Skill: atomise (audit mode) + ap-toolbox CED cross-check @ 1fed1da
Unit 4 — operator adoption of the parked decisions (option b, James 2026-09-24)
James chose option (b): the operator makes the register calls with his veto reserved; he checks the atomisation alongside. Each decision below is ADOPTED as stated unless he vetoes. One recommendation is FLIPPED on fresh evidence (D1); one carries a logged follow-up (D3).
CED cross-check (first live use of the Incept ap-toolbox, 2026-09-24)
The register maps all 21 Unit 4 essential-knowledge codes of the 2024–2025 CED, exactly — no
gaps, no extras — across all nine topics (4.1–4.9). Source: ced-library.inceptstore.com
apchem_enriched.json, diffed locally against unit4_design_v1.json mapsToStandards; result
in ced_crosscheck_u4.json. (The whole-course /v1/analyze endpoint expects a nine-unit
payload — noted for a later whole-course validation; unit-level diff is the meaningful check
here.)
The six decisions
D1 — Percent yield: FLIPPED to INCLUDE. The original recommendation (exclude — no CED anchor) predates the exam-registry evidence: 2022 long FRQ Q1(b), 1 point, CED-tagged Unit 4 topic 4.5, asks the student to JUSTIFY a percent yield below 100% (dissolution during rinsing). James's meta-directive — bias toward what the actual AP asks — governs. Adopted: the two-atom letter-suffix insertion after CRX-046, shaped to the evidence (the real part is reasoning, not plug-in arithmetic): - CRX-046A (definition + calculation): "Calculate a percent yield as actual yield divided by theoretical yield times 100%." - CRX-046B (reasoning): "Explain why an actual yield can fall below the theoretical yield [e.g., product lost in rinsing or transfer, incomplete reaction], and justify whether a given percent yield is consistent with a described lab step." Implementation: in the first spec wave, with the validator run; percent yield comes OFF the explicitExclusions list. IDs letter-suffixed so the register stays stable.
D2 — Evidence categories as ONE profile atom: ADOPTED as recommended (Unit 3 v2 precedent; no separately-failable performance in splitting; documented exception to the enumeration-split signal).
D3 — Full redox balancing deferred to Unit 9: ADOPTED, with a follow-up logged. Checked today: Unit 9's register has the electrochemistry atoms (TEC-074–085 use half-reactions) but does NOT yet carry the deferred solution-balancing atoms (adding H₂O/H⁺/OH⁻) nor a retrieval treatment of CRX-080–084. FOLLOW-UP (blocking for the U9 build, not for U4): amend unit9_design with the balancing atoms + retrieval note before U9 specs are written.
D4 — Write/Balance verbs + equation_construction_task / equation_evaluation_task modalities + the pen-and-paper production contract: ADOPTED as recommended (mirrors James's drawing ruling; keeps the unit's core production behaviours visible to the modality layer and the production-item requirement).
D5 — Forward arrow (→) for ionization in Unit 4, ⇌ introduced in Unit 7: ADOPTED as recommended (a symbol before its defining atom is the AST-001 defect class; progressiveSharpening handoff carries the upgrade).
D6 — Reorder: particle-level basis (CED 4.4) immediately after evidence of change (4.1): ADOPTED as recommended (James's 2026-07-31 what-then-why ruling; JM-P15 authorizes departing from CED order; crosswalk keeps traceability; validator confirms no internal dependency violated).
What happens next (Build process v2)
On James's register sign-off (or silence + no veto on this record): blueprints per topic
(unit4_build/blueprints/BLUEPRINT_<topic>.md) go to James for approval BEFORE any prose —
starting with U4-S01 (evidence of change) as the sample; then write-lesson runs from approved
blueprints; the question stage runs the Incept qgen trial (generated candidates vs authored,
both through review-questions + blind read, James judges).