dot-iu-cutter v0.5 — First Constitution Snapshot MARK Dry-run · Command Review Report

Phase: v0_5_constitution_first_snapshot_MARK_dryrun_command_review · Nature: command_review_only__no_execution · Date: 2026-05-18 · doc 5 of 5

review_status: READY_FOR_GPT_USER_DRYRUN_EXECUTION_APPROVAL
dryrun_executed: false ; decision_authority: GPT / User ONLY ; self_advance: PROHIBITED

---

1. Outcome

The first Constitution snapshot-MARK dry-run is command-review complete and READY for a GPT/User execution-approval decision. The historic R9 blocker is CLOSED: the no-DB-write snapshot→manifest entrypoint is now committed (afb7bfc) and 21/21-green. An exact, runnable command was authored against the committed argparse surface — not the superseded design pseudo — so it will not crash on unknown flags.

ready_for_execution_approval: YES
blocking_items: NONE (only 2 informational micro-items routed to GPT — §4)

2. Key correction vs the prior design pseudo (material)

prior_pseudo_flags_NOT_in_committed_code (would cause argparse error / exit 2):
  --identity-region  --exclude-markers  --address-template  --leaf-floor  --emit-levels
committed_code_behaviour: hardcodes BEGIN/END sentinels, the 4-marker map, exclusion logic,
  address_template at.icx.const.v4, the DIEU leaf floor, and OD-G3 (emits NGUYEN_TAC +
  KIEN_TRUC_SECTION + DIEU) — so those flags are unnecessary AND must be omitted.
--emit is accepted but IGNORED; 5 fixed-name files are always written.
⇒ The exact command in doc 2 §1 is the authoritative one; the design-doc pseudo is superseded.

3. Verified state (read-only; nothing executed)

repo:           /Users/nmhuyen/iu-cutter-build/repo/iu-cutter
branch:         feature/constitution-snapshot-mark-dryrun
HEAD:           afb7bfcc9b7bbb953bb00159479c9611e6ac4bd1   (== committed entrypoint, QG1)
main:           4367c83… untouched (not merged)
working_tree:   CLEAN before and after readiness checks (no artifacts produced)
dryrun.py sha:  f1f42e83…2efa1422 (committed, unchanged)   test sha: 454d9fc8…f2843a4a (ratified)
unit_CI rerun:  python3 -m unittest tests.test_dryrun_snapshot_mark → 21/21 OK (this phase)
fixture region: recomputed (read-only) → sha 17660443…cae80c · len 17522 ·
                markers {enacted:19,controlled_draft:1,draft:1,obsolete:1}  → EXACT vs pinned
no_DB_env:      PG_DSN/DATABASE_URL/DIRECTUS_URL/PGPASSWORD all unset

4. Micro-items routed to GPT/User (NOT self-decided; non-blocking)

M-1 snapshot_artifact provenance string (doc 1 §5):
  committed code reads a LOCAL file path; the KB snapshot path is MCP-only (no local
  `knowledge/` dir) ⇒ input MUST be the committed byte-faithful local fixture. Region
  identity is bound by the sha gate (17660443…), not the path string.
  recommendation OPT-P1: pass tests/fixtures/constitution-normalized-17660443e0f23e99.md ;
  accept that string in provenance (sha is the cryptographic binding). [alt OPT-P2/P3 listed]
M-2 out-dir is a LOCAL ephemeral scratch ($WD/manifest, 0700, outside the repo); the 5
  artifacts are uploaded to the KB output folder by the operator AFTER an approved run.
  recommendation: accept (keeps git tree clean; KB folder is the deliverable destination).
both are informational; neither blocks execution approval.

5. Deliverable summary (5 docs in this package)

1 command-readiness     : repo/branch/HEAD/CLI-surface/source-identity/R1..R9 → READY
2 command-package       : the EXACT runnable command (vs committed argparse) + behaviour + outputs
3 verification-plan     : P-1..P-8 pre-checks + V-1..V-20 post-run + net-zero proof
4 risk-cleanup-plan     : risk register, artifact-only cleanup, no production rollback, stop conds
5 command-review-report : THIS — readiness verdict; route GPT/User
allowed_outputs (when an approved run is executed): manifest.json · review_evaluation.json ·
  coverage_proof.json · determinism_digest.md · dryrun_report.md  (5 fixed names)
KB output folder: knowledge/dev/laws/dieu44-trien-khai/v0.5-constitution-first-snapshot-mark-dryrun-output/

6. Quality gates

QG1 uses committed entrypoint afb7bfc            : PASS (command pinned to HEAD + sha precheck)
QG2 exact command defined                         : PASS (doc 2 §1, validated vs committed code)
QG3 no-DB / no-CUT / no-VERIFY guard defined      : PASS (doc 2 §2; mandatory flags; DB-env guard;
    import-isolated stdlib-only ⇒ DB write impossible by construction)
QG4 output folder + files defined                 : PASS (doc 2 §3, doc 3 §2)
QG5 verification + cleanup defined                 : PASS (doc 3 + doc 4)
QG6 dry-run NOT executed                           : PASS (authored only; no `cutter_agent.dryrun`
    run, no parse, no DB, no network)
QG7 STOP after uploading 5 files                   : PASS (exactly 5 docs; STOP here)

7. Explicit non-execution statement

• Performed: read-only KB (SSOT) study; read-only git/branch/HEAD/sha checks; authorized unit-test rerun (21/21, no DB); read-only python recompute of the fixture region sha (NOT the dry-run entrypoint); read of committed dryrun.py to derive the exact CLI.
• NOT performed: no cutter_agent.dryrun execution; no CUT; no VERIFY; no production IU; no production DB write; no source_document/source_version mutation; no schema change; no Directus/vector mutation; no deploy/restart; no merge to main; no push/tag/remote; no code change; no commit; no self-advance. Git tree CLEAN; HEAD afb7bfc; main untouched.
• Exactly 5 docs under …/v0.5-constitution-first-snapshot-mark-dryrun-command-review/.
• doc 5 of 5 — STOP. Route to GPT / User for dry-run execution approval (and the two non-blocking micro-rulings M-1/M-2). Self-advance to execution PROHIBITED.

Companion docs: command-readiness (1), command-package (2), verification-plan (3), risk-cleanup-plan (4).