Wave 47 Phase 2 tick#4 — Dim R federated_mcp partner ETL actual run¶

date: 2026-05-12 branch: feat/jpcite_2026_05_12_wave47_phase2_dim_r_ingest ref: PR #167 (storage substrate landed) — this STATE doc is the actual ETL run snapshot that loads the 6 curated MCP partner rows. parent feedback: feedback_federated_mcp_recommendation

Scope¶

Wave 47 Phase 2 closes Dim R by actually running the seed ETL (scripts/etl/seed_federated_mcp_partners.py) against the local data/autonomath.db. The storage substrate (migration 278) was landed in PR #167 and only created the empty am_federated_mcp_partner + am_handoff_log tables; this tick exercises the seed path so the catalogue actually contains the 6 partner rows the recommendation surface needs.

This run does NOT touch production. It is the local dev-DB seed that exercises the same code path the production cron will execute, and the row count + idempotency are recorded here as the audit trail.

Hard constraints honoured¶

NO external MCP server call. Pure local SQLite UPSERT.
NO LLM API import. grep -iE "^(import|from)\s+(anthropic|openai|langchain|llama_index|claude)" on seed_federated_mcp_partners.py + 278_federated_mcp.sql returns zero hits (matches feedback_no_operator_llm_api).
No legacy brand. Identifiers/comments use jpcite only.
No PRAGMA quick_check on autonomath.db (per feedback_no_quick_check_on_huge_sqlite).
No rm/mv. Worktree isolation via /tmp/jpcite-w47-phase2-dim-r/ (separate from main worktree).

Steps executed¶

1. git worktree add /tmp/jpcite-w47-phase2-dim-r \
       -b feat/jpcite_2026_05_12_wave47_phase2_dim_r_ingest main
2. sqlite3 data/autonomath.db < scripts/migrations/278_federated_mcp.sql
3. python3 scripts/etl/seed_federated_mcp_partners.py \
       --db data/autonomath.db --dry-run
   ->  INFO: federated_mcp partner seed dry-run:
            inserted=6 updated=0 (total=6)
   ->  exit=0
4. python3 scripts/etl/seed_federated_mcp_partners.py \
       --db data/autonomath.db
   ->  INFO: federated_mcp partner seed applied:
            inserted=6 updated=0 (total=6)
   ->  exit=0
5. python3 scripts/etl/seed_federated_mcp_partners.py \
       --db data/autonomath.db   # idempotency probe
   ->  INFO: federated_mcp partner seed applied:
            inserted=0 updated=6 (total=6)
   ->  exit=0

Row count snapshot¶

$ sqlite3 data/autonomath.db \
      "SELECT COUNT(*) FROM am_federated_mcp_partner;"
6

6 partner rows verified¶

partner_id	name	server_url	capability_tag	last_health_at
freee	freee 会計	https://mcp.freee.co.jp/v1	accounting\|invoice\|tax	NULL
mf	マネーフォワードクラウド	https://mcp.moneyforward.com/v1	accounting\|invoice\|payroll	NULL
notion	Notion	https://mcp.notion.com/v1	doc\|kb\|collab	NULL
slack	Slack	https://mcp.slack.com/v1	chat\|notify\|workflow	NULL
github	GitHub	https://mcp.github.com/v1	code\|issue\|pr\|review	NULL
linear	Linear	https://mcp.linear.app/v1	issue\|product\|cycle	NULL

last_health_at = NULL is the seed default. The out-of-band HTTPS-HEAD probe (scripts/cron/) fills it later; callers treat NULL as DEGRADED per the migration 278 contract.

Index + audit log shape verified¶

$ sqlite3 data/autonomath.db \
      "SELECT name FROM sqlite_master \
       WHERE type='index' AND tbl_name='am_federated_mcp_partner';"
sqlite_autoindex_am_federated_mcp_partner_1   -- PK
idx_am_federated_mcp_partner_capability
idx_am_federated_mcp_partner_health

$ sqlite3 data/autonomath.db "PRAGMA table_info(am_handoff_log);"
0|handoff_id|INTEGER|0||1
1|source_query|TEXT|1||0
2|partner_id|TEXT|1||0
3|response_summary|TEXT|1|''|0
4|requested_at|TEXT|1|strftime('%Y-%m-%dT%H:%M:%fZ','now')|0

am_handoff_log is the append-only audit trail. Wave 47 Phase 2 tick#4 does not write to it directly (no recommendation traffic in this run); a separate seed run is unnecessary because the table is populated on live recommendation calls.

Bug-free verify¶

Gate	Result
ETL `--dry-run` exit	0
ETL `--apply` exit (1st run)	0
ETL `--apply` exit (2nd run, idempotency)	0 (inserted=0 / updated=6)
6 partner row presence (freee/mf/notion/slack/github/linear)	All 6 present
`am_federated_mcp_partner` CHECK constraint	`length(capability_tag) > 0` holds
LLM API import (anthropic/openai/etc.)	0 hits in ETL + migration
External MCP server call	0 (pure local sqlite3)
PRAGMA quick_check on autonomath.db	not run (per feedback)
Legacy brand (税務会計AI/AutonoMath/zeimu-kaikei.ai)	0 in ETL + STATE
`am_handoff_log` schema (5 cols incl. AUTOINCREMENT)	verified
Indexes (capability, health)	both created

Cost posture¶

feedback_no_operator_llm_api honoured: 0 Anthropic/OpenAI/Claude SDK import, 0 paid API call, 0 yen incurred. Future cron partner_health_probe.py will use HTTPS HEAD (non-LLM) for the periodic last_health_at refresh.