Unit 1 is built end to end and QA'd: 10 of 10 sequences authored · 66 lessons · 435 checks · 123 figures · 5 gates · 3 practice pages · 1 test. Cold solve 435/435 · blind read 433/433 across the ten sequences (parsed from the per-sequence records, never retyped; the gates, practice and test have their own tallies on their cards).
Do two things. First, rule on the 31 cards that genuinely need you — one tap each, and every tap sends straight to the inbox.
Second, skim the verdict cards (one per sequence, gate, practice page and the test) and tap Looks right or Needs work on each. The adjudication calls sit between, each with a one-tap veto.
This page IS the reviewer view, on by default: QA numbers, records and build calls are all visible. Students never see any of it. The course page shows it their way.
Each card is a genuine fork: your answer changes what a student meets, or what the next unit inherits. The filled button is my recommendation; tap it to confirm, tap the other to overrule, or use Other.
The deck's five gate clusters stay, but each cluster now splits into one-job sequences: ten in all, 66 lessons. QA and gates still ran per cluster, so a student's checkpoints are unchanged.
This is the first of the fourteen §MAP calls; the other thirteen sit on the §MAP adjudication card below with their own veto.
My recommendation: Keep ten — one conceptual job per sequence wrote cleaner and QA'd cleaner.
AST-003A (molar mass from a formula), AST-006A (atoms within molecules) and AST-041A (ion vs parent-atom radius) are new teaching atoms. Dan flagged all three as CED coverage the legacy unit skipped.
Each got its own lesson and checks and went through the full gauntlet with the rest.
My recommendation: Keep all three — each closes a named CED gap.
The old atom drifted into Unit 2 bonding. The reshape pins it to the CED's own 1.8 statement; the electronegativity-difference paragraph is licensed as a preview only and is never tested.
My recommendation: Stands — scope honesty with the preview kept.
The legacy extraction found it is an ordinary, sound spec atom, not the pilot artefact it was suspected to be. It kept its lesson.
My recommendation: Keep it.
S01 (the balance), S04 (the balloon) and S10 (sodium burning in chlorine) each fell to the figlib drawing after two honest QC fails: content rows passed twice; only the background-blur rule failed.
The takes and both verdicts sit in assets/U1-S01, -S04 and -S10. A swap is one line each — the photos are on disk.
My recommendation: Keep the drawings — the QC rule exists for a reason, and U5 shipped with the same residual class.
Two blueprints wrote verdict+because options at lesson tier; your 2026-09-28 rule bans the bundle there (it feeds the longest-option tell Dan kept catching). The build trimmed six bundles to bare verdicts (§E-C2-6 and §E-C4-3, one ruling family).
Bundles still live at gates, practice and the test, where the exam form licenses them. One lesson-tier exception stands because its blueprint licenses it by name (S07 L01's value+because).
My recommendation: The ban stands. A veto here restores the clause forms everywhere at once.
No unconsumed released question is Unit-1-pure, but unit-pure parts exist. The test's 2021 Q2 excerpt (reserved since 18 Sept) and the practice pages are built on part-level excerpts (CARD-ASSESS-1); the veto path, authored-to-archetype pages, is recorded but not built.
Declared with it: S06 teaches on silicon's own spectrum, so the excerpt's parts (a.ii) and (g) are partially rehearsed — inherent to any released question on taught content.
My recommendation: Approve part-level use — your standing rule prefers real questions, and the pattern has two accepted precedents.
The platform cannot take a drawing (CARD-ASSESS-2). The conversion follows your shipped U5-P2 precedent, quotes the released wording inside the prompt, and keeps both scoring features: position (rightmost, lowest binding energy) and height (2, same as 3s).
A veto drops the excerpt from 4 points to 3.
My recommendation: Accept — same class as the conversion you have already shipped.
The released test excerpt carries none of the four AP verbs (CARD-ASSESS-3). T1 attributes the checker's warning, and the Determine coverage lands at practice U1-P1, which carries a real released Determine.
My recommendation: Confirm — the coverage is real, just not on the test page.
2022 Q3 (a)+(b): the released paper is not archived, so the stems are rebuilt from the scoring guidelines and chief-reader report, tagged RECONSTRUCTED-PENDING-CHECK and never labelled verbatim (your U5 2022 Q5 precedent).
The 2021 Q2 test excerpt is NOT in this class: its paper was read and the stems checked verbatim; only its figures are reconstructions, and those were verified against the released drawings.
My recommendation: Ship with the tag; check against the released paper before anything leaves the course.
2023 Q1 (a.i)/(a.ii) demands a d-block electron configuration. The adjudicated register's fence 7 excludes d-block configurations (d is named, never drilled), so the parts were dropped from practice and the drop declared.
Your call whether the register's scope note should record that a released exam has asked it.
My recommendation: Add the one-line scope note; keep the exclusion.
G3 item 1's closer stands keyed-wrong on written grounds: the taught proportional method genuinely settles the comparison, so 'cannot' is a real misconception, not filler. After G5 item 9's third instance was killed, two instances remain across the gates; T1 item 17(d) is the same class.
The reviewers ask for a standing rule rather than more case-by-case calls.
My recommendation: Allow the closer where the stem's data genuinely settle the question; never as safe-looking filler.
The blueprint said 2.38, which is potassium's 3p — a transcription slip caught at spec. 2.08 is the one-source AP PES table's value, confirmed four ways, including the physics identity: outermost binding energy = first ionisation energy (CRC 2080.7 kJ/mol).
The headache ratio now reads 'about forty times' (84.0 / 2.08).
My recommendation: Confirmed — the C3 accuracy reader re-verified this value specifically.
Two standard datasets, rounded differently. The ruling (§E-C3-6): one provenance line at the AP table's first use ('rounded reference values … close to, not identical to, the removal energies taught later'), and no single item may print both datasets' values for one element.
My recommendation: One line is enough — unifying the datasets would falsify one source.
Dan's polish note asked for a pass mark and an honest header. The headers are honest; the engine simply has no pass-mark field, so all five gates ship without one — as U5's did.
Adding the field is one engine change, but it applies to every course on this engine and needs a policy number from you.
My recommendation: Ship without until you set the policy — name the number and the field follows.
The figures reviewer upheld the Group 16 highlight (§E-C4b): the lesson's worked pair is Group 17, so the figure gives the student a second case of the up-a-group pattern instead of a picture of the answer. A relabel to Group 17 invites picture-recognition over claim-reading.
The one-field alternative (figure beside its worked pair) is recorded and ready if you prefer it.
My recommendation: The highlight stands.
The drawn heights sum to 100 and the stem says 'heights idealized', so the reading is inferable — but never stated. The blind reader solved it anyway; the one-clause addition is written and ready.
My recommendation: Leave it — the 100-sum forces the reading.
Axis numbers currently print as 15.1, 1.46, 0.79, with prose and alt text aligned to the prints (§C3-S2). Fixed two-decimal labels are a one-line engine change plus a unit-wide re-render.
My recommendation: Keep the short form — it matches how the AP table itself prints.
S03's AST-019 C02 and C03 are different problems that happen to land on the same answer. The blind reader solved both and flagged only the variety. A re-instance is a fresh-data edit plus a fresh cold read.
My recommendation: Leave them — two honest problems, two honest solves.
The natural print is 6.022 × 10²², but the validator's Avogadro fence rightly refuses a bare 6.022 beside a power of ten (it guards the constant). The value is identical and the printed arithmetic true either way — this is cosmetics (§E-S01c addendum).
My recommendation: Keep the fence-safe form.
T1-Q12 and G5-Q09 both carry distractors that are right under the retired UK/old-US labelling ('Group 7', 'Group 4') and wrong under the AP 1–18 table the course teaches at AST-035. Correct by AP convention — but a UK-taught student could feel cheated.
My recommendation: Add the glossary line — one sentence in the AST-035 article.
The spec contract's form names the reader and commentates; the vault's write-explanation rule says state the fact, no commentary. Units 4 and 5 ship the contract form, and this unit kept it for consistency (§S10-S2, the standing U4/U5 fork).
Your ruling here settles it for all three units and updates the vault.
My recommendation: Keep the contract form until you rule — then one sweep aligns all three units.
The standing engine card from U5 (§S10-S4): alt text never answers an item, so scaffold figures carry generic descriptions, and count items may enumerate bars but never state a total or verdict (§E-C2b-5).
A real accessibility channel — full blind descriptions behind the answer reveal — is an engine feature, not a per-course edit.
My recommendation: Stands for now; the channel goes on the engine's list when you want it.
Its options c/d were near-verbatim G4 item 7's — a student off G4 minutes earlier re-met the same statement set with new numbers (Dan's recall-not-mastery class, one gate removed).
Before: 'Both atoms take in energy when they gain an electron.' After: 'Both values record energy taken in by the atoms.' Same two error paths, fresh strings, key untouched; the re-strung pair took a fresh cold read and matched.
My recommendation: The fix stands.
The item keys on fluorine releasing more energy. The feedback teaches the convention; the stem assumes negative-means-released. A triage note from the blind read, not a defect.
My recommendation: Add the one line to the stem.
Q17's key is 'between 376 and 419' — the least committal option, which test-wise students favour. Here that IS the construct (know how much precision the data support), so the blind reader would not change the options.
The flag is the pattern: if hedged-interval keys recur across the bank, the heuristic starts paying.
My recommendation: Watch it — a bank-level tally rides the next QA pass; no change now.
AST-051's taught skill is value-free chaining, so the item withholds the values and the feedback prints them. That puts items-do-the-atom above a stem-only reading of the values-always-printed rule (S09-3).
My recommendation: Stands — printing the values would hand over the chain.
This build ported the Grades 3–5 blueprint checker and every one of the ten blueprints exits 0 (§S01-10). Unit 5 shipped without one.
Either answer is a vault-sync note.
My recommendation: Make it standing — it caught real slips here at the cheapest stage.
Authors worked from its SKILL.md directly this build (§S01-S5). Linking makes it invocable as a skill; that is your call jointly with ops.
My recommendation: Link it — the pack is already the photo-real route of record.
S08 L05's redesign (§C4-S1): the key is the one error that produces the observed wrong order; every distractor is a verified non-producer whose feedback states what it would give instead. Dan's bug, inverted into a design rule.
My recommendation: Write it into the vault — it generalises to every diagnosis item.
Your 14 Oct ruling made Dan's feedback a standing rule (“You can take Dan's feedback as a rule - he is a chemistry expert, so happy to go with him”). Dan's 4 Oct note on AST-036: after the configuration lessons taught 1s 2s 2p, a shells-only sodium drawing “reads as a reversion to 2,8,8”. Triage: reviews/dan/U1_RAW_NOTES_TRIAGE_2026-10-09.md.
The chemistry floor held: the figure is correct (11 electrons drawn 2·8·1 — sodium's true shell occupancy, the shell grouping of 1s² 2s² 2p⁶ 3s¹), and S07–S08's mechanism story runs on shells, distance and shielding, so the shell view stays. The resolution honouring both is the middle path this card offered: ONE bridging line now sits at the figure — each ring is a shell; shell 2 holds 2s and 2p together (2 + 6 = 8); 2, 8, 1 is 1s² 2s² 2p⁶ 3s¹ counted shell by shell; both pictures show the same atom.
Record: reviews/dan/DAN_RULE_APPLICATIONS_2026-10-14.md. Gates re-run green after the insert.
Applied without asking (rulings you have already given): your 8 Oct rules (no quantity agency, no lesson-number references, named-instance restatements) hold unit-wide and live in the lexicon sweeps; the 2026-08-06 Coulomb ruling applied at its true scope (calculated force values banned, factor reasoning licensed — AP asks the factor form); the 2026-09-28 verdict+reason lesson-tier ban enforced (the card above is its veto); 'Period 4's main groups' kept over the bare clause because the bare clause is false across the d-block; effective nuclear charge is always written out in full, never a subscript symbol.
Every call below was made on the record under your option-(b) delegation and already executed. Tap Stands to confirm, or Veto with a line. The record itself sits collapsed under each card.
§MAP
The call: All fourteen of the atomiser's map recommendations adopted: five gate clusters kept, ten sequences, 66 lessons.
What changes: Ten blueprints, five QA packets; the four highest-signal calls sit on your Needs-you cards above.
## §MAP — sequence map (adjudicated 2026-10-08, operator; P3 = atomise @ 95bc97a) All fourteen P3 recommendations ADOPTED as written (02_decisions_for_operator.md §1–14): five gate clusters kept; TEN sequences; new atoms AST-003A (after 003, single source of standard M values), AST-006A (after 006), AST-041A (between 041 and 042); AST-009A kept; AST-053 reshaped to EK 1.8.A.1 with the ΔEN paragraph licence; S10 order 053→055→055A→054; AST-024 after 026A before 027; legacy IE-before-radius kept; G5 whole-unit mixed with ≥1 topic-1.8 item; CED authority = the repo's Fall-2024 extraction; mapping source = unit1_atoms_full.json (63 atoms); no merges or removals. 66 lessons total. James veto points: all fourteen surface as cards on U1_REVIEW.html, highest-signal first (ten-vs-five granularity; the three new atoms; 053 reshape; 009A keep). **MAP-2a (operator ruling): work-packet granularity.** Blueprints are written per SEQUENCE (ten — one conceptual job each). Spec authoring and the five-record QA gauntlet run per GATE CLUSTER (five packets: C1=S01 → G1; C2=S02+S03 → G2; C3=S04+S05+S06 → G3; C4=S07+S08+S09 → G4; C5=S10 → G5). Rationale: the gate is the natural closure point (a gauntlet that ends at its gate QAs the gate's feeding lessons together), and cluster sizes (10/13/19/20/4) match the U5 sequence sizes the playbook proved (S04 there was 27 lessons). The five-records-per-packet discipline is unchanged; records name the cluster and enumerate its sequences. **MAP-2b (operator ruling): validator delta.** The 12-fence exclusions list in 01_register_deltas.md is relayed to the P1 engine agent for the validator + mutation suite (incl. mass-spectra z = +1 scope, no Aufbau exceptions taught, the banned false rules).
§S01
The call: All thirteen moles-sequence flags adopted; the molar-mass canon became a validator fence.
What changes: S01 authored to the adopted plan; the canon is enforced mechanically.
## §S01 — blueprint flags (adjudicated 2026-10-08, operator; blueprint = atomise @ 95bc97a) - S01-1 ADOPTED: L01 hook photo-real via David's pack (QC'd, file-referenced); figlib balance drawing is the recorded fallback if QC fails twice. - S01-2 ADOPTED: both claim items stand — beliefs documented in the legacy bank and the lesson's own teaching. - S01-3 ADOPTED: coin "constituent particles" and "formula unit" in L02 (CED-named terms). - S01-4 ADOPTED: molar-mass canon fixed at L04 (S 32.06, Ca 40.08, (NH₄)₂SO₄ 132.14). RELAYED to P1 as a numeric fence: the validator carries the canon table; any student-facing molar mass that contradicts it is an error. - S01-5 ADOPTED: L07/L09 type (a) fluency. - S01-6 ADOPTED: no fresh L09 figure (numeric-lesson precedent). - S01-7 ADOPTED: L10's four bare chains = the R176 mixed set. - S01-8 ADOPTED: worked thread water/CO₂/glucose/NaCl; check pools disjoint per lesson. - S01-9 ADOPTED: L07 no-shared-mantissa pool rule; spec author fixes the final pool. - S01-10 RULED: blueprint checker — P1 engine agent ORDERED to port/adapt the blueprint checker (source: the Grades 3–5 build's build_tools_check_blueprint.py, adapted to this course's blueprint shape) and run it over every U1 blueprint as each lands; until it exists, hand-checks against the 6b spec stand ON RECORD (this blueprint's hand-check accepted). If the checker cannot be adapted faithfully, P1 reports back and the operator waives per-blueprint on record. James card: U5 shipped without one — vault-sync candidate either way. - S01-11 ADOPTED: fluency rows from R233 coining parts; register deltas unchanged. - S01-12 ADOPTED: article-only, video slots reserved. - S01-13 ADOPTED: 63 checks; exam-form load sits at G1/practice/test.
§S02
The call: The eleven verified legacy mass-spectrum figures stay; worked-example numbers never resurface in checks.
What changes: Binds every cluster's spec author, not just S02.
## §S02 — blueprint decisions (adjudicated 2026-10-08, operator; blueprint = atomise @ 95bc97a) - S02-1 ADOPTED: the eleven verified legacy mass-spectrum figureSpecs carry their items unchanged through the new figlib; write-mcqs/QA re-verify every key regardless. - S02-2 ADOPTED: element Q keeps Dan's named 191 (3.0) / 193 (7.0) peaks — his own specification; the neutral "element Q" + rounding tag covers the idealised split. - S02-3 ADOPTED (sequence-wide rule, applies to all clusters): sound legacy instances may stay in checks; worked-example numbers NEVER resurface in checks; scaffolded cases sit on a second instance, bare cases on a third. - S02-4 ADOPTED: rounding convention — idealised stems say "rounded for this problem"; real masses where they key exactly. - S02-5 ADOPTED: one photo-real mass-spectrometer hook (David's pack, QC'd, file-referenced); spectra stay figlib. - S02-6 ADOPTED: headache scenes as written (chlorine thread L01→L02→L04; taught numbers 35.45 / 3:1 / 34.97·36.97×0.7577·0.2423 are check-banned per S02-3).
§S03
The call: Composition adopted: scoped rules, two taught uncertainty triggers, one solve-for-x form.
What changes: Engine fences relayed: demanded-precision keys, caption dedupe, zero option-letter tokens in feedback.
## §S03 — blueprint decisions (adjudicated 2026-10-08, operator; blueprint = atomise @ 95bc97a)
- S03-1 ADOPTED: network-solid caveat dropped (fence 4 keeps covalent networks in Unit 2; the
scoped "usually" + named ammonium exception satisfies the no-exceptionless-rules rule).
- S03-2 ADOPTED: measurement uncertainty from exactly two taught triggers — a stem-stated ±
value, else agreement at the precision shown.
- S03-3 ADOPTED: one solve-for-x rearrangement; second component always 1 − x; Dan's corrected
NaCl-fraction row as the diagnostic twin.
- S03-4 ADOPTED: single-best-answer at L02; the roman-numeral multi-select form reserved for
practice/test (AP form kept where exam fidelity matters).
- S03-5 ADOPTED: one photo-real supplier-jar hook (QC'd, file-referenced).
- S03-6 ADOPTED: headache scenes as written. AST-012 drawing-correction item: options stay
verbal at spec stage; if the spec author draws the wrong drawing it must be captioned as the
student's, and review-figures owns the call (as flagged).
Engine notes from §S02/§S03 flags RELAYED to P1: fences 3 and 8 live before S02 specs; the
bromine two-atom-species licence (permitted as distractor, banned as taught content); caption
dedupe check; numeric keys stored and compared at demanded precision ("40.00"); half-unit-
last-digit tolerance convention; zero option-letter tokens asserted across fb/why fields.§S04
The call: Coulomb adopted: factor items licensed, calculated force values stay banned.
What changes: S04 teaches the factor form ('halving the distance quadruples the force').
## §S04 — blueprint decisions (adjudicated 2026-10-08, operator; blueprint = atomise @ 95bc97a)
- S04-1 ADOPTED: rubbed-balloon L01 scene with one photo-real hook (QC'd, file-referenced);
atom diagram stays figlib.
- S04-2 ADOPTED: factor items ("halving the distance quadruples the force") licensed — the
2026-08-06 ruling bans k and calculated force VALUES, not proportional reasoning; AP asks
the factor form.
- S04-3 ADOPTED: Mg²⁺/O²⁻ vs Na⁺/Cl⁻ promoted to the taught worked comparison; checks use
fresh ion pairs.
- S04-4 ADOPTED: L01 coins nucleus/proton/neutron/electron as working vocabulary — state,
label, drill the charges; no ceremony.
- S04-5 ADOPTED: headache scenes as written.§S05
The call: Configurations adopted with the 3d rung named but never filled.
What changes: As blueprinted; one transition-metal caveat line, no item on it.
## §S05 — blueprint decisions (adjudicated 2026-10-08, operator; blueprint = atomise @ 95bc97a) - S05-1 ADOPTED: energy ladder shows the 3d rung above 4s, named but never filled (fence: no Aufbau exceptions taught; d named, not drilled). - S05-2 ADOPTED: sodium as the worked spine (L01, L02, L04, L05, L07). - S05-3 ADOPTED: the misfiled-3d filling-order check stays. - S05-4 ADOPTED: one transition-metal caveat line in L06, no item on it. - S05-5 ADOPTED: one which-electrons-leave-first item tests the p-before-s order — CONSTRAINT: the item asks which electron leaves FIRST from a main-group configuration (e.g. Al's 3p before 3s); it must not present an unusual charge state as a common ion. - S05-6 ADOPTED: headache scenes as written.
§S06
The call: PES adopted on one data source; legacy gate figures became lesson items and G3 authors fresh spectra.
What changes: The spectrum renderer port became the C3 critical path, and landed.
## §S06 — blueprint decisions (adjudicated 2026-10-08, operator; blueprint = atomise @ 95bc97a) - S06-1 ADOPTED: all printed binding energies from the standard AP PES table, MJ/mol, one source (the blueprint's recomputations against Dan's figures accepted, e.g. 84.0/2.38 ≈ 35×). - S06-2 ADOPTED: legacy gate-item figureSpecs reused as LESSON items; G3 authors fresh spectra (gate stays cold; all 17 verified figures stay in play). The S06 do-not-copy list binds the cluster-3 gate author. - S06-3 ADOPTED: L04 (AST-031) prediction items use described-spectrum options, not per-item figures — kills the image-option give-away class U5 met. - S06-4 ADOPTED: AST-034 items carry binding-energy data in the stem, no per-item figure. - S06-5 ADOPTED: headache scenes as written. Engine orders from cluster-3 flags RELAYED to P1: the figlib `spectrum` renderer port (legacy deterministic svg-template → new RENDERERS registry; CRITICAL PATH for the C3 spec wave); fences 1/2/5/7/9 confirmed live before C3 specs; render_blueprint step noted as owed with the review-page work, not blocking.
§S10
The call: The bonding preview adopted, with the K₂O honesty requirement binding every string.
What changes: S10 is 4 lessons / 32 checks; the reservations bind the G5 author.
## §S10 — blueprint decisions (adjudicated 2026-10-08, operator; blueprint = atomise @ 95bc97a)
- S10-1 ADOPTED: L01 two-part split (EK 1.8.A.1 teaching + coin fluency).
- S10-2 ADOPTED: the "neither bonds readily" third label in the classification frame.
- S10-3 ADOPTED: photo-real L01 hook (sodium burning in chlorine), QC'd, file-referenced,
figlib fallback on two QC fails.
- S10-4 ADOPTED: Ca₃N₂ and ALL Sr instances reserved for G5 (binding do-not-copy list for the
cluster-5 gate author; no adjacent element reuse at the gate).
- S10-5 ADOPTED: ΔEN licensed-paragraph mechanics as the validator implements ({"licence":
"u2-preview"}, AST-053 only, never tested) — ALREADY LIVE per qa/ENGINE_PORT.md, so the C5
critical path is clear.
- S10-6 ADOPTED: legacy Rb₂S item kept with Dan's verbatim feedback fix.
- S10-7 ADOPTED: the MgBr verify-and-reject scope-verdict closer.
- S10-H (operator, binding): the K₂O honesty caution is a REQUIREMENT on every S10/G5 string —
K₂O is the charge-balance analogue, but no string may claim K₂O is what burning potassium
mainly forms (KO₂ is).
Engine follow-up RELAYED to P1 from the S10 flags: raw-HTML-in-student-strings fence and the
integer-key/tolerance-0 assert class (ion counts exact), with mutation cases. The four S10
figlib extensions (bondpanel, chargemap, ionbalance, columnswap) are the C5 SPEC AUTHOR's to
write as figlib_u1_* modules (auto-discovered; no engine edit).§S07/§S08/§S09
The call: All thirteen trend-sequence recommendations adopted; verdict and reason split at lesson tier.
What changes: G4's bounds-when-not-linear capstone became mandatory.
## §S07/§S08/§S09 — blueprint decisions (adjudicated 2026-10-08, operator; blueprint = atomise @ f618217) All thirteen recommendations ADOPTED: - S07-1 ADOPTED: lesson checks split verdict from reason (the 2026-09-28 James pattern; bundled verdict+because reserved for G4/practice/test where the architecture licenses it) — kills Dan's stock-sentence/longest-option class at the root. James card. - S07-2 ADOPTED: Ga/Al named once in the up-a-group scope line, untested. - S07-3 ADOPTED: Ba/Ca/Br bare chain kept. - S07-4 ADOPTED + EXECUTED: IE data canon (CRC) relayed to P1 as a numeric fence. - S08-1 ADOPTED: honest radius coin (half the bonded-nuclei distance). - S08-2 ADOPTED: legacy 18-electron set kept as the scaffolded case (S02-3 rule). - S08-3 ADOPTED: right-answer-wrong-reason form on I⁻/Br⁻ kept. - S08-4 ADOPTED: "anions are always larger than cations" kept as a flagged distractor whose feedback names the counterexample class (never taught). - S09-1 ADOPTED + EXECUTED: EA data canon relayed to P1 as a numeric fence. - S09-2 ADOPTED + EXECUTED: EN unitless S135 licence registered with the engine. - S09-3 ADOPTED: values printed even for named established series — no "established" ranking without data on the page. - S09-4 ADOPTED: model-error honesty lines in L09 kept. - S09-5 ADOPTED: the double Si/P appearance in L02 kept (different asks). Plus both instance-overlap repairs ADOPTED as named (S07 L06 item-5 swap; S08 Si/S swap) — binding on the C4 spec author. The shared group-mechanism wording rule (Z_eff roughly similar down a group; distance decides) BINDS S07 L04 / S08 L02 / S09 L07 — the AST-038/041 contradiction class stays dead. G4 do-not-copy instance lists bind the cluster-4 gate author; the G4 bounds-when-not-linear capstone item is MANDATORY.
§S01-S
The call: S01 spec flags adopted; the design JSON is generated from the adjudicated map, never hand-written.
What changes: unit1_design_v1.json is derived mechanically; the generator refuses on divergence.
## §S01-S — spec-stage rulings (adjudicated 2026-10-08, operator; spec = write-lesson @ f618217) - S01-S1 EXECUTED (engine order): FORWARD_TERMS corrects "formula unit"'s definer AST-012 → AST-002, per the standing §S01-3 adjudication (L02 coins it; AST-006A's own register statement uses it); \bionic\b stays at AST-012. The 29 one-class S01 errors clear with it. - S01-S2 AMENDED RULING (supersedes the "no design JSON" note of P1 close): build_course.py needs a design file for the real build, so unit1_design_v1.json is GENERATED from the adjudicated blueprints/00_sequence_map.md by the engine agent — mechanically derived, never hand-drifted; the map stays the single source of truth; the owed re-validation runs on landing and any divergence is adjudicated. - S01-S3 ADOPTED: the L01 hook ships the figlib balancescene fallback — the two-QC-fail rule fired honestly (content rows passed twice; overall failed twice on background sharpness). The take-2 photo + both verdicts stay in assets/U1-S01/; JAMES CARD: same residual class U5 shipped with on record — his call whether the photo replaces the drawing. - S01-S4 ADOPTED: the S135 prose-good trade (10 warnings) and the 27 R113 two/three-option licences stand as licensed classes. - S01-S5 NOTED (ops): David's image pack is installed but not linked into the Skill list — authors work from its SKILL.md as the C1 author did; linking is James's/ops call, carded. - S01-S6 ADOPTED: "ionic" deliberately absent from the L02 coin cell (lives at AST-012). - S01-S7 NOTED as vault-sync candidates: check_parallel slot-counting on chemistry pairs; the keys-spread note.
§S10-S
The call: S10 spec flags adopted; the feedback-voice fork stays your card, not a unilateral vault change.
What changes: The contract feedback form stands unit-wide pending your ruling.
## §S10-S — spec-stage rulings (adjudicated 2026-10-08, operator; spec = write-lesson @ 16bb23f) - S10-S1 ADOPTED: the covalent coin's "meet each other" minimal deviation (avoids the S50 bare-"meet" lexicon error); recorded, not a defect. - S10-S2 RULED: the fb-voice fork (SPEC_CONTRACT "This option is wrong — you…" vs the vault write-explanation no-commentary rule) is the SAME standing open line U4 and U5 carry — contract form stands unit-wide for consistency; the fork goes on James's card; no unilateral vault change. - S10-S3 ADOPTED: L01's five-case summary datatable (three-or-more-things-is-a-table); review-figures reads it in the gauntlet. - S10-S4 ADOPTED: the deliberately generic scaffold-map alt (alt never answers an item); the blind-description accessibility channel stays the standing engine-level James card from U5. - S10-S5 NOTED: photo-real hook again fell to the figlib fallback on the same background-sharpness-only QC class (content rows perfect twice) — one JAMES CARD now covers both instances (S01, S10): swap-to-photo is one line each if he prefers the photos.
§S02/S03-S
The call: C2 spec flags adopted; the canon outranks a blueprint arithmetic slip.
What changes: The canon-fence phrasing convention goes to the vault.
## §S02/S03-S — spec-stage rulings (adjudicated 2026-10-08, operator; spec = write-lesson @ 16bb23f)
- S02-S1 ADOPTED: L04's dead-rule claim item rephrases Dan's banned wording rather than
licensing it — no licence mechanism added for a FALSE rule; the fence stays absolute.
- S02-S2 ADOPTED: legacy AST-010 lesson visual dropped for the fresh eqstack (75.0/25.0
labels contradicted the taught 75.77/24.23) — the blueprint's own else-branch.
- S02-S3 ADOPTED: L03 keys the other copper isotope (31.0%) — worked numbers never resurface.
- S02-S4 ADOPTED: CuSO₄ expected % corrected to 39.82 per the §S01-4 canon (canon outranks a
blueprint arithmetic slip; class noted for the C2 QA reader).
- S02-S5 ADOPTED: L07 consistent-verdict check on fresh CaCO₃; KCl 47.5 pair stays taught-only.
- S02-S6 ADOPTED + vault-sync candidate: the canon-fence phrasing convention ("masses are
63.0 and 65.0, in amu" — never a non-canon value unit-adjacent to a canon element name).
- S02-S7 EXECUTED (engine order): lesson TITLES are student-facing and must flow through the
same mathtext/typesetting path as body text — the AST-004 title "Moles from mass: n = m/M"
currently breaks the whole-course build on the raw-maths survival assert. Engine routes
titles through mathtext (with a mutation case); no title rewording needed.
- S02-S8 NOTED: the supplier-jar label's ambiguous lowercase-l print (sans-serif bar) judged
no-reject by the QC seat; recorded for review-figures to confirm on the rendered page.§C3-S
The call: C3 spec flags adopted; neon's 2p corrected to 2.08 MJ/mol at spec.
What changes: The one-source table now prints consistently everywhere.
## §C3-S — spec-stage rulings (adjudicated 2026-10-08, operator; spec = write-lesson @ f618217) - C3-S1 ADOPTED: Ne 2p = 2.08 MJ/mol (the one-source AP table AND the reused legacy neon figure agree; the blueprint's 2.38 was K's 3p — a blueprint slip, corrected at spec); headache ratio "about forty times" (84.0/2.08 = 40.4). JAMES CARD (data provenance). The C3 accuracy reader re-verifies this value specifically. - C3-S2 ADOPTED: spectrum axis g-format prints with prose/alts aligned to the prints (S06-1); two-decimal axis labels are a one-line engine change if James prefers — on his card. - C3-S3 ADOPTED as the course convention: Z_eff never prints as a symbol; student prose says "effective nuclear charge" in full (clearer than a subscript glyph; AP wording). Noted on the record; blueprint's subscript note superseded. - C3-S4 RULED: the S04 balloon hook stays prose-only FOR NOW (the author's session lacked the image-pack skills); the generation brief sits in qa/SPEC_U1-S04.md — the C3 FIX ROUND (after QA wave 1) attempts the photo per that brief under the two-QC-fail rule; prose-only is the standing fallback. Same James photo-card class as S01/S10. - C3-S5 ADOPTED: the 2.2 px anneq bbox-kiss (inkless NBSP advance, screenshot-verified clean) is licensed; engine suggestion logged for close-out.
§C4-S
The call: C4 spec flags adopted; Dan's diagnosis bug inverted into a design rule.
What changes: The shared-pair renderer gained a guard so the drawn pull is measurably true.
## §C4-S — spec-stage rulings (adjudicated 2026-10-08, operator; spec = write-lesson @ 16bb23f) - C4-S1 ADOPTED: the S08 L05 diagnosis redesign (key = the one error that produces Cl⁻ > S²⁻; distractors verified non-producers whose feedback states what they would give) — Dan's bug inverted into a design rule. James card (strong pattern, vault-sync candidate). - C4-S2 ADOPTED: the L09 licence drills as a true minimal pair (one frame, one slot, opposite keys) over the blueprint's asymmetric options. - C4-S3 ADOPTED: ptmap draws main groups only, marked footnoted d-block gap, no Period 1; plotted elements land in true columns by construction. - C4-S4 ADOPTED (blueprint errata): 52 checks follow the per-lesson plans (the 50 was a header slip); Mg→Cl is FIVE extra protons (shipped right); the sharedpair geometry deepened with a renderer guard so the drawn pair is measurably nearer chlorine (the figure asserts the true pull). - C4-S5 NOTED for ENGINE_COVERAGE: the canon proximity matcher can mis-pair in "A's is X and B's is Y" prose — C4 stems use the tight list form; a coverage line is owed at close-out.
§ASSESS
The call: The assessment plans adopted: an 18-slot test spread, five binding gate plans, real part-level excerpts for practice.
What changes: One card (the part-level ruling above) governs all released-question use.
## §ASSESS — assessment plans (adjudicated 2026-10-08, operator; prep = write-mcqs @ cca3af6) - A-1 ADOPTED: the U1-T1 18-slot spread (S01×3 · S02×1 · S03×2 · S04×1 · S05×2 · S06×2 · S07×2 · S08×1 · S09×3 · S10×1), barred-instance ledgers, and fresh-candidate pools (with the recount-at-use-time rule — pools re-grep at build time against the then-current specs). - A-2 ADOPTED: the 2021 Q2 excerpt conversion as staged (three parts verbatim, (g) draw→state per the U5-P2 precedent with released wording quoted; figures reconstructed-verified; 10/10 recomputes; number-free (h) models honouring E-C3-6). The Si-overlap note and the draw→state conversion go on James's cards. - A-3 ADOPTED: all five gate plans as staged, including G3's honest freshness definition ((element, ask, pair) + always-fresh-figure — the Z = 1–20 PES universe is finite) and G4/G5's binding requirement lines. - A-4 RULED (practice): BRANCH A (real part-level excerpts) is BUILT — James's standing rule prefers real questions and the part-level pattern has two accepted precedents; ONE card covers the part-level ruling for the 2021 Q2 split AND the practice excerpts (veto path = branch B authored-to-archetype, recorded not built). P3's 2026 Q1 builds ONLY if its scoring points verify against the released SG at build time; otherwise P3 falls back to an authored FRQ with the disclosure note and says so. - A-5 NOTED: examiner pass ordered on U1-T1 before its blind read (U5 pattern); pool partition re-grep between gates and test at each build.
§ASSESS addendum
The call: Three practice departures ratified: the d-block part dropped, P2 moved after G4, P3's FRQ re-instanced.
What changes: The staged USAGE.md rows apply at close-out under your part-level card.
## §ASSESS addendum — practice departures (ratified 2026-10-08, operator) - RATIFIED: P1 drops 2023 Q1 (a.i)/(a.ii) — the released part demands a d-block configuration the register's fence 7 deliberately excludes, and the page sits before configurations are taught. JAMES CARD: a released exam part the course's scope excludes (his call whether that changes the register's scope note). - RATIFIED: P2 placed after G4 (2022 Q3(b) is AST-041A content — items-need-only-taught- content outranks the staged placement). - RATIFIED: the authored P3 FRQ re-instanced S²⁻/S after the mid-build G5 P³⁻ collision; the give-away adjacency (P3's model prints an 18-electron string = G5's P³⁻ key) is a BINDING check for the G5 QA pass. - The staged USAGE.md ledger rows apply at close-out (operator), all pending the ONE part-level James card (CARD-ASSESS-1); 2026 Q1 NOT consumed; 2022 Q3 excerpt carries RECONSTRUCTED-PENDING-CHECK like U5's 2022 Q5 precedent.
§E-S01
The call: C1 wave 1: 63/63 cold; one Essential — a scaffold chain carried unrounded intermediates and marked an honest route wrong.
What changes: The fix round applied exactly these lines; 13 changed items joined the blind read.
## §E-S01 — C1 QA wave-1 fix round (adjudicated 2026-10-08, operator)
Cold-solve 63/63; 1 Essential, 13 Worth fixing across the two records. Rulings:
- E-S01-1 (A1, Essential): the L08 scaffold chain carries UNROUNDED intermediates — item 3's
revealSteps/fb show 1.506 × 10²³ carried forward; item 4's tolerance widens to 1e21 so the
honest 3.02e23 route passes; printed products made arithmetically TRUE; one
carry-the-unrounded-count teaching line added. Key unchanged.
- E-S01-2 (A2/A3, same class): all printed worked products recomputed with 4 s.f.
intermediates so no printed line asserts false arithmetic.
- E-S01-3 (A4 + S-row L09-c1): "sensor records N molecules" stems re-instanced to honest
framings (counts derived, never instrument-read) — the lesson's own premise and EK 1.1.A.1.
- E-S01-4 (A5): the hook balance gains its taring line.
- E-S01-5 (A6): quantity-agency reword ("its grams buy more moles" class) per the 8-Oct rule.
- E-S01-6 (S2, engine FACT): there is NO runtime option shuffle — key positions are
spec-managed. The author-record claim is corrected; the flagged middle-slot pile-up and the
6/27 key-tell items get the records' exact trims/rebalances. House behaviour (static order)
matches U5; bundler shuffle stays on the close-out improvement list.
- E-S01-7 (L10-c6): fresh instance adopted (4.216e23 CH₄ → 11.2 g class), ending the
L09-given reuse.
- E-S01-8: the "This option is wrong —" opener stays (house contract; standing James card).
Fix agent applies ONLY these lines + the records' exact fixes; every changed value re-verified
by script; validator + sweeps re-run; changed items listed for the blind read.§E-S10
The call: C5 wave 1: 32/32 cold; two Essentials, both position-fact honesty fixes.
What changes: Changed items took their fresh cold read at the blind stage.
## §E-S10 — C5 QA wave-1 fix round (adjudicated 2026-10-08, operator) Cold-solve 32/32; accuracy 2E/7W/6L; questions 0E/3W. Rulings — ALL the records' exact fixes ADOPTED, specifically: - E-S10-1 (A1): delete "directly" — "Rubidium sits below sodium, in sodium's column." (Potassium is directly below; the lesson says so itself.) Key Rb₂S stands. Ruling note: a structural adjudication never immunises a false position fact. - E-S10-2 (A2+A7): the metals-left/nonmetals-right line is TAUGHT IN PLACE (back-reference dropped — nothing in the rebuild teaches it) with the hydrogen top-left exception clause. - E-S10-3: the seven accuracy W fixes as written (similar-strength sharing fb; LiO₂ existence overclaim; the closed "two other real compounds" count; metal-with-NONMETAL sorting scope; electron wording; CaO₂ parity line; plus the record's seventh row). - E-S10-4: the three question W fixes as written — argon key trimmed; L02-01 re-instanced to cesium (taught-instance rule); FOUR option reorders (key spread 12/30 and the 43% index-2 cluster both repaired). The engine-shuffle caveat is SETTLED: no runtime shuffle exists (E-S01-6) — reorders are required, not optional. - E-S10-5 (note for the G5 author, non-blocking): Sr appears in S07–S09 periodicity items — the 1.8 reservation (SrCl₂→SrO class) stands; the gate author avoids adjacency collisions with those items too. Changed items/paragraphs owe their fresh cold read via the blind stage.
§E-S02/S03
The call: C2 wave 1: 99/99 cold, zero Essentials; the vault's lesson-tier verdict+reason ban outranked the blueprint.
What changes: Three changed-key items read fresh at the blind stage.
## §E-S02/S03 — C2 QA wave-1 fix round (adjudicated 2026-10-08, operator) Cold-solve 99/99; zero Essentials; accuracy 5W/4L across the two sequences, questions 12W. Rulings — ALL the records' acted fixes ADOPTED, specifically: - E-C2-1: the double-rounding worked line prints unrounded products (26.4968 + 8.9578), rounds once (we + eqstack + alt together). - E-C2-2: the Zr claim feedback restated as a GUARANTEE, not an exclusivity (the course's own Mg case breaks "only for two-isotope elements"). - E-C2-3: the bromine stem gains its "(rounded for this problem)" tag (S02-4 convention). - E-C2-4: "Every particle of a pure compound is identical" → same-elements-same-counts wording (isotope-proof); the L08 "mixture … alone" one-word fix. - E-C2-5: the copper 69.0% and sucrose 50.0% taught-number resurfacings re-instanced as proposed (antimony 57.0/43.0; 3.640 g Cl → 60.0%) — key-value changes via fresh stem data; the three changed-key items get their fresh cold read at the blind stage. - E-C2-6 (RULE-VS-BLUEPRINT CLASH, ruled): the vault's 2026-09-28 verdict+reason-at-lesson- tier ban OUTRANKS the adjudicated blueprint — all three bundles rewritten as proposed (the L08 bounds pair recast in the course's Yes/No claim form, which also carries the B2 fix). The clash itself goes on James's cards (same family as U5's licensed-AP-form line: bundles live at gates/tests only). - E-C2-7: the remaining question W rows as written (sort-run string 011011012; FeSₓ wrong-map error renamed; "formula unit" distractor swap; the law-name item takes the record's first proposed route; stem trims). - E-C2-8 (noted, non-blocking): hand key-tells read stands in for the .py-bank tool (engine JSON — tool gap is a vault-sync candidate); the 47%/44% middle-lean and L07's 4-of-5 "impure" run are WATCH notes for the C2 gate author; count-item alt enumeration goes to review-figures; claim-fb shared reasoning = the standing contract-form card.
§E-S01 addendum
The call: Fix-round deviations ratified: the exact print wins over a prescription that would itself be false.
What changes: Carried into the one C1 application round.
## §E-S01 addendum — fix-round deviations (adjudicated 2026-10-08, operator) - E-S01-2a ADOPTED: the fix agent's 1.2044 × 10²³ exact-print deviation stands — the QA reader's prescribed 1.204 would itself print false arithmetic; the "no false printed line" clause governs. (Deviation documented in qa/FIX_U1-S01.md.) - E-S01-9 ORDERED (carried to the C1 closing application round): L08-c5 and L10-c8 gain the anticipated-wrong map "7.22e23" with rounded-too-early feedback — the 4 s.f.-carry student is told the error the lesson now teaches against, not silently rejected. No key change. - E-S01-10 ADOPTED: option terminal-punctuation follows the file's universal convention.
§E-C3
The call: C3 wave 1: 124/124 cold, zero Essentials; one key flip so a real-ion comparison lands true.
What changes: The phosphorus provenance line ordered — your one-dataset card.
## §E-C3 — C3 QA wave-1 fix round (adjudicated 2026-10-08, operator)
Cold-solve 124/124; zero Essentials; 11 W + 2 operator calls. Rulings:
- E-C3-1 ADOPTED (the one key change): S04 L05 item 7's separations swapped so the real-ion
comparison lands TRUE (K⁺/Br⁻ vs Li⁺/F⁻ must not teach the reverse of reality); key flips
with it; item owes a fresh cold read (blind stage covers it, and the fix agent lists it).
- E-C3-2 ADOPTED: the S05 ground-state fb one-clause fix (atoms AND ions); "shielded mainly by
just" loses "just"; L03-4's impossible-error map fixed; L03-5's verdict+reason bundle
rewritten bare (the 2026-09-28 rule, same as E-C2-6); the stem clause printing the previous
answer cut; key-length tells lengthened as given.
- E-C3-3 ADOPTED: all given key-string rebalances verbatim (S04 L03 run; S05 L07/L02; S06 L01
a a a a → a a b b). The unit-wide first-option lean (49/39/18) is NOTED for the gate/test
authors; no mass-reshuffle beyond the given strings.
- E-C3-4 ADOPTED: S06 L01 Z_eff numeric items gain exact:true + tol:0 and the "+3" accepted
entry form (grader contract).
- E-C3-5 RULED (operator call 1): the L06 element-ID collisions are REAL defects in an
identification atom — L03 stems anonymised as proposed, AND the L06 Na/Mg unknowns
re-instanced to the supplied fresh candidates (F, Si): identifying a spectrum the student
saw LABELLED one lesson earlier is recall, not identification — Dan's own cold logic applied
in-lesson. Full recompute required; §S06-2's reuse licence stands for TEACHING figures only.
- E-C3-6 RULED (operator call 2): the P 1.06 (AP PES table) vs 1.01 (CRC IE) cross-dataset
mismatch gets ONE provenance line at the AP table's first use in S06 ("rounded reference
values from one standard PES table; close to — not identical to — the ionization energies
taught later"), and NO single item may print both datasets' values for the same element.
JAMES CARD (one-course-one-dataset scope).§E-S01b
The call: S01 blind read: 61/61 of solved matched; the two unsolved were real missing-data stems.
What changes: All rows joined one C1 application round.
## §E-S01b — S01 blind-read findings (adjudicated 2026-10-08, operator)
Blind read scored by the operator: 61/63 solved, 61/61 MATCH, zero clashes, zero
second-defensibles. The two unsolved are REAL missing-data defects:
- E-S01b-1 (Essential-grade): AST-008 C03/C04 — the canister chain's givens live only in the
preceding article step; the stems gain their data ("A canister holds 33.01 g of CO₂
(M = 44.01 g/mol)…" per the spec's own numbers), exactly as C05/C06 already do. Items carry
their data — the course's own write-mcqs contract.
- E-S01b-2: the five soft context-dependence rows fixed the same way — AST-006A-C03 names its
0.250 mol sample; AST-004-C01 names its scenario; AST-003A-C01/C02 print the atomic masses
they need (the atom is molar-mass-FROM-formula; table values are givens, AP practice);
AST-005-C02 prints M(NaOH) = 40.00 g/mol.
- E-S01b-3: grader accept-forms verified for "120." (accept 120, 120., 1.20e2) — application
agent confirms the engine's accept mechanism; 12.10's half-unit tolerance already admits the
honest 12.096 route (|12.096 − 12.10| = 0.004 < 0.005), no change.
- E-S01b-4: "Avogadro's number" stays — the CED's own name, coined deliberately.
These rows join E-S01-9 and the pending review-lesson F-rows in ONE C1 application round.§E-C4
The call: C4 wave 1: 117/117 cold, zero Essentials; the all-No meta-tell killed in the lessons, not just at the gate.
What changes: S07 L01's licensed value+because bundle stands, named on your card.
## §E-C4 — C4 QA wave-1 fix round (adjudicated 2026-10-08, operator)
Cold-solve 117/117; zero Essentials; 5W + 8W + 12W. Rulings:
- E-C4-1 ADOPTED: all length-tell extensions and the named option reorders exactly as the
records mandate (incl. the L04 C6 3× monster and S07's never-keyed position 0).
- E-C4-2 ADOPTED: the L05 C1 fb-fidelity fix (distractor fb names the error that produces the
wrong Rb⁺-largest order).
- E-C4-3 RULED (the §C4-S2 clash): the 2026-09-28 lesson-tier verdict+reason ban OUTRANKS the
blueprint's clause forms — the three bundles (L04 C5, L09 C5/C6) trim to bare verdicts,
reasons stay in fb; §C4-S2's minimal-pair FRAME survives untouched. Same ruling family as
E-C2-6; ONE James card covers both (bundles live at gates/tests only — his veto restores the
clause forms everywhere at once). L01 C5's value+because stands (blueprint licenses it by
name — on the card as the licensed exception).
- E-C4-4 ADOPTED: L09 C4's two-clause key cut.
- E-C4-5 RULED (the all-No meta-tell): the S08 warm flip candidate is EXECUTED (one claim item
flips to a TRUE Yes-keyed claim, per the record's named candidate, full recompute), AND the
≥1-true-Yes-claim requirement BINDS G4's author. Severity: the meta-tell dies in the lessons,
not just at the gate.
- E-C4-6 NOTED: the S08 radiusstrip alt clause ("strict shrink" vs the drawn P=S) fixed with
the W rows; the "most tables"→"many" one-worder; the blueprint's Mg→P three-protons erratum
recorded so no fix round restores "five"; the S09 L07 G16-highlight-vs-G17-pair Leave goes
to the C4 figures pass for a second eye.
All mandated wordings pass the lexicon sweep before shipping (fix-contract rule).§E-C3/§E-C4 fix-round deviations
The call: Fix-round departures ratified; one engine order — the equation renderer under-measured its numerator.
What changes: The figure board is fully clean.
## §E-C3/§E-C4 fix-round deviations (ratified 2026-10-08, operator)
- C4 departures 1–4 RATIFIED (Period-5 fb slip corrected; Yes-flip on the fresh Lena O²⁻/O
instance, Cl⁻/Cl stays taught; bare Yes/No form; reviewer tally slip recomputed).
- C3-D1 RATIFIED: the E-C3-6 provenance line reworded forward-term-safe ("the removal energies
taught later in this unit") — the ruling's dictated wording itself tripped two fences; the
substance governs (E-S01-2a class).
- C3-D2 RATIFIED: the never-exercised 14.7 script typo corrected to 15.1.
- C3-carried-1 → ENGINE ORDER: layout_eq underestimates the |q₁q₂| numerator width by 4.4 px
against the 8 px pad (AST-021's 2.2 px overlap, invariant across spacings) — fix at template
level (wider "|" width estimate or numerator pad), re-render, figure sweep to 123/0.
- C3-carried-2 CLOSED: balloon photo fell to the two-QC-fail rule on SHARP_NO_DOF only
(content rows passing twice) — the THIRD instance of the S01/S10 residual class; ONE James
photo-card covers all three (take + verdicts kept in assets/U1-S04/).§E-C2b
The call: The C2 closing pass found the unit's one renderer Essential: the label shelf assigned in draw order and mislabelled three spectra.
What changes: The qa gate now accepts per-sequence records — the richer audit trail.
## §E-C2b — C2 closing-pass rulings (adjudicated 2026-10-08, operator) - E-C2b-1 (F1, ESSENTIAL → ENGINE): figlib_u1_spectrum.py's y-label collision shelf assigns in DRAW order — fix at template level to VALUE order, add the renderer's value-order assertion as a refusal, re-render, and re-verify EVERY spectrum unit-wide (S02 masses + S06 PES share the renderer). Three on-page instances were actively mislabelled (Mg 20-above-22 supporting the swap distractor; both Zr 34/35 pairs). - E-C2b-2 (F2, after F1): in-plot haloed height labels on the two near-equal Mg bars (engine: the per-bar label path exists — enable per-figure param; the C2 application agent sets it). - E-C2b-3 (F3–F5 ADOPTED): Na⁺ side-label with leader; the 900 g beaker drawn visibly larger (the pair's changed feature must be on the drawing); tone-word alts reworded to role/size words (FALSE in dark mode — class note for all clusters). - E-C2b-4 (F6 ADOPTED): "The average already contains the abundances" (agency fix); polish rows as listed. - E-C2b-5 ADOPTED + vault-sync candidate: the count-item alt ruling — alts may enumerate bars (data, equivalent access); never a total, the asked quantity's words, or a verdict; ordinals lawful; bare list preferred. - E-C2b-6 (OPS reconciliation): the qa gate expects cluster-named records; the build uses per-sequence records matching the accuracy/questions set — ENGINE adjusts the gate to accept per-sequence records (a cluster's gate opens when every sequence in the cluster has its five record types). Per-sequence naming is the richer audit trail and stands. - E-C2b-7: C2 closure = F1 re-render re-look + F2–F6 application + fb re-read of touched strings. S06's spectrum re-verification folds into the C3 closing pass (relayed).
§E-S01c
The call: C1 closing pass: all eight rows adopted; one genuine catch — an item sat on the worked example's own result pair.
What changes: C1 closed on that single-item read.
## §E-S01c — C1 closing-pass work order (adjudicated 2026-10-08, operator) All 8 F-rows ADOPTED for the ONE C1 application round: F1 (= ordered E-S01-9 maps), F2 (= E-S01b-1 canister givens, record wordings), F3 (= E-S01b-2 five stems), F4 (= E-S01b-3 accept-form verification through the real grader — no engine edit unless it fails), F5 (the genuine catch: L10-c7 re-instanced to 46.25 g (NH₄)₂SO₄ → key "2.11e23", maps verified — KEY CHANGE, owes a fresh cold read), F6 (exact 6.022×10²² print), F7/F8 (polish as written). CHANGED_IDS per the record's §4; F5's cold read is the single owed blind-stage item.
§E-C4b
The call: C4 closing pass: the Group 16 figure ruling upheld; no keys changed.
What changes: No new blind-stage items arose from this pass.
## §E-C4b — C4 closing-pass work order (adjudicated 2026-10-08, operator) - G16 ruling (E-C4-6) UPHELD as the reviewer argued (figure-before-prose held; a G17 relabel invites picture-recognition over claim-reading); the one-field alternative recorded on James's card. - All F-rows ADOPTED for the C4 application round: S09-F1 (sharedpair frame tightened to content+margin in figlib_u1_atomsize.py — the C4 lane's own module — re-render, re-shoot); S07-F3 (two stacked 4-column windows replace the clipped 8-column table); S07-F4 meta-line rewrites; S07-F1/F2 + S08-F1 polish as written. Leaves stand as reasoned. - No keys/options change → no new blind-stage items from this pass.
§E-S01c addendum
The call: Application deviations ratified: the fence-safe 0.6022 × 10²³ print and the wrong-map ordering.
What changes: C1 closed on the single L10-c7 cold read.
## §E-S01c addendum — application deviations (ratified 2026-10-08, operator) - RATIFIED: F6 prints the exact intermediate as 0.6022 × 10²³ (the Avogadro fence rightly bans a bare "6.022 × 10²²"; value identical, printed arithmetic true). Cosmetic note on James's cards (mantissa < 1 form). - RATIFIED: the 7.22e23 map listed BEFORE 7.20e23 (overlapping near-windows; both entries now read the correct rounded-too-early correction). ENGINE CLOSE-OUT CANDIDATE recorded: exact-match pass before near-window pass in the wrong-map matcher. - C1 closes on the single L10-c7 cold read (launched).
§E-C2b addendum
The call: The C2 application agent's own-module edit ratified; the C2 gauntlet closed.
What changes: Five records green per sequence; the qa gate is satisfied.
## §E-C2b addendum — application departure (ratified 2026-10-08, operator) - RATIFIED (D1): the C2 application agent edited figlib_u1_composition.py (its cluster's own authoring module) for F3/F4 — the authoring-lane convention; the refusal-guarded param pattern followed. C2 GAUNTLET CLOSED (five records green per sequence; qa gate satisfied). - NOTED for vault-sync: "remembers/knows/forgets" → the agency sweep's lexicon list (the B7 blind spot); E-C2b-5's count-item alt ruling still owes its vault-sync note (close-out).
§E-G5
The call: G5 QA and blind: 10/10 cold, both binding adjacency checks give-away-safe; two Worth-fixing rewrites executed.
What changes: Items 6 and 9 took their fresh cold read at the T1 blind stage — both matched.
## §E-G5 — G5 QA + blind rulings (adjudicated 2026-10-08, operator) - Both BINDING adjacency checks ruled GIVE-AWAY-SAFE by the QA pass (stepwise rendering; the 18-electron string is canonical course furniture) — accepted; the honest page-order fact (practice renders before the gate) recorded. - W1 ADOPTED (operator, bias to the student — executed, not parked): item 6's options c/d re-strung per the reviewer's exact fix (the cross-gate verbatim repeat was Dan's recall-not-mastery class one gate removed); key and error paths untouched; before/after on James's card. - W2 ADOPTED: item 9's option d → "The group number equals n." — kills the third never-keyed cannot-closer while keeping the period-confusion misconception; the closer-CLASS standing ruling stays on the G3 card (two instances remain). - Blind-flag micro-fixes executed earlier this stage: G5 Q08 tol 0.1; G2-Q08 accept forms; G2-1 rounding tag; G1 3.05e5 ×2; the P1 atomic-mass data line. Q02's heights-as-abundances clause ruled LEAVE (the 100-sum forces it; one-clause fix on James's card); Q06 sign- convention triage note carded. - OWED: a 2-item fresh cold read of G5 items 6 and 9 (distractor text changed) — folded into the T1 blind stage; the regenerated G5 bundle is on disk, leak PASS.
§E-T1
The call: Four test-build deviations ratified; the 2021 excerpt ships as two linked FRQ items.
What changes: U1-T1 closed on that read.
## §E-T1 — test-build deviations (ratified 2026-10-08, operator) All four RATIFIED: (1) the 2021 Q2 excerpt ships as two linked frq items (one figure per frq-item renderer limit — the shipped U5-P2 precedent; scoring 1:1, released labels kept); (2) the "shown above"/bridge-stem adaptations (the U1-P1 declared class); (3) the (g) draw→state disclosure lives inside the prompt (test pages carry no note steps); (4) the shot artifacts beyond the named files (house pattern). The A-1 use-time re-picks stand as documented. Examiner pass ordered BEFORE the blind read (A-5); the blind stage also carries the 2 changed G5 items (§E-G5).
§CLOSE
The call: Whole-unit verification: the 12-point board all green; five findings ruled on the record.
What changes: Every artifact's gauntlet is closed; the board ran 0 errors across validator, clickthrough, figures and the 480px sweep.
## §CLOSE — whole-unit verification findings (adjudicated 2026-10-08, operator) - F1 EXECUTED: the two content-stale bundles (S04 re-lettering; S06 label draw-order) re-cut to the shipping specs, both leak-PASS. No re-reads owed: both change classes were post-blind, adjudicated, prose/param-only, Role-A verified on the built page. - F2 RULED: the four README-layer variants STAND as honest snapshots of their cut-time contract text (S08's README carries an on-record disposition a re-cut would silently drop); content layers are what sequester — now 17/17 byte-current at the content layer. - F3 ADJUDICATED here: the 7 assessment-tier restatement hits are the two already-adjudicated classes (named-substance regex shapes; the option-verdict-"it" class) — LEAVE, no defect; this line is their adjudication row. - F4 RULED: SPEC_CONTRACT.md is a contract, not a record — no Skill line owed. - F5 EXECUTED: the one artifact without a non-author review (practice) gets its review-questions + review-accuracy pass now (launched); the gauntlet discipline holds uniformly.
§E-C3b
The call: C3 closing pass: pair letters re-lettered off element symbols; the blind reads already postdated the fix round.
What changes: No keys changed; no further reads owed.
## §E-C3b — C3 closing-pass work order (adjudicated 2026-10-08, operator) - F1 ADOPTED: AST-022's pair letters re-lettered J/L, M/R, G/T (element-symbol collision — "F" beside real-ion rows); stems/tables/alts/fb together; keys unchanged. - F-L1 ADOPTED: the L04 one-law idea line rewritten as recorded (strength-and-direction contradiction). - S06-F1 ADOPTED: removecompare drawn-near-electron fix at template level in the C3 lane's own figlib module (drop the disc in the distance panel or dot on the rim), re-render both instances. - Polish rows → the engine close-out list (equal-height label dedupe; pairs with the §C3-S2 two-decimal card). Leave rows stand. - BLIND-STAGE NOTE (operator correction to the record): the C3 blind bundles were generated FROM THE POST-FIX SPECS and the blind read (124/124) postdates the fix round — the 16 changed items' fresh cold reads are already DONE; no further read owed from that round. This application round changes no keys/options, so none arise here either.
One verdict card each: what it holds, and its gate cluster's five QA records' headline results, parsed from the records themselves.
Your criteria (as you pasted them, 7 Oct 2026 — the doc): lessons are good enough if they are 8/10 on these.
Gut checks: Are we confident that this is better than the alternative? Are we confident that a student would benefit from taking this lesson?
The doc's tab is the physics one; its accuracy line reads as chemistry accuracy here. Wording otherwise verbatim.
Sequence S01 of 10
10 lessons · 63 checks (27 MCQ + 36 numeric) · 14 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Sequence S02 of 10
5 lessons · 38 checks (24 MCQ + 14 numeric) · 20 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Sequence S03 of 10
8 lessons · 61 checks (49 MCQ + 12 numeric) · 8 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Sequence S04 of 10
5 lessons · 33 checks (33 MCQ) · 12 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Sequence S05 of 10
7 lessons · 46 checks (35 MCQ + 11 numeric) · 7 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Sequence S06 of 10
7 lessons · 45 checks (38 MCQ + 7 numeric) · 27 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Sequence S07 of 10
6 lessons · 34 checks (31 MCQ + 3 numeric) · 10 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Sequence S08 of 10
5 lessons · 31 checks (28 MCQ + 3 numeric) · 5 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Sequence S09 of 10
9 lessons · 52 checks (49 MCQ + 3 numeric) · 12 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Sequence S10 of 10
4 lessons · 32 checks (30 MCQ + 2 numeric) · 8 figures
Read it as a student on the course page.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Gate U1-G1
8 items · 0 figures · cold-built, never-reused items
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Gate U1-G2
10 items · 3 figures · cold-built, never-reused items
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Gate U1-G3
10 items · 4 figures · cold-built, never-reused items
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Gate U1-G4
10 items · 1 figures · cold-built, never-reused items
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Gate U1-G5
10 items · 2 figures · cold-built, never-reused items
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Practice U1-P1
1 FRQ, 2 rubric points · 1 figures
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Practice U1-P2
2 FRQs, 4 rubric points · 1 figures
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Practice U1-P3
1 FRQ, 2 rubric points · 0 figures
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Unit test U1-T1
18 MCQ + 2 FRQ items (the reserved 2021 Q2 excerpt) · 4 figures
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
The whole unit
10 sequences · 66 lessons · 5 gates · 3 practice pages · 1 test — every number below is parsed from the records, never retyped
One tap here rules on the unit as a whole — the swap from the legacy page stays yours either way.
Judge it on your criteria (the 8/10 bar):
The full box sits above the verdict cards.
Everything you tapped or typed on the review tab gathers here, in page order. It saves in this browser as you go; sending twice is safe — later notes supersede earlier ones.
Everything lands under APCHEMU1 with your name on it. If a send fails, the note is kept in this browser and the page retries by itself.