From a28baa61974c347c2fe63dc6a2dcbd67d09e72d2 Mon Sep 17 00:00:00 2001 From: Your Name Date: Wed, 13 May 2026 17:34:46 +0800 Subject: [PATCH] docs(awooop): record t13 quality classification deployment --- docs/LOGBOOK.md | 83 +++++++++++++++++++ ...-04-15-MASTER-ai-autonomous-flywheel-v2.md | 7 ++ 2 files changed, 90 insertions(+) diff --git a/docs/LOGBOOK.md b/docs/LOGBOOK.md index 15053640..fe23a5d2 100644 --- a/docs/LOGBOOK.md +++ b/docs/LOGBOOK.md @@ -7591,3 +7591,86 @@ HTML contains: AwoooP 治理總覽 / 自動化品質 / 不可宣稱完整閉環 - 產線資料仍顯示 `verified_auto_repair_total=0`、`production_claim=false`，因此目前正確說法是「真相可見度已補上」，不是「完整 AI 自動修復閉環已完成」。 - 下一步要進 T13：收斂 `execution_unverified`，把 post-execution verification / auto_repair durable record / learning writeback 的缺口從可見化推進到真正閉環。 - 目前整體進度更新：約 76%。 + +### 2026-05-13 — AwoooP truth-chain T13：NO_ACTION / audit-only 不再誤算成自動修復執行（production verified） + +**目的**： + +- T12d production quality summary 顯示不少 `execution_unverified`，但 live trace 發現其中多數其實是 `NO_ACTION` 或 `ansible_candidate_matched/dry_run` audit row。 +- 這些 row 是「純觀察 / 候選稽核」，不是 AI 自動修復執行；若算成 execution，Operator 會誤以為「AI 修了但沒驗證」。 + +**變更**： + +- truth-chain 新增 effective execution 判定： + - `playbook_executed` 且 `output.reason=NO_ACTION` / action 含 `NO_ACTION` / `OBSERVE` / `INVESTIGATE` → `noop_operation_records` + - `status=dry_run`、`ansible_candidate_matched`、`ansible_execution_skipped` → audit-only，不算有效修復執行 +- `automation_quality.facts` 新增： + - `effective_execution_records` + - `noop_operation_records` + - `audit_only_operation_records` +- `_truth_status()` / `build_automation_quality()` 都改用 effective execution，避免 NO_ACTION 把 stage 推成 `execution_succeeded` 或 verdict 推成 `execution_unverified`。 + +**local verification**： + +```text +DATABASE_URL=postgresql+asyncpg://u:p@localhost:5432/db pytest tests/test_awooop_truth_chain_service.py tests/test_platform_router_order.py -q +21 passed + +ruff check --select F821 src/services/awooop_truth_chain_service.py tests/test_awooop_truth_chain_service.py +OK + +python3 -m py_compile src/services/awooop_truth_chain_service.py tests/test_awooop_truth_chain_service.py +OK +``` + +**production deploy / smoke（完成）**： + +```text +Gitea: +2054 ai-code-review cecadb33 -> success +2053 CD Pipeline cecadb33 -> success + tests -> success + build-and-deploy -> success + post-deploy-checks -> success +Deploy marker: 2314bade chore(cd): deploy cecadb3 [skip ci] + +K8s image: +awoooi-api 192.168.0.110:5000/awoooi/api:cecadb331badac7aa4fb07922b892875c28a891a +awoooi-worker 192.168.0.110:5000/awoooi/api:cecadb331badac7aa4fb07922b892875c28a891a +awoooi-web 192.168.0.110:5000/awoooi/web:cecadb331badac7aa4fb07922b892875c28a891a + +health: +https://awoooi.wooo.work/api/v1/health -> 200 + +summary endpoint smoke, hours=24, limit=30: +schema_version=automation_quality_summary_v1 +incident_total=30 +evaluated_total=30 +verified_auto_repair_total=0 +production_claim.can_claim_full_auto_repair=false +score_buckets={green: 0, yellow: 0, red: 30} +by_verdict: + manual_required_no_action=18 + received_only=12 + execution_unverified=0 +top gate_failures: + auto_repair_recorded=30 + execution_recorded=30 + evidence_collected=22 + approval_state=18 + mcp_gateway_observed=12 + outbound_recorded=12 + timeline_recorded=12 + +detail smoke: +INC-20260513-CF5DCE -> stage=manual_required / verdict=manual_required_no_action / effective_execution_records=0 / noop_operation_records=1 +INC-20260513-553113 -> stage=manual_required / verdict=manual_required_no_action / effective_execution_records=0 / noop_operation_records=1 / audit_only_operation_records=1 +INC-20260513-42FCEC -> stage=manual_required / verdict=manual_required_no_action / effective_execution_records=0 / noop_operation_records=1 +``` + +判讀： + +- T13 已完成並推版：現在 Operator Console / truth-chain 不會再把 NO_ACTION 或 Ansible candidate audit 誤認為真正修復執行。 +- 產線結果更誠實：目前不是「修了但未驗證」，而是「18 筆需人工判斷、12 筆只收到告警、0 筆可宣稱已驗證自動修復」。 +- 下一步 T14 應從「分類校正」進到真正閉環：讓可安全處理的低風險事件產生 durable `auto_repair_executions`、post-execution `verification_result`、KM / learning writeback；不能再用 NO_ACTION 假裝自動修復。 +- 目前整體進度更新：約 78%。 diff --git a/docs/superpowers/specs/2026-04-15-MASTER-ai-autonomous-flywheel-v2.md b/docs/superpowers/specs/2026-04-15-MASTER-ai-autonomous-flywheel-v2.md index a67829d2..eff02b1e 100644 --- a/docs/superpowers/specs/2026-04-15-MASTER-ai-autonomous-flywheel-v2.md +++ b/docs/superpowers/specs/2026-04-15-MASTER-ai-autonomous-flywheel-v2.md @@ -2025,6 +2025,13 @@ Phase 6 完成後 - T12d：Operator Console `/awooop` 首頁新增「自動化品質」面板；summary endpoint 改為 public aggregate 且清空 `examples`，source-level `/truth-chain/{source_id}` 仍需 operator auth。production image `356bfce2c8663c46933df4a9050dfaa9f594436a`、Gitea runs `2050/2049` success、health 200、頁面含 `自動化品質` / `不可宣稱完整閉環`。 - 目前總體判讀：真相鏈可見度已從 Telegram 卡片補到 DB / truth-chain API / Run Detail / Operator Console，但 latest production aggregate 仍是 `verified_auto_repair_total=0`。下一步 T13 必須收斂 `execution_unverified`：post-execution verification、auto-repair durable record、learning / KM writeback 缺口要從「可見」推到「閉環」。 +**T13 NO_ACTION / audit-only quality classification production verified（2026-05-13 台北）**： +- 觸發：T12d production summary 顯示 `execution_unverified`，但 live trace 證實多數其實是 `NO_ACTION` 或 `ansible_candidate_matched/dry_run` audit row，被 truth-chain 誤算成有效修復執行。 +- 修正：truth-chain 新增 effective execution 判定；NO_ACTION / OBSERVE / INVESTIGATE row 計入 `noop_operation_records`，`status=dry_run`、`ansible_candidate_matched`、`ansible_execution_skipped` 計為 audit-only，不再推動 `execution_succeeded` / `execution_unverified`。 +- Production：`cecadb33 fix(awooop): exclude audit-only ops from repair quality` 已部署，Gitea runs `2054/2053` success，deploy marker `2314bade`，API/Worker/Web image 均為 `cecadb331badac7aa4fb07922b892875c28a891a`，health 200。 +- Smoke：quality summary `hours=24&limit=30` 由舊的 `execution_unverified=11` 校正為 `manual_required_no_action=18`、`received_only=12`、`execution_unverified=0`、`verified_auto_repair_total=0`、`production_claim=false`。 +- 判讀：T13 完成的是「真相分類校正」，不是自動修復閉環。下一步 T14 必須讓可安全處理的低風險事件產生 durable `auto_repair_executions`、post-execution `verification_result`、KM / learning writeback；禁止再用 NO_ACTION 或 dry-run audit 假裝自動修復。 + --- ### 2026-04-20 晚 (台北) — C1-C4 全流程串接 — Playbook 鏈路保護（commit de2d34d）