dot-iu-cutter v0.5 WS-2 — Report

Phase: design-only (WS-2 Metadata Key Registry + Source Profile) · 2026-05-18 · Self-contained.

1. Executive summary

Concrete hóa lớp chống hardcode cho Information Unit Fabric: metadata_key_registry (D1), source_family_registry + grammar_profile + authority_semantics mapping (D2/D3/D4), entity_reference_registry core-minimal (D5), canonical_address namespacing giải OD-A1 (D6). Tất cả logical/pseudo-schema, KHÔNG tạo bảng/migration/code/dry-run. WS-1 đã được GPT promote làm authority; WS-2 build trên đó. Self-advance PROHIBITED — STOP, route GPT/User review.

2. What was read

• 6 file WS-1 authority package (do agent này soạn phiên trước; trong context): existing-design-map, WS1 brief, open-decisions, report, edge-APR-minimization, binding-authority-clarification.
• 2 GPT reviews: fabric-addendum-scope-ws1 (PASS_WITH_RULINGS) + WS1-cleanup-edge-binding-authority (PASS, WS-1 authority promoted, open WS-2). Đọc full.
• Nền (reference): v0.5 Label/Metadata concept (§2 entities, §5 promotion, §8 OD-L1..L5); v0.5 source-document ingestion (§2 source_document/version/span, parser_profile, document_version_id, source_authority_class — KHÔNG có field source_family); v0.5 canonicalization/address-grammar (§2 grammar_profile model, §3 canon_pipeline C0..C8, §4 address shape + OD-A1 separator inconsistency, §5 iu_id deriv); v0.4 Hybrid (§1 SQL SSOT, §4 normalize queue, P3/P7 anti-hardcode).

3. What was designed

• D1 metadata_key_registry: concrete 11-field pseudo-schema; promotion governance bảng ai-làm-gì; OD-L1..L5 resolved logical (L1 reuse append-only ledger; L2 both-per-key via index_policy; L3 app-level enforce now + index defer; L4 registry hint+telemetry+sovereign; L5 cutter_governance schema).
• D2 source_family_registry: 9-family seed + authority_semantics_default + grammar_profile_ref + version/status policy; 3-trục clarification (source_family vs source_authority_class vs authority_semantics — không nhầm lẫn).
• D3 grammar_profile: 2 concrete profile rows — incomex-architecture-constitution-v4 (NGUYEN_TAC/KIEN_TRUC_SECTION/DIEU + ✅/📋) và vn-national-law (CHUONG/DIEU/KHOAN/DIEM/DOAN); fail-closed detect → review queue; model giữ canon §2.
• D4 authority_semantics mapping: default từ source_family + authority_override logical storage (unit-level dùng ngay, span-level định nghĩa+defer cơ chế); 3 ví dụ mixed-authority; assembly output bắt buộc gán role.
• D5 entity_reference_registry: core 5 field + 2 deferred + entity_kind_registry seed (sql_entity/code_module/git_file/directus_item/report_path); governance; permission/snapshot defer (mở khi pilot contract thật).
• D6 canonical_address namespacing: scheme <DOCPREFIX>/<L1>-...-<Lk>, docprefix trong source_document registry UNIQUE; ví dụ collision Hiến pháp nội bộ vs Luật DN 2020 "Điều 5"; legacy DIEU_28 giữ nguyên.
• WS-2 dependency note: governed_by/derived_from là P44-4A candidate edges chờ v1.0 (GPT cleanup §3) — tracked, không tự kích hoạt.

4. OD resolved / deferred

• Resolved (logical, recommendation): OD-L1, OD-L2, OD-L3 (policy), OD-L4 (policy), OD-L5 (recommend), OD-A1 (scheme recommended).
• Deferred + justify: physical tables tất cả → schema-design cycle Q5 (sovereign-gated); OD-L4 threshold số → post-pilot benchmark; OD-A1 separator slash/hyphen → GPT/User confirm (canon doc tự mâu thuẫn, không tự quyết cuối); span-level override cơ chế chi tiết → pilot mixed-authority; permission/snapshot policy → pilot contract thật; OD-G3 status-in-address → giữ "không encode status", chi tiết canon OD-G3.

5. Risks

• R-WS2-1: registry shape lệch khi schema-design cycle thật. Mitigate: pseudo-schema bám sát v0.5 concept field names; physical defer có sovereign gate.
• R-WS2-2: governed_by/derived_from candidate edge không được ban hành v1.0 → edge-minimization resolution phải revisit. Mitigate: WS-2 dependency note; gate trước WS-3 execution (GPT ruling).
• R-WS2-3: separator OD-A1 chốt khác recommendation → address scheme phải sửa. Mitigate: flag rõ cho GPT, không tự quyết.
• R-WS2-4: nhầm 3 trục (source_family/authority_class/authority_semantics). Mitigate: §0 bảng phân biệt rõ trong D2.

6. Next steps

1. GPT/User review 4 file WS-2.
2. Nếu PASS: WS-2 + WS-1 = authority; mở schema-design cycle (Q5) cho physical tables (sovereign-gated) HOẶC tiếp WS-3 readiness.
3. APR draft evidenced_by (parallel/conditional, GPT/User approve — không tự làm).
4. WS-3 assembly-by-topic cross-source chỉ sau WS-2 + edge-APR readiness (GPT routing).
5. WS-4 sau WS-1+WS-2 + prerequisite 23-P2.

7. Forbidden respected

KHÔNG: schema migration/CREATE/ALTER/INDEX, production write, code change, CUT/VERIFY/dry-run, Directus mutation, vector/NoSQL, create edge types/approve APR, git commit, deploy/restart, redesign WS-1/v0.5 concept, self-advance WS-3/WS-4. Tất cả output pseudo-schema YAML, không executable SQL.

8. Git / access

• no repo/VPS access in this session (working dir /Users/nmhuyen, không phải git repo, không SSH).
• code_changed: false · commit_made: false · branch/HEAD/git status: không truy cập được → không báo.

9. Quality Gates self-check

• QG1 không redesign WS-1/v0.5 concept, chỉ concrete hóa: ✅ (reference path+section cụ thể)
• QG2 không vi phạm forbidden: ✅
• QG3 mỗi design item ≥1 ví dụ nôm na: ✅ (sổ nhãn kho, thư viện, thẻ kho, địa chỉ nhà)
• QG4 pseudo-schema/YAML, không executable SQL: ✅
• QG5 source_family ≥9 families: ✅ (đúng 9)
• QG6 grammar_profile ≥2 concrete (internal law + external gov law): ✅
• QG7 mỗi file ≤25K (preferred ≤15K): ✅
• QG8 OD-L1..L5 + OD-A1 addressed: ✅ (resolve hoặc justify defer)

10. Routing

4 file đã upload knowledge/dev/laws/dieu44-trien-khai/v0.5-fabric-addendum-scope/. DỪNG. Route GPT/User review. Agent self-advance PROHIBITED.