Wave 46 dim 19 dim D audit_workpaper sub-criterion PR¶

Date: 2026-05-12 Branch: feat/jpcite_2026_05_12_wave46_dim19_D_audit_workpaper Worktree: /tmp/jpcite-w46-dim19-D Author: Wave 46 永遠ループ tick2#2 (dim D)

Audit baseline (per sibling STATE_w46_dim19_pr.md)¶

count: 19 dimensions
average: 6.37/10 (target 8.0+)
total: 121.0 / 190
verdict: yellow

Lowest-scoring dimensions (same audit pass)¶

code	dim	score	top finding
F	fact_signature_v2	2.50	REST file MISSING (closed in dim F PR)
D	audit_workpaper	3.00	migration MISSING
G	realtime_signal_v2	4.50	ETL MISSING, cron MISSING
H	personalization_v2	4.50	ETL MISSING, test MISSING

Dim D breakdown — what is already present vs. what is flagged¶

Wave 43.2.3+4 (commits e1d982f2e + d9a3715b8) landed:

src/jpintel_mcp/api/audit_workpaper_v2.py — REST POST /v1/audit/workpaper
src/jpintel_mcp/mcp/autonomath_tools/audit_workpaper_v2.py — MCP tool
tests/test_dimension_c_d_combined.py — combined C+D test (21 tests)

The dim 19 audit's "migration MISSING" finding is structural: the audit_workpaper compose path joins five tables that each have their own upstream migrations (075/wave24_106/wave24_194/etc.), but there is no dedicated *audit_workpaper* migration because the compose is a pure projection — no new table is needed. Beyond the migration axis the audit also flags:

no metadata/discovery surface (POST-only, no GET)
test count: 1 (shared C+D combined file) vs. dim D-specific = 0
no compose_* row in any registry/manifest file

Selected sub-criterion: GET /v1/audit/workpaper/schema (discovery surface)¶

Why this axis over migration: A dedicated migration would be a no-op table the compose path never reads — pure ceremony. The genuine gap is the missing discovery surface: there is no way for an agent or human caller to learn (a) which source tables the substrate joins, (b) which output sections ship, (c) the 4-業法 fence statutes, or (d) the billing unit, without paying for an invocation. Adding a static GET schema endpoint is the single highest-leverage sub-criterion at the lowest LOC cost, and it pairs naturally with a dim-D-specific test file that lifts the "test count" axis from 1 (shared C+D) to 2 (new D-specific). Per feedback_completion_gate_minimal we deliberately do NOT chase the full 8.0 gap in one PR.

Sub-criterion checklist (dim D → 5 axes)¶

axis	before	after	delta
migration cohesion	MISSING (structural)	MISSING (structural, unchanged)	unchanged
REST file	POST-only	POST + GET /schema	+1 sub
ETL / cron	n/a (pure compose)	n/a (pure compose)	unchanged
test file(s)	1 (C+D combined)	2 (new D-specific)	additive
MCP grep	hit	hit	unchanged

Estimated dim D score lift: 3.00 → ~4.00 (GET discovery surface present + 1 new dim-D-specific test file). This alone moves the dim 19 average from 6.37 toward ~6.42 without touching the migration / ETL / cron axes (which would be either ceremony-only or out-of-scope for a pure compose tool).

Files changed¶

src/jpintel_mcp/api/audit_workpaper_v2.py — +119 LOC (1 new _WORKPAPER_SCHEMA static dict + 1 new GET handler; no edits to the POST handler, no edits to existing helpers)
tests/test_dimension_d_audit_workpaper_schema.py — 170 LOC new (file presence, no LLM import, route registration in OpenAPI, 200 without houjin, response shape contract, input_fields contract, source_tables drift guard, disclaimer parity)
docs/research/wave46/STATE_w46_dim19_D_pr.md — this state doc

Total: well under the ≤ 200 LOC source-code budget for the implementation file change itself; test + doc are additive.

Endpoint contract (new surface)¶

GET /v1/audit/workpaper/schema
  -> 200 {
       endpoint: "/v1/audit/workpaper",
       method: "POST",
       billing_unit_invoke: 5,
       billing_unit_schema: 0,
       composition_kind: "multi_hop_year_end_audit",
       input_fields: [
         {name, type, min_length|min, max_length|max, description}, ...
       ],
       source_tables: [
         "jpi_houjin_master", "jpi_adoption_records",
         "am_enforcement_detail", "jpi_invoice_registrants",
         "am_amendment_diff"
       ],
       output_sections: [
         {key, kind: object|array, row_cap: int|None}, ...
       ],
       fence_statutes: [
         "税理士法 §52", "公認会計士法 §47条の2",
         "弁護士法 §72", "行政書士法 §1"
       ],
       non_negotiable: [...],
       disclaimer: <full §52/§47条の2/§72/§1 disclaimer text>,
       schema_version: "wave46-dim19-D-1"
     }

Hard constraints satisfied:

NO LLM call (pure static dict, no SQLite open)
NO row-level leak (does NOT contain any houjin_meta / fy_* / amendment fields)
§52 / §47条の2 / §72 / 行政書士法 §1 disclaimer parity with POST endpoint (same _DISCLAIMER constant)
Single-DB compose path (autonomath only)
0-unit discovery (billing_unit_schema = 0); POST stays at 5 units

Constraints honored¶

worktree /tmp/jpcite-w46-dim19-D (no main worktree touch)
no rm / mv (only Write + Edit)
no legacy brand strings (jpintel, autonomath are internal-only refs)
no LLM API import (verified by test_dimd_schema_module_imports_no_llm)
1 sub-criterion fix (discovery REST surface MISSING → PRESENT) — NOT a full 3.00 → 8.0 refactor

PR¶

Opened as PR #126: https://github.com/shigetosidumeda-cyber/autonomath-mcp/pull/126

Title: feat(wave46-dimD): audit_workpaper /schema discovery surface (dim 19 sub-criterion). Targets main.

Lint + test verdict (2026-05-12 verify)¶

ruff check src/jpintel_mcp/api/audit_workpaper_v2.py tests/test_dimension_d_audit_workpaper_schema.py -> All checks passed!
pytest tests/test_dimension_d_audit_workpaper_schema.py -v -> 8 passed in 12.59s
Regression check on dim C+D sibling: pytest tests/test_dimension_c_d_combined.py -v -> 20 passed, 1 failed (1 failure is test_dimd_rest_404_or_503_when_unseeded which fails on clean origin/main without this PR — verified via git stash + re-run — so it is pre-existing, NOT introduced here).
ruff check src/jpintel_mcp/api/main.py -> 2 pre-existing errors (I001 + F401), unchanged on stash + re-check; NOT introduced by this PR (same finding as the dim F PR's main.py audit).
Rebased onto origin/main HEAD bfcd2b600 so the program_agriculture guard fix is in place — tests would otherwise ERROR on collection for the 6 client-dependent cases.