0 · Compiler honesty flags (read first)

1. Schema drift, two generator halves. L01–06 frontmatter: anchor:/ anchors_diagnosed:; ledger_write carries auto_mastery: "X -> "; bloom_reached is a real word (analyze, apply…). L07–12: combined standard: … · tags […] frontmatter; ledger_write drops auto_mastery entirely, adds tags:, and sets bloom_reached to the literal string "" — not a placeholder Bloom word — despite bloom_target being logged in every frontmatter.
2. Zero real post-arc numerics on ANY lesson. Unlike the Priya compile (L01–06 wrote real end-mastery, e.g. 0.60→0.78), Devon's L02–06 auto_mastery reads literally "0.NN -> " — the arithmetic step was never generated, only the start carried forward. L07–12 never write the field at all. No anchor has a real numeric end-state anywhere in source. None invented here (Frank's compile's ~0.NN [SIM] illustrative numbers for zero-baseline anchors are the anti-pattern this avoids).
3. 5 of 11 anchors have zero numeric baseline anywhere (no profile entry, no L01 diagnostic reading): C1, C2, C3, D1, D2.
4. † Dagger-flag for AIGHVA — orphaned/mismatched C2. Profile carries AIHC.1.C2 = 0.35; no lesson ever evidences AIHC.1.C2. L09 instead teaches AIHC.2.C2 (Band 2), no Band-1 antecedent, no reference to the 0.35 reading — not a mis-citation (checked, #5), but a profile datum that goes nowhere plus an ungrounded Band-2 anchor. Needs a ruling: backfill Band-1 C2, or amend the profile.
5. Cross-learner contamination check: none found. Every baseline number checked against Maya/Priya/Frank's profile figures — none lifted (contrast: Frank's compile caught a real instance on D1). Stated for the record, not assumed clean.
6. Lower-confidence flag, not gated. L09 claims Devon "already sort[s] flows into permitted / prohibited / needs-a-rule (L01–06)" — no lesson there runs that exercise. Possible continuity overclaim; flagged for lesson-level QA, not corrected here.

1 · Mastery by standard (11 AIHC anchors touched, A1–D2)

† = post-arc value never generated (§0.2). ‡ = zero numeric baseline anywhere (§0.3). Orphaned profile datum AIHC.1.C2 = 0.35 sits outside this table — no lesson evidences it (§0.4).

2 · Viz 1 — Mastery by anchor, start → end (SIMULATED)

Scale: each █ = 0.05 mastery (0–1.0, 20 cells). Bars drawn only where a real number exists; NO DATA written plainly rather than invented.

A1 Specification     0.50 ██████████░░░░░░░░░░ → end NO DATA †
A2 Delegation         0.20 ████░░░░░░░░░░░░░░░░ → end NO DATA †
A3 Elicitation        0.10 ██░░░░░░░░░░░░░░░░░░ → end NO DATA †
B1 Verification       0.30 ██████░░░░░░░░░░░░░░ → end NO DATA †
B2 Failure literacy   0.25 █████░░░░░░░░░░░░░░░ → end NO DATA †
B3 Trust calibration  0.30 ██████░░░░░░░░░░░░░░ → end NO DATA † (field absent, L07)
C1 Provenance         start NO DATA ‡ → end NO DATA †
C2 Governance         start NO DATA ‡ dagger → end NO DATA †
C3 Measurement        start NO DATA ‡ → end NO DATA †
D1 Orchestration      start NO DATA ‡ → end NO DATA †
D2 Regeneration       start NO DATA ‡ → end NO DATA †

Six anchors have a real starting bar; zero have a real ending bar — the emptiness is itself the finding.

3 · Viz 2 — Coverage: 11 anchors × 12 lessons

Anchor          A1   A2   A3   B1   B2   B3   C1   C2   C3   D1   D2
Lesson          L02  L03  L04  L05  L06  L07  L08  L09  L10  L11  L12
Diagnosed L01    █    █    █    █    █    ·    ·    ·    ·    ·    ·    5/11
Taught            █    █    █    █    █    █    █    █    █    █    █   11/11
Start numeric    █    █    █    █    █    █    ░    ░    ░    ░    ░    6/11 ‡
End numeric      ░    ░    ░    ░    ░    ░    ░    ░    ░    ░    ░    0/11 †

█ = present   ░ = missing/placeholder-only   · = out of that lesson's scope

Every anchor has exactly one dedicated lesson; the two numeric rows are where the instrument leaks — worse on the end-numeric row than either prior compile.

4 · Per-lesson record

Calibration gap = auto-score minus self-score, the construct Playbook §8 names first-class (Claude's metacognitive signal, not the learner's — next lesson reads it: under-confidence → encourage, over-confidence → re-verify, per the Cell-29 trace). All values below ; none compute until a real learner runs this arc.

SIM = . ‡ = bloom_reached is the literal placeholder string in source (§0.1), not a real Bloom word — "target" is pulled from frontmatter bloom_target for reference only. "(noted)" cal. gap = L02–06's field carries an explanatory string; "(bare)" = L07–12's carries only the placeholder with no note (§0.1).

Journal slots — ALL EMPTY by design (Playbook §8: never generated for him)

• J01 — Of the five gaps found today, which worries you most as the name on this rollout — why that one, not the others?
• J02 — Which spec element took longest to get right, and what does that say about which one your team will skip without a template?
• J03 — Which workflow was hardest to place on the capability/cost grid, and what would change it in six months?
• J04 — Did the elicitation answer surprise you? If it just confirmed what you suspected, was asking still worth it?
• J05 — Claim 2 was a wrong label, not wrong math. Where else might a number be real but mislabeled — how would you catch that again?
• J06 — Which enforced rule depends on a process outside Claude entirely — what does that say about where real risk lives?
• J07 — Which workflow did you mark "policy only, not platform-enforced" — what would make you notice if it stopped being followed?
• J08 — What did you assume "the platform will just log this" meant, before this lesson — and how would you correct it now?
• J09 — Which policy rule surprised you most once you checked real console enforcement — what if you'd never checked?
• J10 — Before this lesson, what number would you have given the board for "is it working" — does it survive today's entries-vs-scores test?
• J11 — If your drafting staffer had quit mid-quarter, what would have been lost — which handoff artifact saves most of it?
• J12 — Reread "Auto mode for all three, to reduce staff friction" against today's register. What would you tell the admin who wrote the first line?

5 · What the tutor reads next (personalization compounds)

The next generated lesson reads THIS file, not a reset baseline:

1. Cannot rank anchors by post-arc mastery — none exists (§0.2); a real scoring pass must happen before any "weakest anchor" ranking is trustworthy.
2. Can rank by starting weakness among the six scored anchors: A3 (0.10) and A2 (0.20) are lowest — next-touch priority once real post-lesson numbers confirm gains.
3. Must resolve the C2 dagger-flag (§0.4) first — teaching forward on AIHC.2.C2 with no Band-1 antecedent and no baseline risks the wrong scaffold level; route to AIGHVA.
4. Backfill priority for C1/C2/C3/D1/D2: taught (11/11, §3) but zero baseline — only a real scoring pass makes personalization compound instead of reset here.
5. Seed-chaining held even though numerics didn't: each next_lesson_seed consumed the prior lesson's real finding (L01's retention flag → L05's verified gap → L06's enforced rule → L09's console-vs-policy sort → L12's register) — the qualitative thread compounds; only the quantitative one is broken.
6. Standing cadence, not a lesson: L12's regeneration register (monthly beat, WATCH list, mismatch-report trigger) is the next real trigger, firing on platform drift.