0 · Compiler honesty flags (read first)

1. Cleanest single ledger_write schema of the four compiles. All 12 lessons share the identical field set (anchors, bloom_reached, exit_ticket_score, auto_score, self_score, calibration_gap, starting_mastery, journal_prompt, structure_used, referents_used, next_lesson_seed) — no Devon/Priya mid-arc field drop, and bloom_reached is always a real Bloom word, never a placeholder (contrast Devon's L07–12).
2. Drift moved to the arc boundary instead of the schema. L01–L06 headers read sequence: N of 6; L07–L12 read N of 12 — the arc was retconned mid-stream. L06's own next_lesson_seed declares "Arc complete" and names two candidate next topics (Band-1 D1, or a B2 follow-up) — neither is L07 (B3 Trust Calibration). This compile's version of the "two generator halves" seam every prior compile found — a false arc-close, not a schema break.
3. Zero real post-arc numerics anywhere — 0/11 anchors, no arithmetic ever attempted. Every auto/self/exit/calibration field, all 12 lessons, is the literal string "". Unlike Devon's partial "0.NN -> " strings, no lesson here even attempts the math. Cleanest-but-emptiest of the four.
4. Only 3 of 11 anchors carry any numeric baseline anywhere — A1 (0.55), B1 (0.40), D1 (0.30) — an exact match to her profile's three demonstrated_level entries. Unlike Devon/Priya, whose L01 diagnostic expands the profile to 5 baselined anchors, Maya's L01 touches only these same 3 — the narrowest diagnostic reach of the four compiles.
5. † Dagger for AIGHVA — mismatched band key on D1. L11 tags anchors: [AIHC.2.D1] (Band 2, "amended," multi-agent + cold-reader handoffs) but keys its starting_mastery to AIHC.1.D1: 0.30 (Band 1, "chain assistance across one task" — the profile's actual anchor). The prose is honest about the stretch ("her profile holds D1 at band 1... the arc's deliberate stretch lesson"), but the field never establishes a numeric Band-2 baseline; a reader keying strictly off anchors: finds none. Recorded as written; needs a ruling.
6. Cross-learner contamination: clean, both directions. Maya's own profile line (AIHC.1.D1: 0.30 # no multi-agent experience) is the original source of a phrase the Frank compile already flagged as lifted, uncredited, into Frank's L11. Checked in reverse here: none of Maya's 12 lessons cite Devon/Priya/Frank figures. Clean.
7. No continuity overclaims found — cleanest of the four on this axis. All 12 "work at hand" callbacks trace to real prior-lesson content (checked L02 through L12); unlike Devon's L09 flag, none invents antecedent work.
8. exit_ticket_score is a fourth SIM field absent from Devon/Priya/Frank's ledgers — noted for round-3 schema unification, not corrected here.

1 · Mastery by standard (11 AIHC anchors touched, Band 1→2)

† = §0.5's dagger. NO DATA = never generated; nothing invented (§0.3).

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.

A1 Specification      0.55 ███████████░░░░░░░░░ → end NO DATA
B1 Verification        0.40 ████████░░░░░░░░░░░░ → end NO DATA
D1 Orchestration       0.30 ██████░░░░░░░░░░░░░░ → end NO DATA  (key mismatch †, §0.5)
A2 A3 B2 B3 C1 C2 C3 D2 — start NO DATA → end NO DATA, all eight

Three anchors have a real starting bar (the profile's exact three); zero have a real ending bar — matching Devon's 0/11 end-state floor exactly, on a narrower starting base.

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    █    ·    ·    █    ·    ·    ·    ·    ·    █    ·    3/11
Taught            █    █    █    █    █    █    █    █    █    █    █   11/11
Start numeric    █    ·    ·    █    ·    ·    ·    ·    ·    █†   ·    3/11
End numeric      ░    ░    ░    ░    ░    ░    ░    ░    ░    ░    ░    0/11

█ = present   ░ = missing/placeholder-only   · = out of that cell's scope   † = §0.5

Every anchor has exactly one dedicated lesson; the diagnostic and start-numeric rows are the narrowest of the four compiles (3/11, vs. Devon/Priya's 5–6/11) — the emptiness is structural (profile design), not a generation failure, per §0.4.

4 · Per-lesson record

Calibration gap = auto-score minus self-score, the metacognitive signal Playbook §8 names first-class. All values ; none compute until a real self_score exists.

SIM = . "(note)" = calibration_gap carries "computed once Maya's self_score exists" in every lesson, uniformly (no bare-vs-noted split like Priya's).

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

• J01 — Which of the three anchor rows surprised you most when you actually looked at your own work instead of your instinct about your own work — and what's one guess about why?
• J02 — Which of your three contract drafts today took the most rewriting before it passed your own unverifiable-line audit — and what does that tell you about where your specifications tend to go soft?
• J03 — Which checklist item was hardest to place on the 2×2, and did your lesson-1 gut-call match where it actually landed once you applied reversibility/blast-radius instead of instinct?
• J04 — Would you have caught invoice.py's caller without running plan mode first — what does that tell you about how much elicitation is doing for you versus how much you assume your own reading catches?
• J05 — Put your lesson-1 sentence next to today's four-leg record. What did your old habit systematically never check?
• J06 — How many of your three log-entry gap fields actually name the agent, versus your contract, your rule placement, or a mechanism you hadn't mapped — what does that ratio change?
• J07 — Count your approval prompts before/after the lanes went live. Did any surviving ask-lane prompt get MORE attention than it used to — is that what the lanes were for?
• J08 — How did it feel to see your own name on records that can't say what you did versus what the agent did — which direction of mis-attribution were you closer to?
• J09 — What did the paste-ritual give you that the table doesn't, and which row do you most expect pushback on?
• J10 — What you believed about time saved, versus the reconstructed baseline, versus the first logged number — which would you have defended in standup without noticing it was a vibe?
• J11 — What couldn't the fresh session figure out from your handoff artifact — how much of your daily work would fail that same cold-read test?
• J12 — What does "actually trust" mean now that lesson-1 you would not have written — which artifact would you defend hardest if deletion were proposed?

5 · What the tutor reads next (personalization compounds)

1. Cannot rank anchors by post-arc mastery — none exists (§0.3); a real scoring pass is needed before any "weakest anchor" ranking is trustworthy.
2. Can rank by starting weakness among the three scored anchors: D1 (0.30, Band-1-keyed under a Band-2 anchor, †) lowest, then B1 (0.40), then A1 (0.55).
3. Must resolve the D1 dagger (§0.5) first — a Band-2 anchor with no Band-2 numeric baseline risks miscalibrating the next D1 touch's scaffold, the same way Devon's orphaned C2 did; route to AIGHVA.
4. Backfill priority for A2/A3/B2/B3/C1/C2/C3/D2: taught (11/11, §3) but zero baseline and zero post-arc number — one real scoring pass on all eight makes the next arc compound instead of reset.
5. Seed-chaining held for content even where numerics didn't, with one real gap: the qualitative thread runs cleanly L02→L11 (each next_lesson_seed is exactly what the following lesson teaches) — except L06→L07, where L06's seed named two candidates and L07 is neither (§0.2). This compile's answer to the two-halves pattern the others found in schema instead.
6. Standing cadence, not a lesson: L12's weekly freshness beat (rotate-thirds, delta-join, stricter triggers on the disclosed MCP/Subagents unread tails) is the next real trigger.