Checkpoint — Text-as-Code Reuse / Anti-Duplication Audit

Date: 2026-06-09 · Status: REUSE_AUDIT_READY · Production mutation: NO Nature: read-only reuse/anti-duplication audit before any Implementation Package DOT design.

1. Why this ran

The User corrected the first feasibility plan: this work is fundamentally Text-as-Code, the system already spent weeks building the miếng-thông-tin / Information Unit / TAC foundation, and the first plan found only a small part of it. Mandate: deeply audit what already exists and maximize reuse BEFORE any new tool design — to avoid wasted rework and system conflict.

2. Method

Five parallel READ-ONLY auditors swept: (1) TAC/IU foundation + completed-state; (2) Đ44 construction/lifecycle/checker tree; (3) tools/checker/dry-run patterns (KB + local dot-iu-cutter repo); (4) Đ39 KG / orphan / impact / duplicate; (5) Đ43 Context Pack + paired-DOT. KB = agent-data MCP (SSOT for knowledge/dev/…).

3. Decisive findings (correcting the first plan)

• The first plan's "PARTIAL / mostly design-only" framing understated the foundation. Live PG holds ~117 fn_iu_* functions + ~45 IU/TAC tables; edit/apply/save, supersede/enact/retire, merge/split, structure-ops (plan/apply/verify/rollback), three-axis envelope, vector-sync, event-outbox + routing worker + dead-letter, notification, birth/gateway write-guards, and a MARK→CUT→VERIFY pipeline are deployed (fn_iu_op_*, schema iu_core).
• A runnable fail-closed Article-14 verifier already exists: cutter_agent/dryrun.py + cli.py selftest + AST isolation tests.
• fn_tac_log_checker_issue (checker logger → system_issues) is deployed in production (S183).
• A 19-gate preflight runner already RAN (86-unit migration, all PASS) with Validation Contract + exact-key rollback.
• Đ43 paired-DOT is deployed (build rev 11 / verify rev 5, VPS, cron) — the exact template (no-fallback dot_config_get, §5.8 5-guard, generic executor_type dispatch, dual-checksum, staging→promote).
• Graph/orphan/impact/duplicate machinery mostly deployed: universal_edges (CAT-130, ~2,199 rows) + v_kg_edges_all, Đ8 entity_dependencies, Đ19 orphan scanners, Đ14 3-tier (engine TD-083 pending), One-Roof island_detected/double-owner. Single sink = system_issues.

4. Conflicts / overlaps that BOUND scope (pre-existing, not created here)

• Two cutting/verify/manifest lineages: iu_core PG-native (fn_iu_op_*) ↔ cutter_governance external code-agent (/opt/incomex/dot, HEAD e93424b). Building a third = triplication.
• TAC↔IU duality unresolved: tac_logical_unit (86 rows, real corpus) ↔ information_unit.
• Two checker_run_status namespaces (P11E AP-CHECKER-* proof-layer vs P6 real-DOT) — keep separate.
• Two dry-run lineages (P9-G6 DB-schema/bash vs dot-iu-cutter Python/artifact-only) — don't import DB-write assumptions into an artifact-only verifier.
• Duplicate-authority detection already exists (Art.14/Đ14/One-Roof/FIX7) — the tool's duplicate-authority check must INTEGRATE, not fork (else it self-violates).
• Data-quality caveat: audit sweeps reported conflicting information_unit counts (175 vs 98) — a single fresh read-only baseline is required; treat all counts as indicative.

5. True new work (narrow, after reuse)

(1) command-runner that captures exit codes (dryrun.py refuses to run anything); (2) prose-claim↔executable-test binding; (3) document-level path-alias/canonical-id + duplicate-authority resolver that executes; (4) generic package_manifest.json envelope + schema; (5) --selftest N/N + module_sha256 self-pin; (6) audit_dead_links() over declared refs. Everything else = reuse.

6. Recommendation

REUSE_EXTRACTION_FIRST (not PROCEED_TO_SCOPE_SPEC, not BLOCKED). Next: R0 owner decisions (canonical cutting lane; TAC↔IU authority) → R1 one fresh read-only baseline → R2 Extraction Map (each asset: copy/call/adapt + what-not-to-touch; tool = generic Đ23 Inspector DOT calling deployed engines, no new authority) → R3 gap-only mini-spec → R4 FIX7 pilot via reuse.

7. Documents

• Created: reports/text-as-code-reuse-anti-duplication-audit-2026-06-09.md (the audit).
• Created: this checkpoint.
• Updated: 00-index.md (added report + checkpoint rows; noted the first plan's maturity correction).

8. NOT done (hard constraints honored)

❌ no implement · ❌ no install · ❌ no production mutation (read-only only) · ❌ no FIX7 resume · ❌ no new law · ❌ no v0.1 scope-lock spec · ❌ no verifier code · ❌ no schema files · ❌ no PG/system_issues wiring · ❌ no standalone Safety Kit.

9. Next step

User / GPT / Codex answer the six §9 questions (esp. Q1 canonical cutting lane + Q2 TAC↔IU authority). Then R1→R3. The scope-lock spec stays deferred until the Extraction Map exists.