sentence card; the sort table; the thermometer pair; colours per quantity and per place).

Judgement calls, recorded

1. SR4-012 inside the capstone, not a sixth lesson (question 1).
2. "mild" kept (question 2).
3. Two colours for two places (question 3) and every line numbered on built graphs only (question 4).
4. Austin's own snowy-day graph has four empty years: drawn as "no bar" and said so ("the other four years have no bar at all"), which is the honest picture and the one Grade 3's bars_desc describes ("no bar rises above the bottom line").
5. The L03 script speaks Grade 3's rule, not the "also" sentence, for the 150-word budget; beat 3 says "the label year, saying what each bar stands for", and the article and summary carry the "also" sentence word for word.
6. The L05 lexicon holds two sentences (RECORDS_RULE, ODD_RULE); the check step ("A result that does not fit the rest is checked, then set aside.") sits in the rule card and the script but not the lexicon, so the summary stays at 171 words.
7. The odd day is REAL, not a mistake: the check step still runs ("was it a mistake?"), the thermometer hung in the shade as always and the news reported the cold snap; the bank carries both kinds (a sun-struck 96, a real 44).
8. L04's key-length tell: the true month-by-month sentence (PP-45) is the longest option because it has two halves; recorded on the type sheet for the blind read.
9. Written to the topic folder and the tools' own output paths only: the page to /tmp/g4t25.html, the figures to /tmp/g4t25_figs/, the batch to handoff/G4T25/videos/ (NOT sent), the bundle to grade4/topic25_build/blind/2026-10-02/. Nothing committed.

Owed

• A non-author review-figures pass and review-questions pass; the blind read of the bundle (keys unverified until a cold reader agrees).
• The picture pass (10 placeholders) after James answers the questions above.
• James's word before the batch is sent (Grade 4's batch rule).
• The S36 additions: if James accepts "mild", it goes to reference/grade3_word_additions.txt (not written here — outside this topic's folder).

v2 — reviews applied (2 October 2026)

In plain words. The three reviews found a sound topic with a handful of real faults: three bank whys that told the student the wrong option was right, one stem that named its own key, one two-answer item, a scene that disagreed with its own table (the video call said 68, the January table said 62), two videos that skipped a step their items test, three rule cards whose text ran through their borders, and a colour habit — the tan graph always had fewer — that let colour predict the key. All are fixed. The two longest articles are now two lessons each, with a quiz between: the topic is seven lessons, and every article is under about 300 words except L01 (311) and L02 (346, a six-card example sequence). Blind bundle re-cut; nothing sent; nothing committed.

#### Finding → change

#### The seven lessons

Counts: 61 quiz items · 7 attached + 3 summary questions · 12 practice · bank 52 (47 four-option MCQs, 1 multi, 4 numeric; per atom EAR4-070 8 · EAR4-071 12 · SR4-010 8 · SR4-011 8 · SR4-012 7 · EAR4-072 9; easy 15 / medium 25 / hard 12; keys A 11 · B 12 · C 13 · D 11) · 86 figures (79 SVG, 2 photo and 5 cartoon placeholders) · 9 videos in the batch (intro 117, L01–L07 150 / 149 / 128 / 135 / 150 / 117 / 147, summary 174). Lexicon: + LINES_RULE (L04); the summary speaks one rule beat per lesson (174 words).

#### Key strings

• Attached, video order (L01–L07, summary ×3): B A A A C A C A A B.
• L01 two-option A B A B; three-option A B A B A; four-option B. L02 two-option B A A B B A A B B; four-option B. L03 three-option C A B A B A A C. L04 three-option A A B C B C B; yes/no B. L05 three-option A A B A B B; yes/no B B; numeric 11. L06 three-option B A B A C A; four-option C; FRQ. L07 three-option A C A B B B; yes/no B A. Practice three-option B B C B C A A B B; yes/no A; numeric 10.
• Bank numeric: 9, 4, 32, 3.

#### Checker results

• build_tools_check_topic.py → 0 FAIL, 5 WARN (S9 L04 rule words missing from the script — the rule step carries Grade 3's recalled sentence, the script speaks L04's own rule, as v1's call 5; S29 the drawn thermometers' chips; S36 mild ×11, student ×7, tv ×2 — as v1).
• check_g4_articles.py → 0 FAIL; articles L01 311 · L02 346 · L03 245 · L04 250 · L05 277 · L06 216 · L07 253; intro 117, summary 174.
• build_tools_names.py Jensen Lincoln Logan Lorena → exit 0.
• check_svg_ids.py → PASS (86 figures, 0 ids).
• check_key_tells.py → KEY TELLS: 0 of 127 MCQs flagged.
• Site built to /tmp/g4t25_v2.html; every lesson's article route screenshotted at 1200 px and in a true 360 px iframe (Chrome's bare window will not go below ~500 px, so the 360 px shots are framed) and looked at; 27 figure renders opened at 1000 px, bars counted against their tables after the colour change (Duluth 9 11 8 10 9 / Austin 0 1 0 0 0; Portland 14 12 10 / Phoenix 2 1 0; Erie 11 13 12 / Atlanta 1 2 1; Milwaukee 10 12 11 / Little Rock 1 2 0; Albuquerque 13 14 12 / Cleveland 6 5 7; Town J 3 2 4 / Town K 10 12 9 — all match).
• Batch: build_tools_video_batch.py → 9 videos; check_brief.py ✓ on L01–L07 and the summary, the intro's one standing line. Not sent.
• Blind bundle: blind/2026-10-02-v2/BLIND_BUNDLE_G4T25.md — 126 MCQ, 6 numeric, 2 FRQ, 1 multi (135 held). grep of its [Drawing: lines for climate | weather | hotter | colder | wetter | drier | usual | more | fewer → 0 hits (every drawing line is a title, a label and bar heights as numbers).

#### Questions for James — built as recommended

1. SR4-012 is the second half of the split capstone: L07 "The day that does not fit", its own lesson with the check between.
2. "mild" kept and explained on first use: "mild winters: cool days, with almost no snow".
3. One colour per quantity on every two-place graph (the figure review agreed — tan tracked the key).
4. The gridline split kept, and the line's worth said once in L07 before the temperature graph.

#### Item ids changed (stem, options, key, model, figure, blind text or accept)

• Lessons: L01-03, L01-04, L01-09 · L02-06 · L03-01…L03-08 (new lesson; L03-04, L03-05, L03-07 new items; L03-08 options reordered) · L04-01…L04-08 and L04-V-01 (new lesson; L04-01, L04-07, L04-08 new; L04-02 key trimmed) · L05-01 (stem and options), L05-04 (accept) · L06-03 (stem), L06-04 (new), L06-06 (new, q_fresno_jul), L06-V-01 (new) · L07-01 (new), L07-04 (new), L07-07 (new, q_lubbock_oct), L07-08 (new); the rest of L06 / L07 are the old L05 items renumbered.
• Practice: PR-05 (stem), PR-09 (accept).
• Bank: PP-1, 2, 3, 4, 5, 6, 8, 12, 13, 15, 16, 21, 22, 25 (blind text), 29 (accept), 30, 32 (blind text), 33, 35, 39, 43, 44 (accept), 45, 46 (figure, stem, accept), 47, 48, 49, 51, 52 (figure, stem, accept).
• Figures renamed / added: rule_make_graph → rule_two_labels + rule_lines; rule_records split → rule_records + rule_odd; compare_mismatch_verdict, pp_c6, pp_r6, q_fresno_jul, q_lubbock_oct added.

#### Sentence-atomise, v2 lines

Every new idea line read: one idea each, ≤ 22 words (S1 0), subject and verb (S38 0); the one 23-word line ("A day, a week, a month or one whole winter…") split at the "so". New fb / why lines one causal link per <br> (S5 0). Not committed.

v3 — polish (2 October 2026)

In plain words. A third cold reader (Reader C) read the whole v2 bundle, 126 MCQs and the typed and written items, and agreed every key; the v1 number faults are gone. She left one item that would not ship (PP-42: the weather key said "for ten years", the very clue the lesson teaches as a climate signal), two figures to confirm on the render, and four soft notes. All are applied; no key letter, no key meaning, no tag position and no figure changed. One of the three children has a new name: Lorena is now Bianca, so Logan and Lorena can no longer be read as one child on neighbouring items. Blind bundle re-cut; batch rebuilt, not sent; nothing committed.

#### Flag → change

#### Checks after v3

• python3 build_tools_check_topic.py grade4/topic25_build/content/G4T25.py → 0 FAIL, 5 WARN (the same five as v2: S9 L04's rule step is Grade 3's recalled sentence; S29 the drawn thermometers' chips; S36 mild ×11, student ×7, tv ×2).
• python3 grade4/check_g4_articles.py grade4/topic25_build/content/G4T25.py → 0 FAIL (articles 311 / 346 / 245 / 250 / 277 / 216 / 253; scripts 150 / 149 / 128 / 135 / 150 / 117 / 147; intro 117; summary 174, opener ok).
• python3 build_tools_build_topic_site.py grade4/topic25_build/content/G4T25.py /tmp/g4t25_p.html → 7 lessons, 12 practice, 52 pp100, 86 figures.
• python3 ~/Documents/CourseBuilding-Vault/skills/review-figures/check_svg_ids.py grade4/topic25_build/content/figures.py → PASS (86 figures, 0 ids).
• python3 ~/Documents/CourseBuilding-Vault/skills/write-mcqs/check_key_tells.py grade4/topic25_build/content/G4T25.py → KEY TELLS: 0 of 127 MCQs flagged (a: 0, b: 0, c: 0); KEY LONGEST (any margin) 31 of 127.
• python3 build_tools_names.py Jensen Lincoln Logan Bianca → all four free, exit 0.
• python3 build_tools_video_batch.py … → 9 videos, key string B A A A C A C A A B; check_brief.py → every video ✓ except the intro's standing closing-bullets line. Not sent. (Rebuilt because the name sits in the blurb, intro beat 1 and four lesson scripts.) grep -c Lorena batch_G4T25.json → 0; grep -ci strand → 0.
• Blind bundle re-cut with build_tools_blind_bundle.py to blind/2026-10-02-v3/: 126 MCQ, 6 numeric, 2 FRQ, 1 multi (135 held); keys sequestered. grep '^\[Drawing:' | grep -icE 'climate|weather|hotter|colder|wetter|drier|usual|more|fewer|missing' → 0 of 49 drawing lines. "Lorena" → 0 hits.
• The three figures confirmed on headless-Chrome renders at 900 px (/tmp/g4t25_v3figs/q_cmp_mn.png, q_tulsa_nolabel.png, pp_g5.png), opened and looked at.

#### Counts and key strings after v3

Unchanged from v2: 7 lessons · 61 quiz items · 7 attached + 3 summary questions · 12 practice (10 MCQ, 1 numeric, 1 written) · bank 52 (47 four-option MCQs, 1 two-of-four select, 4 typed numeric): EAR4-070 8 · 071 12 · SR4-010 8 · 011 8 · 012 7 · EAR4-072 9; easy 15 / medium 25 / hard 12; bank MCQ keys A 11 · B 12 · C 13 · D 11; every quiz, practice and attached key string as in v2 (no option moved position).

#### Item ids changed in v3 (for the mini re-read)

Lesson quizzes: L01-02 (option C, fb C) · L01-05 (claim-maker's name) · L04-V-01 (name) · L05-05 (name). Bank: PP-20 (stem, options, why) · PP-32 (stem) · PP-42 (option C, why) · PP-49 (name). Lesson text, not items: the blurb, intro beat 1, L02's two belief lines, L03 and L04's headache, article and script lines, L06's article and script lines, L07's article lines (name only). Figures: none redrawn; lorena_graph_cartoon renamed bianca_graph_cartoon, its note and brief and the two other cartoon briefs say Bianca. Records: TYPES_G4T25.md (Type 3's line and exemplar; the name), grade4/NAMES_POOL.md (Topic 25 row).

#### Judgement calls, v3

1. PP-42 keeps its hard tag and its place. Its old "hard because" rested on the ten-years decoy, which is gone. The item is still the bank's one sort whose weather key is a MONTH (L02's near miss) against three distractors with three different climate clues, which is a step past PP-19's "all this week". A retag would move the item and renumber the bank for no gain in diagnosis.
2. PP-20 is a week, not a day, so the bank keeps one "few days" record item beside PP-21's "today"; PP-19 is also a week but asks a different move (which of four sentences tells the weather).
3. "millimeters" stays on PP-32's drawing (the honest unit of rain per day, Grade 3 Topic 22's) with the stem carrying it, rather than a change of quantity.

#### Owed after v3

• A mini re-read of the eight changed items above (blind/2026-10-02-v3/), by a reader who has not seen the keys.
• James's questions on the Needs-you page (records/REVIEW_NEXT_G4T25.md → course_preview/review/REVIEW_NEXT_G4T25.html); the picture pass (7 placeholders) waits on them.
• James's notes on the scripts are the cue to send the nine-video batch.

Nothing committed.

Topic 25 — guidance, questions and accuracy review, 2 October 2026

A sound topic: the coin is Grade 3's "typical weather" sentence word for word with "season" widened to "place"; the graph words and sentences are Grade 3's exactly; the invented records are plausible and never called real; every misconception is met; one atom per item, keys spread, tags rising, every typed value on a numbered line, item alt and blind text numbers only — all re-performed and true; 0 FAIL. But three bank whys tell the student the wrong option is right, one bank stem names its own key, one item has two true options, the scene contradicts its own dataset, and two videos skip a step their items test.

The author's four questions, the reviewer's view: (1) SR4-012 inside the capstone — agree, as the second half of a split L05 with a check between; (2) "mild" — keep, with its plain meaning said once; (3) two colours for two places — go back to ONE colour per quantity (tan is the lower set in eight of nine graphs, so a child learns "tan means fewer"; colour never carries a key, so no item breaks either way); (4) the gridline split — keep, but say what a line is worth once.

Topic 25 — figure review, 2 October 2026

The graphs are clean and honest: every bar lands on a numbered line, the Duluth graph grows step by step with the lines ruled before the first bar, and Lincoln's wrong 0-to-2 graph is plainly wrong beside the 0-to-12 redraw. Then the eye snags on three yellow rule cards whose first lines run through the card border on both sides. Page pass PASS (13 routes, light and dark; every figure renders as alone; check_svg_ids.py PASS). Bars counted against six tables — all match; thermometers read by the taught routine; the five numeric items each give one reading; blind texts numbers only; data captioned as class weather tables.

1. Three rule cards overflow their card (rule_make_graph, rule_compare, rule_records; Topic 18's rule_card has a fixed 460 width). A local wrapper widening w to the longest measured line + 48, as the author already did for table().
2. Tan always loses. In every fair two-place item the quantity-coloured left graph has more and the tan right graph has fewer (eight of nine). Colour predicts the key. Put the place with more on the right in about half the items — or, with the guidance review's recommendation, go back to one colour per quantity.
3. The headache hands over its answer: compare_mismatch carries the chip "THE NUMBERS UP THE SIDE ARE DIFFERENT" at its first showing, two sentences before the text names the fault. First show with verdict=None.
4. Crowded side numbers: q_yuma_may (0–18) and pp_r1 / r2 / r3 (0–20) number every line and glyphs overlap. These are read, not built: num=2.
5. Two-place graphs at phone width: 740-unit frames with 12-unit text give ~6 px numbers in a 360 px column. Larger text inside two_graphs, checked at 360 px.

Unplaced (10): austin_jan_table — PLACE before austin_jan_graph_plain in L05 (table → graph is the routine); pp_g3, pp_g4 — place under an easy read-one-graph SR4-010 item (the bank has none) or delete; pp_g1, pp_g2, therm_call_plain, austin_graph, storm_sky_photo, rain_gauge_photo, desert_town_photo — delete.

Minor: therm_28 alt names an undrawn 32 mark; the DOES NOT FIT chip touches neighbouring bars; THE MOST SNOWY DAYS chip spans two columns; q_tulsa_nolabel / pp_g5 blind texts say "no label" (near the option wording); lincoln_compare_cartoon brief vs the half-way bar; the CART suffix's "No text" where graphs need written numbers ("marks standing for writing"); the placeholder prints "; the pipeline draws it" and stacks two captions.

Three best: the build_* sequence, compare_mismatch → compare_same, austin_jan_graph_plain. Three weakest: the three overflowing rule cards, the two-place items, pp_r1–r3's packed numbers.

Topic 25 — Weather and climate — blind read, 2 October 2026

Every key agreed by two blind readers (121 items each; MCQ disagreements 0 for both; the six typed numbers matched; PP-23 multi A, C matched; both written models in the key's words). Every graph and table read by its numbers gave one answer; every odd-day item puts the odd reading far from the rest with the record on the graph. One item both readers would not ship as written. The intro poll is not bundled by design.

Flags carried into the fix pass

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

Takeaway

Almost every item answers cleanly from the bundle alone, and the number faults the v1 read found are gone: every table the stems quote matches its drawing (Reno 10/12/14, Boise 9/7/6, Richmond 8/6/9), and every "day that does not fit" graph has exactly one odd reading. One PP100 item would not ship as written: PP-42's weather key says "for ten years", which is the very clue-word the lesson teaches as a climate signal, so a strong student is pushed off the right answer. Two softer points: the Town M / Town N drawing (L05-08) is described without naming what the bars count, so "less sun" is only answerable if the side label says Sunny days; and three child names start with L (Lorena, Lincoln, Logan), with Logan and Lorena close enough for a nine-year-old to blur.

Items

id | answer | one-line note

--- | --- | ---

L01-01 | A | yes; B absurd

L01-02 | C | weather; "pattern" appears as a distractor before L01-09 coins it (lesson may coin it first)

L01-03 | C | climate; clean

L01-04 | A | only A names year after year; B is a thermometer leftover, absurd

L01-05 | B | no

L01-06 | B | weather; "tomorrow" carries the time

L01-07 | B | climate; no time word, plural "Winters" carries the load (PP-43 tests the same idea)

L01-08 | B | records from ten years; clean

L01-09 | C | pattern

L01-10 | D | climate; same stem as L06-01, PR-06, L01-V-01 (fluency repetition)

L01-V-01 | C | climate; stem says "year after year" twice, fine for a gate

L02-01 | A | climate ("every year")

L02-02 | B | weather ("this morning")

L02-03 | A | weather ("tomorrow")

L02-04 | B | climate ("Most years")

L02-05 | B | climate ("year after year")

L02-06 | B | weather ("last Tuesday")

L02-07 | B | weather; "all this week" is more than one day — needs the lesson's "how long a time" rule, not L01's "one day"

L02-08 | A | climate ("most winters")

L02-09 | B | no

L02-10 | C | usually

L02-V-01 | B | weather

L03-01 | A | 9; drawing read

L03-02 | B | Sunny days

L03-03 | C | Month

L03-04 | A | B describes a position, absurd; C tempting-wrong

L03-05 | C | 0 to 14

L03-06 | C | label up the side; the description says "nothing written beside the numbers" — check the rendered figure carries no such caption

L03-07 | A | what the numbers count

L03-08 | C | clean

L03-V-01 | B | Rainy days

L04-01 | C | windy days

L04-02 | B | rule a line at every number

L04-03 | A | up to the 12 line

L04-04 | C | February drawn to 11, table says 14; table matches L03's Reno table

L04-05 | B | no; tallest bar = month, not day

L04-06 | B | April (12)

L04-07 | C | key is the longest option by a clause; A and B are short

L04-08 | A | four bars

L04-V-01 | B | a line at every number

L05-01 | B | what the numbers count

L05-02 | B | Portland

L05-03 | B | Savannah (fewer)

L05-04 | 11 more days | Portland 12, Phoenix 1 (middle line between 0 and 2)

L05-05 | B | no; J runs 0-4, K runs 0-14

L05-06 | A | numbers up the side the same

L05-07 | C | Fargo more every month

L05-08 | A | Town N (2, 3, 1) IF the bars count sunny days — the description never names the quantity; stem only says "less sun"

L05-09 | B | no

L05-V-01 | C | numbers up the side

L06-01 | A | climate

L06-02 | A | lowest bar is 13 (2024), so 13 or more; clean

L06-03 | A | usually hot

L06-04 | C | one year cannot show what came every year

L06-05 | C | the weather that came every year

L06-06 | C | usually dry (1, 0, 1, 2, 0)

L06-07 | C | usually snowy (6-9); D "nobody can tell from five years" is tempting-wrong — lesson must say five years is enough

L06-08 | "The town's summer climate is usually hot and dry." | award: hot and dry + usually / year after year

L06-V-01 | B | usually snowy

L07-01 | C | usually rainy

L07-02 | A | Oct 12 reads 40; the rest 68-74; exactly one odd day

L07-03 | C | check for a mistake first

L07-04 | B | a mistake

L07-05 | A | no; options run no/yes (reverse of other yes/no items) — fine, just noting

L07-06 | B | yes

L07-07 | A | 2024 (12 vs 3-5); one odd year

L07-08 | A | 95

L07-V-01 | A | usually rainy

SUMMARY-01 | B | weather

SUMMARY-02 | C | what the numbers count

SUMMARY-03 | A | no

PR-01 | C | 46 °F (3 marks of 2 above 40)

PR-02 | B | 7

PR-03 | A | 9; graph matches PR-02's table

PR-04 | C | "most years"

PR-05 | A | weather

PR-06 | A | climate

PR-07 | B | Month

PR-08 | C | Spokane

PR-09 | 10 more days | Spokane 10, San Diego 0

PR-10 | C | Nov 9 reads 66; the rest 36-42; one odd day

PR-11 | A | usually rainy (11-14)

PR-12 | "The town's winter climate does not change: one warm day is one day's weather, and the climate is what the winters are usually like, year after year." | award: no change + one day is weather + climate is many years

PP-01 | C | climate

PP-02 | D | weather

PP-03 | D | key is the longest option, though C is nearly as long

PP-04 | C | clean

PP-05 | C | "most years"

PP-06 | A | "this morning"

PP-07 | A | "most years"

PP-08 | D | most winters

PP-09 | D | today

PP-10 | B | Rainy days

PP-11 | C | the months; A/B are label text with capitals, C/D lowercase descriptions — mixed shape, but the pair split does not point at one

PP-12 | C | 0 to 14

PP-13 | A | rule a line at every number

PP-14 | C | Milwaukee

PP-15 | A | numbers up the side

PP-16 | C | clean

PP-17 | C | clean

PP-18 | A | climate

PP-19 | D | "all this week"; "mild" in distractor B may be unmet, harmless to the key

PP-20 | A | one winter's weather; D "Madison's climate" is tempting — a season-long record called weather is a stretch beyond L02-07's week, but the only fit

PP-21 | D | one day

PP-22 | C | clean

PP-23 | B, C | two-select, clean

PP-24 | B | August drawn 7, table 9

PP-25 | D | label along the bottom; description hands over ("nothing written under them") — check rendered figure

PP-26 | B | what the numbers count

PP-27 | C | Town R (4, 6, 3)

PP-28 | C | numbers up the side

PP-29 | 9 more days | Albuquerque 14, Cleveland 5

PP-30 | A | clean

PP-31 | A | Mar 10 reads 80; rest 48-54; one odd day

PP-32 | B | Thursday 24 mm vs 5-7; "millimeters" may be an unmet unit, harmless (heights decide)

PP-33 | B | check for a mistake

PP-34 | D | 2025 (11 vs 2-4)

PP-35 | A | checked, then set aside; matches PP-48

PP-36 | B | usually rainy (14-17)

PP-37 | B | usually dry (0-2)

PP-38 | A | usually hot

PP-39 | B | the weather that came every year

PP-40 | D | clean

PP-41 | A | clean

PP-42 | B | by elimination only — B's "for ten years" is a climate clue word on the weather key; a strong student reading L02's clue-word rule picks against it. MUST FIX: "This July has been very hot in Tucson."

PP-43 | B | every winter; relies on the lesson teaching that a plain plural with no time word means the usual

PP-44 | 4 lines | 7 − 3

PP-45 | A | G: 9, 11, 8; H: 10, 9, 11 — G ahead only in January; one true option

PP-46 | 32 more days | Erie 11+13+12 = 36; Atlanta 1+2+1 = 4

PP-47 | B | clean

PP-48 | C | drawing's 44 matches the stem; set aside, still in the table

PP-49 | A | all five Februaries 10-13 snowy days; clean

PP-50 | C | usually rainy (13, 12, 14, 2, 13); A tempting-wrong

PP-51 | A | still cool and foggy

PP-52 | 3 | 2023 (15), 2024 (13), 2025 (14)

Cross-item notes

• Tables versus drawings: every quoted table matches its graph description (L03/L04 Reno 10/12/14; PR-02/PR-03 Boise 9/7/6; PP-24 Richmond 8/6/9 with August drawn at 7; PP-48's 44). The v1 "January 12" mismatch is gone.
• "Day that does not fit" items (L07-02, L07-07, PR-10, PP-31, PP-32, PP-34, PP-50): each has exactly one reading far from the rest; the rest sit within a 6-degree or 3-day band.
• Bar and line items: the descriptions give the verdict for every reading item, as a drawing substitute must. For L03-06 and PP-25 the description states the missing part in words; the rendered figure must not carry that sentence.
• Weather-versus-climate time words: every L02 stem carries one. L01-07 and PP-43 have none and rely on a plain plural ("Winters in Fairbanks are…", "Anchorage has long, snowy winters") — fine if the lesson teaches that reading, and PP-43 says so in its stem. PP-42's weather key carries "for ten years", the one real conflict.
• Weather over more than one day: L02-07 (a week), PP-19 (a week), PP-20 (one winter). A student holding L01's title "weather is one day" literally could stumble on PP-20; the options rescue it.
• Key shape: L04-07 and PP-03 have the longest option as key; PP-11 has a capital/lowercase split. None hands the key over alone.
• Terms: "pattern" as a distractor in L01-02 before L01-09 coins it; "mild" (PP-19 distractor); "millimeters" (PP-32 label). "gridline" appears only in the drawing descriptions, not in student text. "record" first appears in L01-04 as "weather records", presumably coined in L01.
• Names: Lorena, Lincoln and Logan all start with L; Logan and Lorena share "Lo" and appear in neighbouring items (L05-05 Lorena, L05-09 Lincoln; L07-05 Logan, L07-06 Lincoln). Consider one non-L name.
• Repetition: "Which word names the weather a place usually has, year after year?" appears four times (L01-10, L01-V-01, L06-01, PR-06). Acceptable as fluency repetition; a PP100 item does not reuse it.

Would not ship as is: PP-42. Confirm before shipping: L05-08 (side label must read Sunny days), L03-06 and PP-25 (rendered figure must not print the missing-part sentence).