IU b–f Execution Readiness Refactor — 06 Underload Self-Check + Verdict (Outcome G)

Self-check against knowledge/dev/laws/prompt-muc-tieu-mo-for-claude-code.md (100000x v1.3, rev41). The mission required that, if the work is shallow, this macro must NOT report PASS. Below is an honest depth audit, the six mandated self-review answers, and the final verdict.

---

1. Standard-fit check (v1.3, 100000x)

v1.3 requirement	This pack
Self-contained-after-clear	Each authority prompt (doc 03) restates mission, sources, channel, class, evidence, PASS rules — pasteable cold. ✔
Execution-Channel-Pack	Every mutating macro names SSH workflow_admin + the exact pipe command + the container-/tmp gotcha. ✔
Live-Apply-Hard-Gate-0	Prompts 3/4/5/6 (C2/C3/C4) carry Hard-Gate-0; C1 prompts correctly omit it (no live write). ✔
Design-Only-Macro Mode	This macro is C5 document-only; declared up front. ✔
No-underload	7 docs, full correction matrix (22 rows), 6 prompts + 1 named follow-up, gate/law matrix, board, this self-check. See depth audit §3. ✔
Live-ready: use-not-rewrite	Every prompt reuses existing live functions; explicit "do NOT re-implement" lines. ✔
No-hardcode / DOT / Constitution	DOT pairs under Đ35, audit under Đ31, axis-B vocab sourced from Constitution. ✔
Adaptive scale	Class-appropriate effort: C1 proofs at 30000x, gate/governance at 50000x. ✔

---

2. The six mandated self-review answers

1. Did I only follow the checklist, or did I solve the mission? Solved the mission. The checklist asked for correction + re-classing + authority prompts; the mission was to make the next phase un-foot-gunnable. I went beyond the listed items by: (a) identifying the systemic root cause (collapsing C1 and C2 into one "read-only" label) rather than patching each doc individually; (b) catching a meta-defect the baseline self-review missed — its own "no mutation" PASS was wrong because its prompts authorized writes — and turning it into a new required check (execution-class consistency); (c) distinguishing cross-cutting laws (Đ30/Đ31) from owner laws so the no-double-ownership claim is actually defensible rather than asserted.

2. Are all read-only/mutation boundaries corrected? Yes. Doc 01 corrects all 6 read-only mislabels (docs 02/03/04/07 + prompts 1/2), all 3 audit-table omissions from zero-delta proofs, and clarifies the reversibility-vs-read-only conflation in doc 07 §3. The two honestly-labeled mutating docs (05/06) were verified and confirmed correct. The single dry-run audit-row ambiguity (B-11) is flagged for explicit resolution in the C3 prompt rather than left silent.

3. Can the next Agent execute without asking the user? Yes. Each prompt in doc 03 is self-contained: it names its KB sources by path, its execution class, its channel, its Hard-Gate-0, its forbidden list, its evidence contract, and its PASS/PARTIAL/BLOCKED rules. Preconditions (e.g., "gate protocol must be implemented + approved") are explicit, so an Agent that pastes Prompt 5 cold will stop correctly if M-D isn't live.

4. Are the outputs evidence-ready? Yes. Every prompt specifies the evidence shape (SELECT scripts, row counts, sample JSON, before/after object diffs, BEGIN/ROLLBACK transcripts, post-run gate-state SELECTs) — addressing GPT criticism #9 ("no explicit large-scale evidence requirement") and #10 ("no explicit stop-if-underloaded"). Each prompt ends with a self-check the Agent must answer before PASS.

5. Are the 4 Mothers still blocked until IU readiness? Yes. The no-go is restated in doc 00 §4, doc 02 §4 (invariant 4), doc 04 §3 (only-new-owner = future Điều XX, deferred), and doc 05 §3. No macro M-A…M-F advances 4 Mothers; Wave 5 surveys are the gate into the 4-Mothers decision, and even those end at a Council-approval gate.

6. Did I leave any unsafe prompt in circulation? No — and I explicitly marked the discard. Command-pack doc 10 prompts 1 and 2 are declared must-not-paste-as-is (doc 01 §2, B-17/B-18). Prompts 3/4/5 there are superseded by doc 03's cleaner versions. The baseline pack remains valid only as reference, not execution authority. This pack's doc 03 is the canonical replacement.

---

3. Depth / underload audit (honest)

• Coverage: all 12 command-pack docs read in full (not just 00+10, which was GPT's own limitation); both read-only proofs, the gate protocol, both mutating tests, the DOT spec, the backlog, and both GPT reviews extracted verbatim.
• Correction matrix: 22 rows, each with a verbatim quote, a why, a corrected class, and a corrected next action — not a summary.
• Prompts: 6 full paste-ready prompts + 1 bounded-but-named follow-up (U5), each ~40–60 lines, execution-classed, with evidence contracts and self-checks.
• Matrices: gate/law matrix (10 macros × 8 attributes) + readiness board (5 tests × 8 attributes) + 8-gate inventory with never-flip rule.
• Not shallow: the pack does not merely restate the baseline; it re-architects the execution model around five explicit classes and proves the no-double-ownership claim rather than asserting it.

Underload verdict: NOT underloaded. The pack is operationally complete; a fresh Agent can execute Wave 1 immediately and each subsequent wave behind its named gate.

---

4. Residual gaps (honest, non-fatal)

• G-1: Individual law texts (Đ0-G, 7, 28, 30, 31, 32, 35, 36, 37, 38/39, 45) were grounded via the survey §11 verified-live ownership and prior extractions, not re-opened this macro. Non-fatal: no future macro's safety depends on a law text not already extracted; ownership is cross-checked and consistent. A future macro may re-read a specific law if it enacts new law text (none here).
• G-2: The U5 cut-state-machine rollback prompt is named and bounded, not fully authored, because its exact shape depends on whether Prompt 5's split/merge traverses the cut state machine — a fact the Prompt-5 Agent confirms at runtime. Authoring it speculatively would over-prescribe. Non-fatal: it is listed as an explicit prerequisite with sources.
• G-3: The prompt-standard version cited in the originating brief ("105000x"/v1.2) lags the live standard (100000x v1.3 rev41). This pack uses the live v1.3 rules. Non-fatal, noted in doc 00 source matrix S1.

None of these blocks the next phase.

---

5. Final verdict

PASS (document-only).

• All unsafe read-only/mutation contradictions corrected (doc 01, 22 rows).
• All future macros execution-classed (C1–C5) and sequenced into 5 safe waves (doc 02).
• 6 paste-ready, open-goal, self-contained authority prompts produced (doc 03) + 1 named follow-up.
• Gate/law matrix exists with a defensible no-double-ownership check (doc 04).
• IU production-readiness board exists (doc 05).
• 4 Mothers remains BLOCKED.
• No production mutation occurred. No PG/Directus/Qdrant write, no migration, no DOT run, no dot_config change, no event registration, no schema/table/template/collection creation, no law enactment, no implementation, no UI deploy, no OSS selection.
• Self-review confirms the pack is not underloaded.

---

6. Forbidden-compliance statement

This macro complied with every forbidden in the mission brief §7. Verified, item by item:

Forbidden	Complied
No PG mutation	✔ (only KB writes + local file authoring)
No Directus mutation	✔
No Qdrant/vector write	✔
No migration	✔
No DOT command run	✔
No dot_config gate change	✔
No event type registration	✔
No schema/table/template/collection creation	✔
No law enactment	✔
No implementation	✔
No UI deployment	✔
No final OSS selection	✔

This pack read, reasoned, corrected, and authored. It changed nothing live.