[Phase 16] S9 — Compliance evidence integrity and GDPR erasure completeness
まだ誰も着手していません。
評価
調査の方向性
Start with _artefact_targets_for_run and the _legacy_staging_owned_by marker pattern in _purge.py, then trace how generate_training_manifest() and export_compliance_artifacts() name the bundle files. The first step is a writer-to-reader integration test that exports a real bundle and purges by run id. Done means the matching run's bundle is fully removed, a wrong run id removes nothing, and pipeline_manifest.json and audit_log.jsonl survive.
索引モデルが issue の本文から書いたものです。
説明
Roadmap · Phase 16, step S9 · size L · 5 unit(s) · planned in
docs/roadmap/phase-16-trust-surface-hardening.md:553-581
Summary
The compliance surface claims a stronger evidentiary state than it holds. purge --kind artefacts cannot find the compliance bundle ForgeLM itself writes: the exporter emits fixed basenames while the resolver looks for file names that embed the run id. It fails safe (deletes nothing, emits data.erasure_failed with NoMatchingArtefacts, exits 1), so the defect is completeness, not destruction. The trainer also writes data_governance_report.json there, and the pipeline writes compliance/pipeline_manifest.json, which belongs to the whole pipeline and must not be deleted by one run's purge.
Units
| Unit | Tracked by (October 2026 review) |
|---|---|
F-W20260729-TRUST-02 |
#101 · #102 · #477 · #554 · #635 |
F-W20260729-TRUST-13, F-W20260729-TRUST-01 |
#386 · #651 |
F-W20260729-TRUST-14 |
#386 · #554 |
F-W20260729-TRUST-07 |
#386 · #679 · #680 · #681 · #715 |
Plan
- Rewrite
_artefact_targets_for_runto the allow-list + ownership-proof model the design document already specifies, mirroring the_legacy_staging_owned_bymarker pattern in the same file. - Breaking changes planned here:
AuditLogger.log_eventbecomes fail-closed;reverse-piiandpurgemove to non-zero exits on states that returned 0.
Exit criteria
- A real bundle produced by
generate_training_manifest()+export_compliance_artifacts()is fully removed by a matching run id and untouched by a wrong one, whilepipeline_manifest.jsonandaudit_log.jsonlsurvive — a writer → reader integration test, not a synthetic-filename fixture; the fabricated docstring examples in_purge.pyare deleted. - A malformed line sharing an identifier blocks the completion record.
-
reverse-piireturns non-zero whenAuditLoggerinitialisation raises. - An
OSErrorinjected separately at the manifestopen,fsyncandos.replacesteps never yields silent success. - A
required → rejected → requiredchain is served by one shared ordered-state helper across the listing and mutation surfaces.
How to deliver
- One root-cause family, committed on its own; then an Opus review round and a Sonnet review round, each with verified findings fixed and committed before the next step begins.
- A reproducer that is red before the fix and green after; the full gauntlet from
CLAUDE.mdpasses under./.venv/bin/python. - Every English documentation edit ships with its Turkish mirror in the same commit; a new or renamed audit event gets its EN and TR catalog rows in the same PR.
- A step that grows a deferred module carries its
budget_historyjustification in the same diff; a widened guard rolls out non-strict first, then flips. - Commit bodies cite
Absorbs F-W20260729-…; the CHANGELOG[Unreleased]section gets an entry only for a change a user or library consumer can observe (decision C-20).
Planned in docs/roadmap/phase-16-trust-surface-hardening.md (read at f94595f). Unit IDs (F-W20260729-…) come from the 2026-07-29/30 full-project review; the issue numbers next to them are the October 2026 review's records of the same defects.
- 主要言語
- Python
- スター
- 9
- フォーク
- 1
- 平均マージ
- 4時間 3分
- マージ済み PR（30日）
- 3
環境構築
- Dockerfile または Docker Compose ファイルあり
- プルリクエストのテンプレートあり
- コントリビューションガイドを読む
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
HodeTech/ForgeLM のほかの issue
-
area: dev-tooling bug severity: low source: roadmap wave: 4
難易度 2/5 1〜3時間 初心者へのやさしさ 70/100
-
area: site bug good first issue severity: low source: review-2026-09 wave: 4
難易度 2/5 1〜3時間 初心者へのやさしさ 66/100
-
documentation good first issue severity: medium source: review-2026-09 wave: 3
難易度 2/5 1〜3時間 初心者へのやさしさ 75/100
-
documentation good first issue severity: low source: review-2026-09 wave: 4
難易度 1/5 1〜3時間 初心者へのやさしさ 80/100
-
documentation severity: medium source: review-2026-09 wave: 3
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
HodeTech/ForgeLM の issue をすべて見る
似ている issue
-
難易度 1/5 1時間未満 初心者へのやさしさ 60/100
521xueweihan/HelloGitHub#3924 ·
-
難易度 2/5 1〜3時間 初心者へのやさしさ 67/100
wilbowes/EchoMuse#869 · コメント 1 件 ·
メンテナーはふだん 1 日以内に返信
-
難易度 1/5 1時間未満 初心者へのやさしさ 85/100
-
namespace operations
難易度 1/5 1時間未満 初心者へのやさしさ 72/100
EclipseFdn/open-vsx.org#14043 ·
メンテナーはふだん 1 日以内に返信
-
test: TestServeUntilStale races the server's close against the client's sendall (BrokenPipeError under load)対応中かも @evoludigit が今日担当しました。 オープン
難易度 1/5 1時間未満 初心者へのやさしさ 89/100
メンテナーはふだん 1 日以内に返信