thth 2.5.0__tar.gz → 2.7.0__tar.gz

{thth-2.5.0 → thth-2.7.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: thth
-Version: 2.5.0
+Version: 2.7.0
 Summary: 承認済みの投稿だけを投げる門（Threads・Bluesky・Mastodon）。人が承認し、記録は利用者の git に残る。
 Project-URL: Homepage, https://thth.me
 Project-URL: Source, https://github.com/aokings/thth
@@ -33,7 +33,7 @@ watchtower（`~/Developer/watchtower`）の隣に同じ流儀で並べる。watc
 
 ## 版
 
-このソースの版は **2.5.0**。現在の設計は [自分の泉](docs/設計_自分の泉_2026-09-16.md)、導入は [自分の Meta アプリで動かす](docs/導入_自分のMetaアプリで動かす.md)、文書の索引は [docs/README.md](docs/README.md)。
+このソースの版は **2.7.0**。現在の設計は [自分の泉](docs/設計_自分の泉_2026-09-16.md)、導入は [自分の Meta アプリで動かす](docs/導入_自分のMetaアプリで動かす.md)、文書の索引は [docs/README.md](docs/README.md)。
 
 版の正本は `thth/VERSION`（`server.json` はテストで一致を強制）。開発中の `main` には未配布の変更も含まれます。インストール済みの版は `thth --version`、VM の版と revision は `thth board` で確認してください。
 
@@ -45,7 +45,7 @@ watchtower（`~/Developer/watchtower`）の隣に同じ流儀で並べる。watc
 
 ## 実装と運用の状態（2026-09-18）
 
-2.5.0 の実装 revision の全件テストは **2690 件**（`python -m pytest tests/ -q -n auto -p no:cacheprovider`・1 skip・rc=0）。開発中の変更の検証は対象 revision の CI を参照してください。
+2.7.0 の実装 revision `f331e8c` の全件テストは **2743 件**（`python -m pytest tests/ -q -n auto -p no:cacheprovider`・1 skip・rc=0）。開発中の変更の検証は対象 revision の CI を参照してください。
 
 **媒体**: Threads（稼働）・Bluesky・Mastodon（同席用の台帳あり・未稼働）。
 
@@ -59,6 +59,20 @@ timer（systemd・`thth systemd` で生成）は10分ごとに実行し、投稿
 
 **トピックの棚**（`thth topics`）: 観測者ごとに並ぶ・打ち消し `retract-note`・`history`。
 
+### v2.7.0 の変更
+
+- 指標採集に720時間（30日）を追加し、対象窓を刻みから38日と計算。正常時は指標6回、返信は従来の5回です。返信の採集窓は独立して維持します。
+- 新しい観測に、採取時刻以前48時間以内で最も新しい日次フォロワー数を文脈として添えます。`followers_count_at` と `staleness_hours` を含み、追加APIや過去行の後補完はありません。該当しない場合は `context_reason` に理由を残します。
+- Threads・Bluesky・Mastodonの刻みを揃え、媒体にないviewsはnullのまま維持。[合成検証と限界](docs/検証_v31_C実装_2026-09-18.md)。実SNS APIでの30日後採集は未確認です。
+- C3・C4は実装保留。[独立規約照合報告](docs/照合_MetaPlatformTerms_枝と語の観測_2026-09-18_Codex.md)の未確認事項とApp Reviewの結果を踏まえて別途判断します。
+
+### v2.6.0 の変更
+
+- 既存台帳から刻み別の値、中央値の IQR・min・max、枝別の帰結、採集成否と鮮度、層別比較、採集時刻を上限とした枝の形を追加。既存キーは維持し `schema_version=1` のままです。
+- `cannot_say_details` にコードと既存の説明文を併記。施策レポートには未成熟投稿が24時間に達する予定を添え、採集保証とは分けます。
+- `handoff-report --since-last-read` で保存済みsnapshotとの差分。cursor保存は CLI の `--mark-read --by 名前` を明示したときだけ。MCP・HTTPは読むだけです。
+- API呼出を増やさない分析の改良です。VM5accountの読取検証と限界は [検証記録](docs/検証_v31_A実装_2026-09-18.md) に記載。Bluesky/Mastodon の非空実台帳と実採用施策の検証は未確認です。
+
 ### v2.5.0 の変更
 
 - 読み取り専用のレポート 3 本: `analytics-report`（活動のスナップショット・`--compare-previous` で隣接期間の比較）・`handoff-report`（ローカル運用記録の引継ぎ）・`study-report`（施策の宣言と本人の観測の結合）。MCP に `analytics_report`・`operations_handoff`・`study_report`。数値は期間・母数・欠測・根拠を連れて歩き、因果や推奨は出しません。

{thth-2.5.0 → thth-2.7.0}/README.md

@@ -9,7 +9,7 @@ watchtower（`~/Developer/watchtower`）の隣に同じ流儀で並べる。watc
 
 ## 版
 
-このソースの版は **2.5.0**。現在の設計は [自分の泉](docs/設計_自分の泉_2026-09-16.md)、導入は [自分の Meta アプリで動かす](docs/導入_自分のMetaアプリで動かす.md)、文書の索引は [docs/README.md](docs/README.md)。
+このソースの版は **2.7.0**。現在の設計は [自分の泉](docs/設計_自分の泉_2026-09-16.md)、導入は [自分の Meta アプリで動かす](docs/導入_自分のMetaアプリで動かす.md)、文書の索引は [docs/README.md](docs/README.md)。
 
 版の正本は `thth/VERSION`（`server.json` はテストで一致を強制）。開発中の `main` には未配布の変更も含まれます。インストール済みの版は `thth --version`、VM の版と revision は `thth board` で確認してください。
 
@@ -21,7 +21,7 @@ watchtower（`~/Developer/watchtower`）の隣に同じ流儀で並べる。watc
 
 ## 実装と運用の状態（2026-09-18）
 
-2.5.0 の実装 revision の全件テストは **2690 件**（`python -m pytest tests/ -q -n auto -p no:cacheprovider`・1 skip・rc=0）。開発中の変更の検証は対象 revision の CI を参照してください。
+2.7.0 の実装 revision `f331e8c` の全件テストは **2743 件**（`python -m pytest tests/ -q -n auto -p no:cacheprovider`・1 skip・rc=0）。開発中の変更の検証は対象 revision の CI を参照してください。
 
 **媒体**: Threads（稼働）・Bluesky・Mastodon（同席用の台帳あり・未稼働）。
 
@@ -35,6 +35,20 @@ timer（systemd・`thth systemd` で生成）は10分ごとに実行し、投稿
 
 **トピックの棚**（`thth topics`）: 観測者ごとに並ぶ・打ち消し `retract-note`・`history`。
 
+### v2.7.0 の変更
+
+- 指標採集に720時間（30日）を追加し、対象窓を刻みから38日と計算。正常時は指標6回、返信は従来の5回です。返信の採集窓は独立して維持します。
+- 新しい観測に、採取時刻以前48時間以内で最も新しい日次フォロワー数を文脈として添えます。`followers_count_at` と `staleness_hours` を含み、追加APIや過去行の後補完はありません。該当しない場合は `context_reason` に理由を残します。
+- Threads・Bluesky・Mastodonの刻みを揃え、媒体にないviewsはnullのまま維持。[合成検証と限界](docs/検証_v31_C実装_2026-09-18.md)。実SNS APIでの30日後採集は未確認です。
+- C3・C4は実装保留。[独立規約照合報告](docs/照合_MetaPlatformTerms_枝と語の観測_2026-09-18_Codex.md)の未確認事項とApp Reviewの結果を踏まえて別途判断します。
+
+### v2.6.0 の変更
+
+- 既存台帳から刻み別の値、中央値の IQR・min・max、枝別の帰結、採集成否と鮮度、層別比較、採集時刻を上限とした枝の形を追加。既存キーは維持し `schema_version=1` のままです。
+- `cannot_say_details` にコードと既存の説明文を併記。施策レポートには未成熟投稿が24時間に達する予定を添え、採集保証とは分けます。
+- `handoff-report --since-last-read` で保存済みsnapshotとの差分。cursor保存は CLI の `--mark-read --by 名前` を明示したときだけ。MCP・HTTPは読むだけです。
+- API呼出を増やさない分析の改良です。VM5accountの読取検証と限界は [検証記録](docs/検証_v31_A実装_2026-09-18.md) に記載。Bluesky/Mastodon の非空実台帳と実採用施策の検証は未確認です。
+
 ### v2.5.0 の変更
 
 - 読み取り専用のレポート 3 本: `analytics-report`（活動のスナップショット・`--compare-previous` で隣接期間の比較）・`handoff-report`（ローカル運用記録の引継ぎ）・`study-report`（施策の宣言と本人の観測の結合）。MCP に `analytics_report`・`operations_handoff`・`study_report`。数値は期間・母数・欠測・根拠を連れて歩き、因果や推奨は出しません。

{thth-2.5.0 → thth-2.7.0}/mcp/server.py

@@ -143,7 +143,8 @@ TOOLS = [
         "name": "operations_handoff",
         "description": "ローカル運用記録の状態・承認待ち・inflight・通知未処理を根拠と鮮度の制約付きで返す。同期・承認・再送はしない",
         "inputSchema": {"type": "object", "properties": {
-            "account": {"type": "string"}, "project": {"type": "string"}}},
+            "account": {"type": "string"}, "project": {"type": "string"},
+            "since_last_read": {"type": "boolean"}}},
     },
     {
         "name": "study_report",
@@ -159,6 +160,7 @@ TOOLS = [
         "inputSchema": {
             "type": "object",
             "properties": {
+                "by": {"type": "string", "enum": ["kind", "hour_band", "topic"]},
                 "compare_previous": {"type": "boolean", "description": "直前の同じ日数との比較（既定false）"},
                 "account": {"type": "string"},
                 "project": {"type": "string"},
@@ -397,6 +399,11 @@ def validate_arguments(name: str, arguments) -> dict:
         raise ToolInputError(
             f"{name}: arguments は object で渡してください"
             f"（受け取った: {type(arguments).__name__}）")
+    if name == "analytics_report" and "by" in arguments:
+        if arguments["by"] not in ("kind", "hour_band", "topic") or arguments.get("compare_previous") is not True:
+            raise ToolInputError("analytics_report: by は compare_previous=true と kind/hour_band/topic が必要です")
+    if name == "operations_handoff" and "since_last_read" in arguments and type(arguments["since_last_read"]) is not bool:
+        raise ToolInputError("operations_handoff: since_last_read は boolean です")
     props = schema.get("properties") or {}
     for key in schema.get("required") or []:
         if arguments.get(key) is None:
@@ -533,6 +540,10 @@ def call_tool(name: str, arguments: dict | None) -> dict:
         text = proc.stdout
     elif name == "operations_handoff":
         args = ["handoff-report"]
+        if "mark_read" in arguments or "by" in arguments:
+            raise ValueError("operations_handoff is read-only")
+        if arguments.get("since_last_read"):
+            args.append("--since-last-read")
         if arguments.get("project"):
             args += ["--project", arguments["project"]]
         else:
@@ -558,6 +569,8 @@ def call_tool(name: str, arguments: dict | None) -> dict:
                 args += ["--" + key.replace("_", "-"), str(arguments[key])]
         if arguments.get("compare_previous"):
             args.append("--compare-previous")
+        if arguments.get("by"):
+            args.extend(["--by", arguments["by"]])
         args.append("--json")
         proc = run_cli(args)
         text = proc.stdout

thth-2.7.0/tests/test_27_acceptance_fixes.py

@@ -0,0 +1,45 @@
+import datetime as dt
+import json
+from pathlib import Path
+from thth import collection_context, jst
+
+NOW=jst.parse('2026-10-01T01:00:00+09:00')
+
+def daily(value,age,**extra):
+    return {'account':'a','date':'ignored','collected_at':jst.iso(NOW-dt.timedelta(hours=age)),
+            'metrics':{'followers_count':value},**extra}
+
+def test_recent_daily_across_month_and_date_ignored(tmp_path):
+    (tmp_path/'a-2026-09.ndjson').write_text(json.dumps(daily(120,2,date='2026-09-29')))
+    value=collection_context.followers('a',tmp_path,NOW)
+    assert value=={'followers_count':120,'followers_count_at':'2026-09-30T23:00:00+09:00',
+                  'staleness_hours':2.0,'source':'insights_account'}
+    (tmp_path/'a-2026-10.ndjson').write_text(json.dumps(daily(121,1.234)))
+    value=collection_context.followers('a',tmp_path,NOW)
+    assert value['followers_count']==121 and value['staleness_hours']==1.23
+
+def test_recent_daily_inclusive_48h_and_no_old_fallback(tmp_path):
+    p=tmp_path/'a-2026-09.ndjson'
+    p.write_text(json.dumps(daily(0,48)))
+    assert collection_context.followers('a',tmp_path,NOW)['followers_count']==0
+    p.write_text(json.dumps(daily(2,49)))
+    assert collection_context.followers('a',tmp_path,NOW) is None
+    p.write_text('\n'.join(map(json.dumps,[daily(2,3),daily(True,1)])))
+    assert collection_context.followers('a',tmp_path,NOW) is None
+
+def test_recent_daily_collect_reason_and_no_retrofit(tmp_path,isolated_account_factory,monkeypatch):
+    from tests.test_v31_collection import setup_route,Adapter,POSTED,daily_path,daily_row
+    from thth import collect,accounts
+    a=setup_route(tmp_path,isolated_account_factory,monkeypatch,'sent')
+    now=POSTED+dt.timedelta(hours=1)
+    p=daily_path(a)
+    p.write_text(json.dumps(daily_row(a,9,collected_at=jst.iso(now-dt.timedelta(hours=49)))))
+    collect.collect_once(a['name'],adapter=Adapter(),now=now,log=lambda _:None)
+    out=Path(accounts.data_dirs(a,a['name'])['insights_posts'])/'P.ndjson'
+    before=out.read_bytes()
+    assert json.loads(before)['context'] is None
+    assert json.loads(before)['context_reason']=='no_recent_daily_record'
+    p.write_text(json.dumps(daily_row(a,0,collected_at=jst.iso(now))))
+    collect.collect_once(a['name'],adapter=Adapter(),now=POSTED+dt.timedelta(hours=6),log=lambda _:None)
+    assert out.read_bytes().startswith(before)
+    assert json.loads(out.read_text().splitlines()[-1])['context']['followers_count']==0

{thth-2.5.0 → thth-2.7.0}/tests/test_analytics_comparison.py

@@ -85,7 +85,7 @@ def test_invalid_metric_does_not_become_zero(isolated_account_factory, value):
     seed(a, "p", NOW - datetime.timedelta(days=3), value=value)
     p = report(a, min_n=1)["by_account"][a["name"]]["posts"]["current"]
     assert p["n_eligible"] == 1  # time-eligible does not imply every metric exists
-    assert p["metrics"]["views"] == {"n_total": 1, "n_eligible": 0, "n_missing": 1, "median": None}
+    assert p["metrics"]["views"] == {"n_total": 1, "n_eligible": 0, "n_missing": 1, "median": None, "iqr": None, "min": None, "max": None, "spread_reason": "below_min_n"}
     json.dumps(p, allow_nan=False)
 
 

{thth-2.5.0 → thth-2.7.0}/tests/test_analytics_report.py

@@ -32,9 +32,19 @@ def test_boundaries_evidence_missing_and_no_writes(isolated_account_factory):
     node = result["by_account"][account["name"]]
     original = after_cli.answer(account["name"], now=NOW, window_days=7)
     for key in ("posts", "engagements", "comparable", "cannot_say"):
-        assert node[key] == original[key]
+        def legacy_projection(actual, original):
+            if isinstance(original, dict):
+                return {k: legacy_projection(actual[k], v) for k, v in original.items()}
+            if isinstance(original, list):
+                assert len(actual) == len(original)
+                return [legacy_projection(a, b) for a, b in zip(actual, original)]
+            return actual
+        actual = node[key]
+        if key == "cannot_say":
+            actual = [x for x in actual if not x.startswith("collection_records_")]
+        assert legacy_projection(actual, original[key]) == original[key]
     assert {p["post_id"] for p in node["posts"]["by_post"]} == {"start", "end"}
-    assert node["posts"]["views_24h"] == {"median": None, "n": 1}
+    assert node["posts"]["views_24h"] == {"median": None, "n": 1, "iqr": None, "min": None, "max": None, "spread_reason": "below_min_n"}
     end = next(p for p in node["posts"]["by_post"] if p["post_id"] == "end")
     assert end["views_24h"] is None and end["covered"] is False
     assert node["engagements"]["by_branch"][0]["post_id"] == "reply"
@@ -46,7 +56,16 @@ def test_boundaries_evidence_missing_and_no_writes(isolated_account_factory):
     assert result["generated_at"] == "2026-09-17T08:00:00+09:00"
     text = json.dumps(result)
     assert "PRIVATE BODY" not in text
-    assert '"username"' not in text and '"text"' not in text
+    assert '"username"' not in text
+    def assert_text_only_in_details(value, details=False):
+        if isinstance(value, dict):
+            assert "text" not in value or details
+            for key, child in value.items():
+                assert_text_only_in_details(child, key == "cannot_say_details")
+        elif isinstance(value, list):
+            for child in value:
+                assert_text_only_in_details(child, details)
+    assert_text_only_in_details(result)
     markdown = analytics_report.render_markdown(result)
     recovered = "\n".join(line[4:] for line in markdown.splitlines() if line.startswith("    "))
     assert json.loads(recovered) == result

{thth-2.5.0 → thth-2.7.0}/tests/test_collect_sent.py

@@ -220,7 +220,8 @@ def test_d_従来の台帳は置き場も行の中身も変わらない(tmp_path
     assert set(rows[0]) == {
         "post_id", "file", "source", "account", "medium", "topic", "reply_to",
         "text_length", "has_link", "collected_at", "posted_at", "age_hours",
-        "marks", "metrics"}, sorted(rows[0])
+        "marks", "metrics", "context", "context_reason"}, sorted(rows[0])
+    assert rows[0]["context"] is None
     assert rows[0]["source"] == "queue"
     assert rows[0]["file"] == "a.md"
     assert rows[0]["age_hours"] == 2.0

{thth-2.5.0 → thth-2.7.0}/tests/test_study_report.py

@@ -149,7 +149,8 @@ def test_broken_source_immature_and_sparse_remain_unknown(isolated_account_facto
     assert r["observations"]["changed"]["n_immature"] == 1
     assert r["observations"]["baseline"]["metrics"]["views"]["n_eligible"] == 1
     assert all(v["absolute_median_change"] is None for v in r["comparison"].values())
-    assert len(r["cannot_say"]) == 2
+    assert len(r["cannot_say"]) == 3
+    assert r["cannot_say"][-1] == "forecast_assumes_collection_runs"
 
 
 def test_invalid_cli_is_error_without_input_disclosure(isolated_account_factory, capsys):

thth-2.7.0/tests/test_v31_analysis.py

@@ -0,0 +1,219 @@
+import datetime as dt
+from thth import analytics_comparison as c, jst
+
+NOW = jst.parse('2026-09-18T12:00:00+09:00')
+POSTED = NOW - dt.timedelta(days=3)
+
+def row(mark, age, value=0):
+    return {'posted_at': jst.iso(POSTED), 'collected_at': jst.iso(POSTED + dt.timedelta(hours=age)),
+            'marks': [mark], 'metrics': {'likes': value}}
+
+def test_a1_strict_marks():
+    post = {'rows': [row(1, 1.25, 8), row(1, 1, 0), row(6, 7.5), row(24, 25, 9)]}
+    result = c._marks_population([('p', POSTED, post)], POSTED, NOW, NOW, 1)
+    assert result['by_mark']['1']['metrics']['likes']['median'] == 0
+    assert result['by_mark']['1']['metrics']['views']['median'] is None
+    assert result['marks_by_post'][0]['marks']['6'] is None
+    assert result['by_mark']['24']['metrics']['likes']['median'] == 9
+    assert result['by_mark']['168']['n_eligible'] == 0
+    assert c._marks_population([('p', POSTED, post)], POSTED, NOW, NOW, 2)['by_mark']['1']['metrics']['likes']['median'] is None
+
+def test_a1_reject_collapsed_future_conflicting():
+    rows = [row(1, 1), row(1, 2), row(1, 1)]
+    rows[0]['marks_collapsed'] = True
+    rows[2]['posted_at'] = jst.iso(POSTED-dt.timedelta(hours=1))
+    assert c._observation({'rows': rows}, POSTED, NOW, 1)[0] is None
+    assert c._observation({'rows': [row(168, 168)]}, POSTED, NOW, 168)[0] is None
+
+def test_a5_spread_gate_and_odd_center():
+    assert c._spread([1, 2, 3, 4], 5) == {'iqr': None, 'min': None, 'max': None, 'spread_reason': 'below_min_n'}
+    assert c._spread([0], 1) == {'iqr': None, 'min': 0, 'max': 0, 'spread_reason': 'too_few_for_quartiles'}
+    assert c._spread([1, 2, 3], 1)['iqr'] is None
+    assert c._spread([1, 2, 50, 80, 100], 1)['iqr'] == 88.5
+    assert c._spread([1, 2, 3, 4], 1)['iqr'] == 2
+
+def test_a8_details_preserve_order_text():
+    from thth.report_details import attach
+    messages = ['timer_health_unknown', '自分の投稿の 24h views が欠測: 2 本', '未知の説明']
+    result = attach({'cannot_say': list(messages)})
+    assert result['cannot_say'] == messages
+    assert result['cannot_say_details'] == [
+        {'code': 'timer_health_unknown', 'text': messages[0]},
+        {'code': 'post_views_missing', 'text': messages[1]},
+        {'code': 'report_limitation', 'text': messages[2]}]
+
+def test_a1_malformed_mark_types():
+    for mark in (1, 6, 24, 72, 168):
+        for bad in (True, False, str(mark), float(mark)):
+            value = row(mark, mark)
+            value['marks'] = [bad]
+            assert c._observation({'rows': [value]}, POSTED, NOW + dt.timedelta(days=10), mark)[0] is None
+
+def test_a2_reaction_identity_and_missing(monkeypatch):
+    from thth import analytics_threads as t
+    ledger = {'broken': [], 'unreadable_accounts': [], 'fetches': [], 'replies': []}
+    monkeypatch.setattr(t.replies, 'load', lambda *a, **k: ledger)
+    e = {'post_id': 'mine', 'author_key': '0123456789abcdef'}
+    assert t.reaction('a', 'threads', e, POSTED, NOW) == (None, None)
+    ledger['fetches'] = [{'post_id':'mine', 'collected_at': jst.iso(NOW)}]
+    assert t.reaction('a', 'threads', e, POSTED, NOW) == (False, None)
+    ledger['replies'] = [{'post_id':'mine', 'collected_at': jst.iso(NOW), 'timestamp': jst.iso(POSTED+dt.timedelta(hours=2)), 'own': False, 'replied_to': {'id':'mine'}, 'author_key': e['author_key'], 'text':'SECRET'}]
+    assert t.reaction('a', 'threads', e, POSTED, NOW) == (True, 2)
+    ledger['replies'][0]['replied_to'] = {'id': 'someone_else'}
+    assert t.reaction('a', 'threads', e, POSTED, NOW) == (False, None)
+
+def test_a3_skip_not_success(tmp_path, monkeypatch):
+    import json
+    from thth import collection_status as s
+    monkeypatch.setattr(s.accounts, 'state_dir_for', lambda name: tmp_path)
+    assert s.summarize('a', NOW)[0] is None
+    p = tmp_path/'runs-2026-09.ndjson'
+    rows = [{'account':'a', 'mode':'collect', 'action':'collect', 'run_id':'collect-'+jst.iso(POSTED), 'status':'ok','collected':0},
+            {'account':'a', 'mode':'collect', 'action':'collect', 'run_id':'collect-'+jst.iso(NOW), 'status':'ok','collected':None}]
+    p.write_text('\n'.join(map(json.dumps,rows)))
+    value, reasons = s.summarize('a', NOW)
+    assert value['last_success_at'] == jst.iso(POSTED)
+    assert value['last_attempt_ok'] is None
+    rows[-1]['collected'] = True
+    p.write_text('\n'.join(map(json.dumps,rows)))
+    assert s.summarize('a', NOW)[0]['last_attempt_ok'] is None
+
+def test_a6_unknown_stratum_reconciles(isolated_account_factory):
+    from tests.test_analytics_comparison import seed, report
+    from thth import analytics_report
+    a = isolated_account_factory()
+    seed(a, 'p', NOW-dt.timedelta(days=3), value=0)
+    result = analytics_report.answer(a['name'], now=NOW, compare_previous=True, by='kind', min_n=1)
+    group = result['by_account'][a['name']]['posts']
+    assert group['stratified']['strata']['unknown']['current']['n_total'] == 1
+    assert group['stratified']['reconciliation']['current'] == {'sum_n_total':1,'n_total':1}
+    assert group['stratified']['strata']['unknown']['current']['metrics']['views']['median'] == 0
+
+def test_a7_collected_time_bounds_history(monkeypatch):
+    from thth import analytics_shapes as s
+    early = POSTED+dt.timedelta(hours=23)
+    late = POSTED+dt.timedelta(hours=25)
+    ledger = {'broken': [], 'fetches':[{'post_id':'p', 'collected_at':jst.iso(early)}, {'post_id':'p', 'collected_at':jst.iso(late)}], 'replies':[
+        {'post_id':'p','id':'r','own':False,'username':'private','text':'SECRET','replied_to':{'id':'p'},
+         'timestamp':jst.iso(POSTED+dt.timedelta(hours=1)), 'collected_at':jst.iso(late)}]}
+    monkeypatch.setattr(s.replies,'load',lambda *a,**k:ledger)
+    value=s.shape_at('a','p',POSTED,NOW)
+    assert value['24']['replies_total']==0
+    assert value['72']['replies_total']==1
+    assert value['168'] is None
+    assert 'SECRET' not in str(value) and 'private' not in str(value)
+    ledger['fetches']=[]
+    assert s.shape_at('a','p',POSTED,NOW)['24'] is None
+
+def test_a9_forecast_only_immature():
+    from thth.study_report import _eligibility_forecast
+    late=NOW-dt.timedelta(hours=1)
+    value=_eligibility_forecast({'evidence':[
+        {'status':'immature','posted_at':jst.iso(NOW-dt.timedelta(hours=2))},
+        {'status':'immature','posted_at':jst.iso(late)},
+        {'status':'incomplete','posted_at':jst.iso(POSTED)}]})
+    assert value=={'posts_immature':2, 'all_eligible_at':jst.iso(late+dt.timedelta(hours=24)), 'basis':'posted_at_plus_24h'}
+    assert _eligibility_forecast({'evidence':[]}) is None
+
+def test_a4_explicit_cli_write_and_readonly_delta(isolated_account_factory, monkeypatch, capsys):
+    import json
+    from pathlib import Path
+    from types import SimpleNamespace
+    from thth import operations_handoff as h, handoff_cursor as cursor, accounts
+    from tests.test_operations_handoff import seed
+    a=isolated_account_factory()
+    name=a['name']
+    monkeypatch.setattr(h.jst,'now_jst',lambda:NOW)
+    seed(a,'draft.md','draft')
+    path=Path(accounts.state_dir_for(name))/'handoff_cursor.json'
+    args=SimpleNamespace(account=name,project=None,json=True,mark_read=False,by=None,since_last_read=True)
+    assert h.cmd_handoff_report(args)==0
+    assert not path.exists()
+    args.mark_read=True
+    assert h.cmd_handoff_report(args)==2
+    assert not path.exists()
+    args.by='reader'
+    assert h.cmd_handoff_report(args)==0
+    saved=path.read_bytes()
+    node=h.answer(name,now=NOW,since_last_read=True)['by_account'][name]
+    assert node['changes_since']['changes']==[]
+    assert 'no_previous_session_cursor' not in node['cannot_say']
+    seed(a,'draft2.md','draft')
+    node=h.answer(name,now=NOW,since_last_read=True)['by_account'][name]
+    assert next(r for r in node['changes_since']['changes'] if r['field']=='queue_counts.draft')['delta']==1
+    assert path.read_bytes()==saved
+    path.unlink()
+    path.symlink_to(path.parent/'outside')
+    assert cursor.read(name,NOW)[1]=='cursor_unreadable'
+    args.mark_read=True
+    assert h.cmd_handoff_report(args)==2
+    assert not (path.parent/'outside').exists()
+
+def test_a4_unknown_counts_not_zero():
+    from thth.handoff_cursor import changes
+    import copy
+    base={'queue_counts':dict.fromkeys(['draft','approved_waiting','overdue','malformed','unattributed_malformed']),
+          'inflight':{'present':None,'since':None},'notification_last_event_id':None,
+          'notification_recorded_state':None,'run_last_attempt_at':None,'run_recorded_state':None,
+          'last_post_observed_at':None,'sent_count':None}
+    current=copy.deepcopy(base);current['queue_counts']['draft']=0
+    assert changes(base,current)==[{'field':'queue_counts.draft','previous':None,'current':0,'delta':None}]
+
+def test_a4_ancestor_swap_does_not_redirect(tmp_path, monkeypatch):
+    import os
+    from thth import handoff_cursor as cursor
+    from pathlib import Path
+    state=tmp_path/'state';state.mkdir()
+    (state/'a').mkdir()
+    outside=tmp_path/'outside';outside.mkdir();(outside/'a').mkdir()
+    original=tmp_path/'original'
+    monkeypatch.setattr(cursor.accounts,'state_dir_for',lambda name:state/name)
+    real_open=os.open
+    swapped=False
+    def race(path, flags, *args, **kwargs):
+        nonlocal swapped
+        # The state descriptor must already be held when its child is opened.
+        if path=='a' and not swapped:
+            state.rename(original)
+            state.symlink_to(outside,target_is_directory=True)
+            swapped=True
+        return real_open(path,flags,*args,**kwargs)
+    monkeypatch.setattr(cursor.os,'open',race)
+    fd=cursor._directory('a',create=True)
+    try:
+        child=real_open('proof',os.O_CREAT|os.O_WRONLY,0o600,dir_fd=fd)
+        os.close(child)
+    finally:os.close(fd)
+    assert swapped
+    assert (original/'a'/'proof').exists()
+    assert not (outside/'a'/'proof').exists()
+
+
+def test_a4_read_mode_limitations(isolated_account_factory):
+    from thth import operations_handoff as h
+    name=isolated_account_factory()['name']
+    assert h.answer(name,now=NOW)['limitations'][0]=='ローカル保存記録の現在の読み取り。前回セッション以降の差分ではない'
+    assert h.answer(name,now=NOW,since_last_read=True)['limitations'][0]=='保存済みsnapshotと現在の値の差分。間に起きた全イベントを復元するものではない'
+
+def test_a4_ancestor_swap_before_open_is_rejected(tmp_path, monkeypatch):
+    import os
+    import pytest
+    from thth import handoff_cursor as cursor
+    state=tmp_path/'state';state.mkdir()
+    outside=tmp_path/'outside';outside.mkdir();(outside/'a').mkdir()
+    monkeypatch.setattr(cursor.accounts,'state_dir_for',lambda name:state/name)
+    real_open=os.open
+    swapped=False
+    def race(path, flags, *args, **kwargs):
+        nonlocal swapped
+        if path=='state' and not swapped:
+            state.rename(tmp_path/'original')
+            state.symlink_to(outside,target_is_directory=True)
+            swapped=True
+        return real_open(path,flags,*args,**kwargs)
+    monkeypatch.setattr(cursor.os,'open',race)
+    with pytest.raises(OSError):
+        fd=cursor._directory('a',create=True)
+        os.close(fd)
+    assert swapped
+    assert list((outside/'a').iterdir())==[]