grift-cli 0.5.4__tar.gz → 0.5.6__tar.gz

{grift_cli-0.5.4 → grift_cli-0.5.6}/CHANGELOG.md

@@ -4,13 +4,54 @@
 
 タグ・PyPI・リポ public 化は本ファイルの記載対象外（人間ゲート）。
 
+## [0.5.6] — 2026-08-23
+
+公開後検収で発見の仕様記述↔実挙動の矛盾 3 件（D-1/D-2/D-3）を解消するホットフィックス。
+
+### Fixed
+
+- **動詞の意味を確定: `analyze` = 表示（stdout）・`report` = 記録（`.grift/`）**。README 日英の 2 語動詞表の analyze 行（「`.grift/` を生成」と誤記）を訂正
+- **D-3b**: 裸 `grift analyze` の既定スコープを **repo** に統一（パス明示の従来モードは既定 tenant のまま不変）
+- **D-3a**: `grift report --scope {repo,tenant}`（既定 repo）を追加
+- D-2 文書: report の help「default: ./out」誤記を `.grift/` に訂正・docstring/examples を動詞意味確定版に更新
+- **errata（0.5.5 の記載について）**: 0.5.5 の README/EVIDENCE に「裸 `grift analyze` が `.grift/` を自動生成する」との記述があったが、実挙動は stdout 出力のみ（`.grift/` は `grift report` が作る）。本版で文書をバイナリに合わせ、動詞意味を上記のとおり確定した
+- CHANGELOG の受け口 URL 表記を tep-contributions に統一
+
+### Added（報告規律・恒久）
+
+- **「wheel 実測」節の各 claim には実行コマンドと生ログを添付する**（0.5.5 で裸 analyze の claim が公開 wheel で再現しなかった件の再発防止）
+
+### テスト（反証）
+
+- 裸 analyze が `.grift/` を作らず stdout へ出すピン / 裸 analyze の scope=repo ピン / パス明示の既定 tenant 維持ピン / 裸 report が `.grift/` を書き scope=repo のピン / `report --scope tenant` のピン / help 文言の golden（`./out` 残存なし）
+
+## [0.5.5] — 2026-08-23
+
+UX 統一（2語動詞 + `.grift/` 出力）+ contribute 受け口の実運用化 + 全ドキュメント日英対応。
+
+### Changed
+
+- **基本操作はすべて `grift <動詞>` の2語で完結**: `grift analyze`（対象=カレント・出力=`.grift/`）・`grift report`（常に HEAD を再分析）・`grift verify`（`.grift/report.json` をカレントリポで再計算）・`grift contribute`（`.grift/report.json` から payload 組立）。明示指定は従来どおり引数で
+- **出力先を `.grift/` に統一**（旧 `./out`・`.grift-out` は読み取り互換のみ）。`.gitignore` への `.grift/` 追加を推奨・リポ標準 `.gitignore` に同梱
+- Action の作業ディレクトリも `.grift/` へ・既定ピン 0.5.5
+
+### Added
+
+- **contribute 受け口リポ [grift-contributions](https://github.com/Cor-Incorporated/grift-contributions) を開設**: README（日英）・CONTRIBUTING・schema 検証 CI（`tep-contribution-v1`・メール形状拒否・禁止キー検査）。`grift contribute` の確認文が提出先と手順を明示
+- **全公開ドキュメントの英語版を `docs/en/` に追加**: norms / metrics-guide（v2・帯規約・インベントリ・関連ツール節を含む完全英訳）/ report-schema / export-schema / identity-schema / corpus-protocol / coverage-map / persona-answers
+- 読み方ガイド v2（帯規約の統一定義・全観測インベントリ・関連ツール節・表記確定）
+
+### Removed
+
+- publish-testpypi ワークフロー（TestPyPI シークレット不在による恒久失敗のため削除）
+
 ## [0.5.4] — 2026-08-23
 
 `grift report` の stale 出力問題を解消 + 表示の版残存を一掃。
 
 ### Fixed
 
-- **`grift report`（引数なし）は既存の `./out/report.json` があっても常に現在の HEAD を再分析**する（提唱者実測で検出: 旧仕様は既存 JSON を再表示し、2コミット目以降の SHA が古いまま残った — 自己証明の罹患として重大）。再分析なしの再レンダリングは引数指定時のみ
+- **`grift report`（引数なし）は既存の `./out/report.json` があっても常に現在の HEAD を再分析**する（提唱者実測で検出: 旧仕様は既存 JSON を再表示し、2コミット目以降の SHA が古いまま残った — 自己証明の信頼性を損なう重大な問題として処理）。再分析なしの再レンダリングは引数指定時のみ
 - README 日英: Action 使用例を `@v0.5.4` へ・**PyPI 欄を版に依存しない表記に変更**（「最新 0.5.2」等の古い版番号が残り続ける問題を構造的に解消）・`grift report` を「初回の簡単導線」として条件付きで説明（主 CTA は `grift analyze . --scope repo --out ./out` を維持）
 - action.yml 既定ピンを 0.5.4 へ
 

{grift_cli-0.5.4 → grift_cli-0.5.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: grift-cli
-Version: 0.5.4
+Version: 0.5.6
 Summary: Deterministic TEP evidence metrics from a single git repository. No LLM.
 Project-URL: Homepage, https://github.com/Cor-Incorporated/grift-cli
 Author: TEP contributors
@@ -36,17 +36,27 @@ Named after the Grift product line.
 
 コミット数・行数・「活動量」は、エージェントと生成コードの時代に簡単に膨らむ。grift は git だけから、TEP の帰属（誰の仕事か）と検証層（test co-change ほか）を**決定論的に**出す。LLM は使わない。合成スコアも等級語彙も出さない。
 
-## 5分で始める
+## 5分で始める（2語で基本操作）
 
 ```bash
 pipx install grift-cli
-grift analyze ./my-repo --scope repo --out ./out
+cd your-repo
+grift analyze     # 分析結果を stdout に表示（repo スコープ）
+grift verify      # .grift/report.json を現在のリポジトリで再計算・突合
 ```
 
-- `--scope tenant`（既定）: `.tep/identity.toml` に載った人の仕事 = **証拠用**
-- `--scope repo`: bot 以外の全人間コミット = **プロセス観測・参照分布用**
+基本操作はすべて **`grift <動詞>` の2語**で完結します。動詞の意味は「**analyze = 表示（stdout）**・**report = 記録（`.grift/`）**」です:
 
-**初回の簡単導線**: リポジトリ内で `grift report` だけでも現在の HEAD を分析して `./out/report.md` と `report.json` を書き出します（既存の `./out/report.json` があっても常に再分析します）。**確実に指定条件で測りたいとき・CI では `grift analyze . --scope repo --out ./out` を使ってください**（`report` は repo スコープ固定の省略形です）。
+| 2語コマンド | 動作 |
+|---|---|
+| `grift analyze` | カレントリポジトリを**stdout に表示**（repo スコープ・`.grift/` は作らない） |
+| `grift report` | 分析して `.grift/report.{json,md}` に**記録**（常に現在の HEAD を再分析・既定 repo スコープ・`--scope tenant` 可） |
+| `grift verify` | `.grift/report.json` を同条件で再計算し改ざんを検出（VERIFIED / MISMATCH / CANNOT_VERIFY） |
+| `grift contribute` | `.grift/report.json` から opt-in 提出 payload を組む（**何も送信しない**） |
+
+- **`.grift/` は出力専用ディレクトリ**です（無ければ自動作成）。リポジトリの `.gitignore` に `.grift/` を追加することを推奨します
+- カスタム指定（対象パス・スコープ・出力先を明示）は従来どおり引数・オプションで: `grift analyze path --scope tenant --identity .tep/identity.toml --out dir`
+- `--scope tenant`（既定）: `.tep/identity.toml` に載った人の仕事 = **証拠用**／`--scope repo`: bot 以外の全人間コミット = **プロセス観測・参照分布用**
 
 参照分布 v2026.09 は **repo スコープ同士**でのみ照合する。tenant の値を repo 分布に載せない（混ぜたら分布が嘘になる）。
 
@@ -72,7 +82,18 @@ report.json の全フィールド定義は [docs/report-schema.md](docs/report-s
 
 ## データ提出（grift contribute・明示的 opt-in）
 
-`grift contribute <repo-scopeのreport.json> --out contribution.json` は **TEP Report 集計と参照分布 vNext** への提出 payload を組み立てる。**CLI は何も送信しない（自動送信は恒久禁止）** — payload 全文を表示し、「この提出は公開リポジトリに載る」ことを明示した上で、提出はあなた自身が PR で行う。payload は repo スコープ集計値 + context_profile（クラス級）+ 定義版のみで、canonical_id・メール・パス・repo 名・tenant スコープ値は含まない（`docs/norms.md` 保持・削除条項参照）。
+**TEP Report 集計と参照分布 vNext** への提出は、次の3段階で完結します:
+
+```bash
+grift report                                  # ① repo スコープの report を作る
+grift contribute --out .grift/contribution.json  # ② payload を組み・全文を確認
+# ③ payload を [tep-contributions](https://github.com/Cor-Incorporated/tep-contributions) に PR で提出
+```
+
+- **CLI は何も送信しません（自動送信は恒久禁止）**。payload 全文が表示され、「この提出は公開リポジトリに載る」ことが明示されます
+- payload は repo スコープ集計値 + context_profile（クラス級）+ 定義版のみ。**canonical_id・メール・パス・repo 名・tenant スコープ値は含まれません**
+- 受け口リポ（tep-contributions）の CI が schema（`tep-contribution-v1`）と個人情報形状（メール等）を機械検証します
+- 用途は「TEP Report 集計と参照分布 vNext」に限定。保持期間は次回年次 Report まで・撤回は issue で受け付けます（`docs/norms.md` 保持・削除条項）
 
 例（click、tenant スコープ。ゴールデン G1）:
 
@@ -82,6 +103,10 @@ report.json の全フィールド定義は [docs/report-schema.md](docs/report-s
 
 母数 20 未満では率も分布位置も出さない（`insufficient_population`。件数の生表示のみ）。
 
+## 指標の読み方（非エンジニア向け）
+
+すべての指標について「何を測っているか・高い/低いで何がわかるか・公開リポジトリ117件の分布上の目安」を [docs/metrics-guide.md](docs/metrics-guide.md) に平易にまとめています。初めてレポートを見る方はまずこちらを。誤読しやすい点（例: corrective rework はバグ件数ではない、dormant は放置ではない）も表にしてあります。
+
 ## 指標と限界
 
 | 指標 | 何の証拠か | 限界 |

{grift_cli-0.5.4 → grift_cli-0.5.6}/README.en.md

@@ -13,18 +13,28 @@ The conditions for calling a use "TEP-compliant" are defined in [docs/norms.md](
 
 Commit counts, line counts, and “activity” inflate easily once agents and generated code are in the loop. grift reads git only and emits TEP provenance (whose work) and a verification layer (test co-change and related metrics) **deterministically**. No LLM. No composite score. No skill-rank labels.
 
-## Start in five minutes
+## Start in five minutes (two-word verbs)
 
 ```bash
 pipx install grift-cli
-grift analyze ./my-repo --scope repo --out ./out
+cd your-repo
+grift analyze     # analyze and print to stdout (repo scope)
+grift verify      # recompute .grift/report.json against the current repo
 ```
 
-**Quickest first run**: inside a repository, `grift report` alone analyzes the
-current HEAD and writes `./out/report.md` and `report.json` (it always
-re-analyzes, even if `./out/report.json` already exists). For explicit
-control and CI, use `grift analyze . --scope repo --out ./out` (`report` is a
-fixed repo-scope shorthand).
+Every basic operation is just **`grift <verb>`**. Verb semantics:
+**analyze = display (stdout)** · **report = record (`.grift/`)**:
+
+| Verb | Action |
+|---|---|
+| `grift analyze` | Analyze the current repository and **print to stdout** (repo scope; does not create `.grift/`) |
+| `grift report` | Analyze and **record into `.grift/report.{json,md}`** (always re-analyzes the current HEAD; repo scope by default, `--scope tenant` available) |
+| `grift verify` | Recompute `.grift/report.json` under recorded provenance (VERIFIED / MISMATCH / CANNOT_VERIFY) |
+| `grift contribute` | Build an opt-in submission payload from `.grift/report.json` (**never sends**) |
+
+- **`.grift/` is the dedicated output directory** (auto-created). Adding `.grift/` to your `.gitignore` is recommended
+- Custom invocations keep the explicit form: `grift analyze path --scope tenant --identity .tep/identity.toml --out dir`
+- `--scope tenant` (default): the identity.toml members' work = **evidence** / `--scope repo`: all human commits = **process observation & reference distributions**
 
 - `--scope tenant` (default): work matched in `.tep/identity.toml` = **evidence**
 - `--scope repo`: all non-bot human commits = **process observation and reference distributions**
@@ -59,6 +69,36 @@ The full field-by-field definition of report.json lives in [docs/report-schema.m
 
 `grift analyze <repo> --export <dir>` writes commits.ndjson (one row per commit: origin / actor / cochange), actors.json (attribution and engagement per canonical_id), and export-meta.json (with `config_digest`). **No raw emails or author strings are ever exported.** Aggregates in report.json remain authoritative for narrative. Details: [docs/export-schema.md](docs/export-schema.md).
 
+## Opt-in data submission (grift contribute)
+
+Submissions to the **TEP Report and the next reference distribution** take three steps:
+
+```bash
+grift report                                      # 1) create a repo-scope report
+grift contribute --out .grift/contribution.json   # 2) build & review the payload
+# 3) submit the payload as a PR to tep-contributions
+#    https://github.com/Cor-Incorporated/tep-contributions
+```
+
+- **The CLI never sends anything** (automatic transmission is permanently
+  forbidden). The full payload is printed and the flow states explicitly that
+  it will appear in a public repository
+- The payload carries repo-scope aggregates + class-level context + definition
+  versions only — **no canonical_id, emails, paths, repo names, or
+  tenant-scope values**
+- The intake repository's CI mechanically validates the schema
+  (`tep-contribution-v1`) and rejects email-shaped content
+- Use is limited to "TEP Report aggregation and the next reference
+  distribution"; retained until the next annual Report; withdrawal via issue
+
+## How to read the metrics (for non-engineers)
+
+[docs/metrics-guide.md](docs/metrics-guide.md) explains every metric in plain
+language: what it measures, what high/low values suggest, and rough ranges
+from a 117-repository public corpus. Common misreadings (e.g. corrective
+rework is not a bug count; dormant does not mean abandoned) are listed in a
+table. Read this first if you are new to the reports.
+
 ## Metrics and limits
 
 | Metric | What it is evidence of | Limit |

{grift_cli-0.5.4 → grift_cli-0.5.6}/README.md

@@ -13,17 +13,27 @@ Named after the Grift product line.
 
 コミット数・行数・「活動量」は、エージェントと生成コードの時代に簡単に膨らむ。grift は git だけから、TEP の帰属（誰の仕事か）と検証層（test co-change ほか）を**決定論的に**出す。LLM は使わない。合成スコアも等級語彙も出さない。
 
-## 5分で始める
+## 5分で始める（2語で基本操作）
 
 ```bash
 pipx install grift-cli
-grift analyze ./my-repo --scope repo --out ./out
+cd your-repo
+grift analyze     # 分析結果を stdout に表示（repo スコープ）
+grift verify      # .grift/report.json を現在のリポジトリで再計算・突合
 ```
 
-- `--scope tenant`（既定）: `.tep/identity.toml` に載った人の仕事 = **証拠用**
-- `--scope repo`: bot 以外の全人間コミット = **プロセス観測・参照分布用**
+基本操作はすべて **`grift <動詞>` の2語**で完結します。動詞の意味は「**analyze = 表示（stdout）**・**report = 記録（`.grift/`）**」です:
 
-**初回の簡単導線**: リポジトリ内で `grift report` だけでも現在の HEAD を分析して `./out/report.md` と `report.json` を書き出します（既存の `./out/report.json` があっても常に再分析します）。**確実に指定条件で測りたいとき・CI では `grift analyze . --scope repo --out ./out` を使ってください**（`report` は repo スコープ固定の省略形です）。
+| 2語コマンド | 動作 |
+|---|---|
+| `grift analyze` | カレントリポジトリを**stdout に表示**（repo スコープ・`.grift/` は作らない） |
+| `grift report` | 分析して `.grift/report.{json,md}` に**記録**（常に現在の HEAD を再分析・既定 repo スコープ・`--scope tenant` 可） |
+| `grift verify` | `.grift/report.json` を同条件で再計算し改ざんを検出（VERIFIED / MISMATCH / CANNOT_VERIFY） |
+| `grift contribute` | `.grift/report.json` から opt-in 提出 payload を組む（**何も送信しない**） |
+
+- **`.grift/` は出力専用ディレクトリ**です（無ければ自動作成）。リポジトリの `.gitignore` に `.grift/` を追加することを推奨します
+- カスタム指定（対象パス・スコープ・出力先を明示）は従来どおり引数・オプションで: `grift analyze path --scope tenant --identity .tep/identity.toml --out dir`
+- `--scope tenant`（既定）: `.tep/identity.toml` に載った人の仕事 = **証拠用**／`--scope repo`: bot 以外の全人間コミット = **プロセス観測・参照分布用**
 
 参照分布 v2026.09 は **repo スコープ同士**でのみ照合する。tenant の値を repo 分布に載せない（混ぜたら分布が嘘になる）。
 
@@ -49,7 +59,18 @@ report.json の全フィールド定義は [docs/report-schema.md](docs/report-s
 
 ## データ提出（grift contribute・明示的 opt-in）
 
-`grift contribute <repo-scopeのreport.json> --out contribution.json` は **TEP Report 集計と参照分布 vNext** への提出 payload を組み立てる。**CLI は何も送信しない（自動送信は恒久禁止）** — payload 全文を表示し、「この提出は公開リポジトリに載る」ことを明示した上で、提出はあなた自身が PR で行う。payload は repo スコープ集計値 + context_profile（クラス級）+ 定義版のみで、canonical_id・メール・パス・repo 名・tenant スコープ値は含まない（`docs/norms.md` 保持・削除条項参照）。
+**TEP Report 集計と参照分布 vNext** への提出は、次の3段階で完結します:
+
+```bash
+grift report                                  # ① repo スコープの report を作る
+grift contribute --out .grift/contribution.json  # ② payload を組み・全文を確認
+# ③ payload を [tep-contributions](https://github.com/Cor-Incorporated/tep-contributions) に PR で提出
+```
+
+- **CLI は何も送信しません（自動送信は恒久禁止）**。payload 全文が表示され、「この提出は公開リポジトリに載る」ことが明示されます
+- payload は repo スコープ集計値 + context_profile（クラス級）+ 定義版のみ。**canonical_id・メール・パス・repo 名・tenant スコープ値は含まれません**
+- 受け口リポ（tep-contributions）の CI が schema（`tep-contribution-v1`）と個人情報形状（メール等）を機械検証します
+- 用途は「TEP Report 集計と参照分布 vNext」に限定。保持期間は次回年次 Report まで・撤回は issue で受け付けます（`docs/norms.md` 保持・削除条項）
 
 例（click、tenant スコープ。ゴールデン G1）:
 
@@ -59,6 +80,10 @@ report.json の全フィールド定義は [docs/report-schema.md](docs/report-s
 
 母数 20 未満では率も分布位置も出さない（`insufficient_population`。件数の生表示のみ）。
 
+## 指標の読み方（非エンジニア向け）
+
+すべての指標について「何を測っているか・高い/低いで何がわかるか・公開リポジトリ117件の分布上の目安」を [docs/metrics-guide.md](docs/metrics-guide.md) に平易にまとめています。初めてレポートを見る方はまずこちらを。誤読しやすい点（例: corrective rework はバグ件数ではない、dormant は放置ではない）も表にしてあります。
+
 ## 指標と限界
 
 | 指標 | 何の証拠か | 限界 |

{grift_cli-0.5.4 → grift_cli-0.5.6}/action.yml

@@ -16,7 +16,7 @@ inputs:
   version:
     description: "grift-cli version to install (pinned)"
     required: false
-    default: "0.5.4"
+    default: "0.5.6"
   comment:
     description: "Opt-in: post the shared block as a PR comment (default off)"
     required: false
@@ -36,16 +36,16 @@ runs:
         GRIFT_SCOPE: ${{ inputs.scope }}
       run: |
         set -euo pipefail
-        mkdir -p .grift-out
-        grift analyze . --scope "$GRIFT_SCOPE" --format both --out .grift-out
+        mkdir -p .grift
+        grift analyze . --scope "$GRIFT_SCOPE" --format both --out .grift
         # Shared block = the copy-pasteable summary section of report.md.
         # Extract from '## Shared block' to the next '## ' heading.
-        awk '/^## Shared block/{flag=1;next}/^## /{flag=0}flag' .grift-out/report.md > .grift-out/shared-block.md
-        cat .grift-out/shared-block.md >> "$GITHUB_STEP_SUMMARY"
+        awk '/^## Shared block/{flag=1;next}/^## /{flag=0}flag' .grift/report.md > .grift/shared-block.md
+        cat .grift/shared-block.md >> "$GITHUB_STEP_SUMMARY"
         {
           echo "### grift (observation only — no pass/fail)"
           echo '```'
-          cat .grift-out/shared-block.md
+          cat .grift/shared-block.md
           echo '```'
           echo "_読み方: 本レポートは証拠であり判定ではない（TEP norms 参照: https://github.com/Cor-Incorporated/grift-cli/blob/main/docs/norms.md）_"
         } >> "$GITHUB_STEP_SUMMARY"
@@ -54,7 +54,7 @@ runs:
       uses: actions/upload-artifact@v4
       with:
         name: grift-report
-        path: .grift-out/
+        path: .grift/
         retention-days: 30
 
     - name: Post PR comment (explicit opt-in only)
@@ -66,7 +66,7 @@ runs:
           const body = [
             '### grift (observation only — no pass/fail)',
             '```',
-            fs.readFileSync('.grift-out/shared-block.md', 'utf8').trim(),
+            fs.readFileSync('.grift/shared-block.md', 'utf8').trim(),
             '```',
             '_読み方: 本レポートは証拠であり判定ではない（TEP norms 参照）_'
           ].join('\n');

grift_cli-0.5.6/docs/en/corpus-protocol.md

@@ -0,0 +1,43 @@
+# Corpus protocol (v0.5) — write kill criteria before running
+
+Japanese original: [corpus-protocol.md](../corpus-protocol.md).
+
+Goal: test **discriminatory power** and **game resistance** — not whether
+anyone is "senior." Named-person scorecards are forbidden.
+
+## Groups
+
+| Group | Represents |
+|---|---|
+| A | High verification culture |
+| B | Notable individual works (repo URLs only in public docs) |
+| C | AI-heavy / vibe-coding (2024+) |
+| D | Template / fork / no-tests |
+| E | Private ground truth (Grift-only; never copied here) |
+
+## Kill criteria (registered before measurement)
+
+- Test co-change is discarded if the A-vs-D medians differ by < 0.10 or the
+  ranges fully overlap
+- `corrective_rework_rate` is **observational only** (subject-convention
+  confounding; demoted from evidence). `path_retouch_rate` likewise. Line-level
+  retry is v0.6
+- Survival is opt-in (`--survival`); killed if a repo exceeds 120s wall time
+- If a metric is killed, keep origin/attribution as the floor. Never invent a
+  composite score to replace it
+
+## Tier definitions
+
+Tier 1 (observation status): D expected `not_observed` or pending
+attribution; identity-bearing A expected observed + narratable. Population
+< 20 → no rate, no decile, raw counts only. Tier 2 (rates, only if both
+groups have n ≥ 2 narratable): direction ≥ 0.10, ranges not fully contained,
+Cliff's δ ≥ 0.33 — else `inconclusive_small_n`.
+
+## Verbatim record
+
+Verdicts (separated and fail alike) are recorded verbatim in
+`corpus/DISCRIMINANT-*.md`. v2026.09 history: A vs C `separated` / A vs D
+`inconclusive_small_n` / corrective A vs C `fail_tier2`. v2026.11: A vs D
+`separated` (δ 0.9198, n 18/18) / A vs C `fail_tier2` (δ 0.291 — the grown C
+pool reduced separation; criteria were not lowered).

grift_cli-0.5.6/docs/en/coverage-map.md

@@ -0,0 +1,38 @@
+# Coverage map — questions → observations (v1, introduced with PR-A)
+
+Japanese original: [coverage-map.md](../coverage-map.md). 21 questions
+(adjudicated). ★ = not yet implemented. Every feature addition starts by
+updating this table.
+
+## Hirer/manager questions
+
+| # | Question | Answering observations | Status |
+|---|---|---|---|
+| Q1 | Can they actually build? | test co-change · survival · activity days · core period (existing) | Implemented |
+| Q2 | Which domains/technologies? | language timeline ★ · domain paths ★ · test_frameworks · language_composition / dependency_manifests / monorepo markers (PR-A) | Partial |
+| Q3 | Can they work in a team? | cross_author ★ · co_authored_share ★ · pr_flow_share (PR-A) · collaboration_class (PR-A) | Partial |
+| Q4 | Maintainer or fire-and-forget? | self_maintenance ★ · dependency_update_share ★ · post_release_fixes ★ · fix/revert authorship ★ | v0.6a |
+| Q5 | Transparent about AI-era work? | declared_ai_assist ★ × its co-change · generated_or_vendor | v0.6a (PR-B) |
+| Q6 | Do they persist? | tenure · cadence descriptors ★ · release_cadence / actor_turnover (PR-A, repo level) | Partial |
+| Q7 | Founding experience? | founder timing ★ · scaffold ★ · release creation ★ · pr_flow_share (PR-A) | Partial |
+| Q8 | What scale/state of repos? | context_profile v2 (PR-A) · contribution-time repo scale ★ | Partial |
+
+## Engineer questions (same observations, self-attestation reading)
+
+| # | Question | Status |
+|---|---|---|
+| E1 | Assets, not volume | survival (implemented) · adopted_creations ★ |
+| E2 | Breadth and depth | language_composition (PR-A) · domain composition ★ · portfolio (v0.6b) |
+| E3 | Can handle others' code | cross_author ★ (PR-B) |
+| E4 | Uses AI with verification | declared_ai_assist × co-change ★ (PR-B) |
+| E5 | Returns for maintenance | self_maintenance ★ · post_release_fixes ★ (PR-B) |
+| E6 | Founded something | founder timing ★ · scaffold ★ · release ★ (PR-B) |
+
+## Not answerable from git (boundary, stated explicitly)
+
+Code-review quality · communication · requirements analysis (the first two
+are Grift-side API/human evaluation territory; requirements is permanently
+out). Repo significance/popularity **grades** (permanently excluded — a repo
+grade joins with attribution into a personal grade in one JOIN). Absence of
+private careers / gap contents (outside TEP; norms article 1). "No
+declaration = no AI use" inference (permanently forbidden).

grift_cli-0.5.6/docs/en/export-schema.md

@@ -0,0 +1,22 @@
+# export-v1 — opt-in machine-ingestion output (frozen 2026-08-22)
+
+Japanese original: [export-schema.md](../export-schema.md). Field tables are
+mirrored; the Japanese original is authoritative.
+
+- `grift analyze <repo> --export <dir>` writes three files:
+  `commits.ndjson` (one row per commit: sha/ts/origin(11-class)/
+  actor(canonical_id|null)/bot/is_merge/prod+test path counts/
+  cochange(bool|null — only production-changing commits carry a value)/
+  subject_class(fix|revert|other)), `actors.json` (canonical_id,
+  attribution_state, commits, active_days, first_ts/last_ts, core_period —
+  attribution and engagement only; per-actor quality metrics are permanently
+  excluded), and `export-meta.json` (schema `tep-export-v1`,
+  `config_digest`, definition versions, `analysis_scope`)
+- **Raw emails and raw author strings are never exported.** The no-`@`
+  property is enforced by tests
+- report and export share `prepare_inputs()` — a single implementation
+- Narrative numbers must come from report.json aggregates (ruling 5);
+  consumers recompute from commits.ndjson and abort ingestion on mismatch
+- `config_digest`: sha256 over the canonical JSON of the normalized identity
+  + definition versions + flags (full spec in the Japanese original)
+- `--export` never changes the stdout contract

grift_cli-0.5.6/docs/en/identity-schema.md

@@ -0,0 +1,32 @@
+# Identity file schema (identity-v1)
+
+Japanese original: [identity-schema.md](../identity-schema.md).
+
+| Field | Meaning |
+|---|---|
+| `canonical_id` | Durable pseudonymous id (never an email; no personal scorecards) |
+| `emails` | Raw git author emails mapped to this actor (local file only) |
+| `github_login` | Optional |
+| `attribution_state` | `verified` / `claimed` / `inferred` / `unresolved` / `external` / `bot` |
+
+## canonical_id validity (F-P10-3, version-managed)
+
+`canonical_id` must **fullmatch**:
+
+```
+^[a-z0-9][a-z0-9._-]{0,63}$
+```
+
+Lowercase alphanumeric start; then lowercase alphanumerics and `.` `_` `-`;
+length 1–64. **`@`, whitespace, control characters, uppercase, and multi-byte
+characters are invalid** (this pattern — not a blocklist — is what makes the
+export's no-`@` contract data-independent). Duplicates are rejected. The
+regex must match the implementation (`tep_core.identity.CANONICAL_ID_PATTERN`)
+verbatim; `tests/test_identity_validation.py` enforces the doc↔code parity.
+
+| Version | Pattern | Introduced |
+|---|---|---|
+| 1 | `^[a-z0-9][a-z0-9._-]{0,63}$` | 2026-08-22 (F-P10-3) |
+
+`tenant.email_patterns` are regexes; a match counts as tenant membership
+(state `inferred`). Empty/missing file: `pending_attribution = true`.

grift_cli-0.5.6/docs/en/metrics-guide.md

@@ -0,0 +1,175 @@
+# How to read grift reports — a plain-language metric guide (v2, 2026-08-23)
+
+Japanese original: [metrics-guide.md](../metrics-guide.md).
+
+**Audience**: non-engineers. **Principle restated**: every number is a record
+of observed fact, not a grade. Standalone pass/fail decisions and
+person-to-person comparisons are non-compliant with the TEP norms
+([en/norms.md](norms.md)). What is not written (= not observed) did not
+"not happen." Reference values come from a 117-repository public corpus
+(v2026.11, repo scope) and field measurements (requests / flask / GitHub CLI
+/ chalk).
+
+---
+
+## Decile notation (the one definition used throughout)
+
+**Decile N = the band containing the top (100−10N)% of values.** Formally,
+with corpus boundaries d₁…d₁₀: **band N = (d(N−1), dN]** (left-open,
+right-closed). Decile 5 = the middle band (50–60% from the bottom); decile
+10 = the top 10% band.
+
+- Example: click's test co-change 0.3178 falls in 0.2894 < 0.3178 ≤ 0.4871,
+  hence **decile 9** (the top-20% band)
+- Every band mention in this guide carries a plain-language translation
+  (e.g. "decile 7 (the top 30–40% band)")
+
+---
+
+## Full observation inventory (what appears in a report)
+
+grift reports consist of three layers. **A layer is a promise about how the
+number may be used.**
+
+### Attribution layer (origin, 11 classes) — "whose/what work is this commit?"
+
+| Class | One-line definition |
+|---|---|
+| tenant_unique | Original (non-merge) commits by identity-matched workers |
+| tenant_merge_or_sync | Tenant merge commits (PR flow; repos without upstream lineage) |
+| upstream_sync | Tenant merges that pull upstream (fork-lineage repos) |
+| inherited_upstream | Other people's (upstream-side) commits under fork lineage |
+| tenant_derivative | Judged only with a parent repo provided: work derived from it |
+| external_upstream_contribution | Judged only with a parent repo: work flowing back upstream |
+| template_inherited | Judged only with a template provided: template-derived parts |
+| generated_or_vendor | Commits touching only generated/vendored files |
+| bot | Automation-account commits |
+| ambiguous_origin | Author information missing — cannot be judged |
+| unresolved | Human commits not matching the identity (no names shown) |
+
+### Evidence layer — "changes accompanied by verification"
+
+| Observation | One-line definition |
+|---|---|
+| test co-change | Share of production changes whose same commit also changed tests |
+| survival | Share of lines still present after 180 days (`--survival` only) |
+
+### Observational layer — "tendency records (not evidence claims)"
+
+| Observation | One-line definition |
+|---|---|
+| corrective rework | Share of commits quickly redoing recently touched paths with fix/revert subjects |
+| path retouch | Share of commits re-touching the same files within 21 days |
+| revert_rate | Share of revert commits |
+
+### Context layer (context profile) — "what kind of repo is this"
+
+| Observation | One-line definition |
+|---|---|
+| collaboration_class | solo / small_team / community |
+| lifecycle_stage (+ active_days_180d / days_since_last_human_commit) | Activity density (the class is always printed with the two raw observations) |
+| resolved_human_actors / top_actor_share | Contributor count (upstream/bot excluded) and top-author share (no names) |
+| pr_flow_share | Share of merge (PR) commits |
+| scale / repo_age_days | Size (commits, span, top dirs, tags) and repo age |
+| actor_turnover | Yearly join/leave counts |
+| release_cadence | Tag frequency (releases/year) |
+| conventional_commit_share / issue_link_density | Conventional subjects (`feat:` etc.) / `#N` references |
+| language_composition / dependency_manifests / monorepo_markers | Language mix / dependency files / monorepo signs |
+| test_file_ratio / docs_share | Test/docs path touches |
+
+---
+
+## Evidence: test co-change
+
+**Unit**: ratio (0.0–1.0). **Population**: commits that changed production code.
+
+**What it measures**: of the commits that changed production code, the share
+that changed tests in the same commit — read from git history, not
+self-reports. Docs-only changes are excluded from the population.
+
+**How to read it**:
+
+- **High (≥ 0.233 = above the top edge of decile 7, i.e. the top-30% band)**:
+  the habit of accompanying production changes with tests is established
+- **Middle (0.109–0.161 = decile 5, the middle band)**: where most public
+  repositories sit
+- **Low / 0**: no test substrate, or tests not synchronized. **Not "bad"** —
+  docs-centric repos, design phases, and tooling legitimately land low
+- **not observed**: no test framework or directory detected. Cannot be
+  measured ≠ zero
+
+**Corpus bands (v2026.11, n=68, repo scope)**: decile 5 = (0.1089, 0.1613] ·
+decile 7 = (0.1613, 0.233] · decile 10 = (0.4871, 0.6948]. Field examples:
+GitHub CLI 0.37 · flask 0.20 · chalk 0.18 · requests 0.12 (2026-08).
+
+---
+
+## Observational: corrective rework & path retouch
+
+**Not evidence claims.** Corrective rework = share of commits redoing paths
+touched within 21 days with fix/revert subjects; path retouch = share
+re-touching the same files in 21 days. Subject-convention dependent —
+unsuitable for comparisons or pass/fail. Use high values as interview
+questions ("what happened in this period?"), not verdicts. Low values do not
+prove few bugs.
+
+**Corpus bands (v2026.11, n=101)**: decile 5 = (0.0683, 0.0972] · deciles
+9–10 = (0.1904, 0.4818].
+
+---
+
+## Survival (optional)
+
+`--survival` only (τ=180d). 1.0-near values = lines mostly still present.
+Reference distribution planned for v2027 — no high/low judgment provided yet.
+
+---
+
+## Context: collaboration & lifecycle
+
+- **collaboration_class**: solo (top author ≥ 90%) / small_team (≤5 people) /
+  community. Examples: requests 802 / flask 870 / GitHub CLI 722 → community
+- **lifecycle_stage**: always read with the two raw observations.
+  active = active_days_180d ≥ 12 / maintained = 3–11 / dormant = ≤ 2.
+  Density does **not** distinguish development from maintenance (a
+  high-density operational repo is legitimately active); density was chosen
+  over recency so bulk housekeeping cannot fake freshness
+
+---
+
+## Related tools (lineage & design differences — no superiority claims)
+
+| Name | In one line |
+|---|---|
+| **machuz/eis (Engineering Impact Signal)** | TEP's survival observation adopts EIS-style blame sampling as its lineage; the designs differ in fact: EIS reduces 7 axes with fixed weights to a 0-100 composite score with type labels and team side-by-sides; TEP forbids composite scores, weighting, and grade vocabulary by design (the six prohibitions / norms), and secures trust via discriminant validation, `grift verify`, and no-transmission |
+| DORA | Four delivery-performance metrics and benchmarks; TEP's norm-style usage guidelines follow the DORA pattern |
+| SPACE | A satisfaction/performance/activity/communication/efficiency/flow framework; agrees with TEP that productivity is not one number |
+| GitClear | Commercial code-health (technical debt, rework) research and SaaS; overlapping observations, but TEP never emits person-comparison tables and the CLI transmits nothing |
+| MSR | The academic field of mining software repositories; much of TEP's git-derived tooling applies its findings |
+| bus factor | The degree to which a project stalls when key developers leave; top_actor_share / collaboration_class answer this question observationally |
+
+---
+
+## Common misreadings
+
+| Misreading | Correct reading |
+|---|---|
+| "Low co-change = can't write tests" | Docs/design/tooling work lands low legitimately; reading absence as inability is non-compliant (norm 1) |
+| "High corrective rework = buggy" | Subject-convention dependent; neither a bug count nor quality |
+| "dormant = abandoned" | Mature repos with long maintenance cycles exist; reading gaps negatively is non-compliant |
+| "Decile 9 > decile 5 person" | Deciles are positions (bands) among same-scope repos; standalone cutoffs and comparison tables are non-compliant (norm 2) |
+| "No decile = hiding something" | Suppressed by rule when population <20 or corpus n<30 |
+
+---
+
+## When a number is absent (not observed)
+
+`no_test_framework_or_directory` (no test substrate) ·
+`insufficient_population` (<20 commits) · `pending_attribution` (identity
+unset) · `scope_is_tenant` (distributions are repo-scope only) ·
+`survival_scan_disabled` (`--survival` only).
+
+---
+
+*This is the plain-language companion of docs/report-schema.md (the
+authoritative, machine-validated reference).*