mixdog 0.9.18 → 0.9.20

package/scripts/bench/round-r24-shapefix.json

@@ -1,677 +0,0 @@
-{
-  "round": "24",
-  "runner": "lead",
-  "opts": {
-    "provider": "openai-oauth",
-    "model": "gpt-5.5",
-    "effort": null,
-    "fast": false,
-    "env": {
-      "MIXDOG_DATA_DIR": "C:\\Users\\tempe\\.mixdog\\bench-data"
-    }
-  },
-  "tasks": 9,
-  "completed": 9,
-  "completion_rate": 100,
-  "sessions": [
-    "sess_60408_1783063353252_1783063353494_ffb01c576742694d462d621830cf0e1c",
-    "sess_4988_1783063353262_1783063353504_2e497c9ba1e6a87f33f6b7f1ee20672a",
-    "sess_39880_1783063353281_1783063353525_1d1b13112d7adc7b0cf1e8742ffe6a5f",
-    "sess_33512_1783063371729_1783063371967_9a4241862453a1eea2e17976f6798c12",
-    "sess_46584_1783063434132_1783063434375_2f5bc8a596d80965dc2b007018f607cd",
-    "sess_8156_1783063464105_1783063464349_363e58bf8ffe09e0bd16e479ab77bef3",
-    "sess_50160_1783063467999_1783063468251_274caf327774153c94df94f9ffb0a9b9",
-    "sess_35916_1783063487207_1783063487441_71e4fa81bf39a50301d9f3fe5118e29f",
-    "sess_59820_1783063534589_1783063534826_5b1388cb0dce5fcf4952797e321186d1"
-  ],
-  "results": [
-    {
-      "id": "r4-debug-forwarding-gap#1",
-      "agent": "reviewer",
-      "ok": true,
-      "ms": 114745,
-      "sessionId": "sess_60408_1783063353252_1783063353494_ffb01c576742694d462d621830cf0e1c",
-      "output": "sessionId: sess_60408_1783063353252_1783063353494_ffb01c576742694d462d621830cf0e1c\n조사 완료: 활성화 직후 출력은 “지연 후 회수”되는 경로와 “EOF 바인딩으로 영구 스킵”되는 경로가 함께 있습니다.\n\n- **최대 영향: 기존 transcript를 늦게 바인딩하면 스킵** — `setContext()`가 기본/불일치 persisted 상태에서 현재 EOF로 커서를 잡아, 그 전에 기록된 assistant 출력은 backlog가 회수하지 못합니다. `src/runtime/channels/lib/output-forwarder.mjs:112`\n\n- **중간 영향: transcript 파일 미생성 상태** — `fs.watch()`는 실패하지만 `schedulePendingTranscriptRearm()`가 파일 생성 후 watch 설치와 `forwardNewText()`를 실행해 보통 회수합니다. 단, 60초 timeout 이후 생성되면 회수되지 않습니다. `src/runtime/channels/index.mjs:357`, `src/runtime/channels/lib/output-forwarder.mjs:656`\n\n- **중간 영향: backend connect 전 ownership gate** — `forwardNewText()`는 `bridgeRuntimeConnected && owned` 전까지 no-op이라 출력이 즉시 전달되지 않습니다. 커서가 0/persisted 위치면 post-connect catch-up으로 회수되고, 이미 EOF로 잡혔으면 회수되지 않습니다. `src/runtime/channels/index.mjs:316`, `src/runtime/channels/index.mjs:801`\n\n- **낮은 영향: watch 설치 직전/직후 쓰기** — 존재하는 파일에서는 `startWatch()`가 즉시 flush를 예약해 stat 이후 기록분은 회수합니다. stat 이전 기록분은 위 EOF 규칙을 따릅니다. `src/runtime/channels/lib/output-forwarder.mjs:665`\n\nVerdict: 현재 설계는 “미생성 transcript”는 대체로 복구하지만, “이미 존재하는 transcript를 활성화 후 늦게 바인딩”한 출력은 EOF 커서 때문에 조용히 유실될 수 있습니다.",
-      "card": {
-        "session": "sess_60408_1783063353252_1783063353494_ffb01c576742694d462d621830cf0e1c",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 104242,
-        "turns": 8,
-        "tool_calls": 7,
-        "tools_per_turn": 0.9,
-        "total_tool_ms": 516,
-        "llm_stream_ms": 102684,
-        "cache_ratio": 0.7824456056127135,
-        "cached_tokens": 202752,
-        "prompt_tokens": 259126,
-        "uncached_tokens": 56374,
-        "output_tokens": 10681,
-        "cache_weighted_input_10": 76649.2,
-        "cache_weighted_input_25": 107062,
-        "prompt_growth": 40068,
-        "antipatterns": 1,
-        "issues": 14,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "broad_tool_result",
-          "service_tier",
-          "missed_parallelism",
-          "tool_churn_cluster"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-impl-usage-summary#1",
-      "agent": "worker",
-      "ok": true,
-      "ms": 18457,
-      "sessionId": "sess_4988_1783063353262_1783063353504_2e497c9ba1e6a87f33f6b7f1ee20672a",
-      "output": "sessionId: sess_4988_1783063353262_1783063353504_2e497c9ba1e6a87f33f6b7f1ee20672a\n재영님, 구현 전 계획 확인이 필요합니다.\n\n- `scripts/session-bench.mjs`의 인자 파싱 스타일을 확인해 `scripts/usage-today.mjs`를 맞춰 작성하겠습니다.\n- `MIXDOG_DATA_DIR` 우선, 기본 `~/.mixdog/data/history/agent-trace.jsonl`에서 오늘 00:00 이후 usage row만 집계하겠습니다.\n- `package.json`에 `usage:today`를 추가하고 `node --check`, 실행 1회로 검증하겠습니다.\n\n이 방향으로 바로 진행할까요?",
-      "card": {
-        "session": "sess_4988_1783063353262_1783063353504_2e497c9ba1e6a87f33f6b7f1ee20672a",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 7960,
-        "turns": 1,
-        "tool_calls": 0,
-        "tools_per_turn": 0,
-        "total_tool_ms": 0,
-        "llm_stream_ms": 6838,
-        "cache_ratio": 0.2695871946082561,
-        "cached_tokens": 2560,
-        "prompt_tokens": 9496,
-        "uncached_tokens": 6936,
-        "output_tokens": 568,
-        "cache_weighted_input_10": 7192,
-        "cache_weighted_input_25": 7576,
-        "prompt_growth": 0,
-        "antipatterns": 0,
-        "issues": 3,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "service_tier"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-explain-compact-engine#1",
-      "agent": "explore",
-      "ok": true,
-      "ms": 80852,
-      "sessionId": "sess_39880_1783063353281_1783063353525_1d1b13112d7adc7b0cf1e8742ffe6a5f",
-      "output": "sessionId: sess_39880_1783063353281_1783063353525_1d1b13112d7adc7b0cf1e8742ffe6a5f\n구조와 호출 흐름을 먼저 좁혀서 확인하겠습니다.자동 발화 조건과 fallback 기록 지점만 추가로 확인하겠습니다.라인 앵커를 위해 공개 re-export와 사후 자동 compact 진입점만 마지막으로 확인합니다.Mapped below; all code claims are anchored.\n\n- **Entry points**: `semanticCompactMessages` and `recallFastTrackCompactMessages` are exported from `engine.mjs` via the facade. `src/runtime/agent/orchestrator/session/compact.mjs:52-55`\n\n- **Live caller path**: `askSession` enters `agentLoop`, whose pre-send compaction calls recall-fasttrack first when configured, otherwise semantic. `src/runtime/agent/orchestrator/session/manager.mjs:1379-1380`, `src/runtime/agent/orchestrator/session/loop.mjs:414-459`\n\n- **Manual/clear callers**: `/clear` pre-compacts through `runSessionCompaction`, and manual `/compact` calls the same runner. `src/runtime/agent/orchestrator/session/manager.mjs:1841-1850`, `src/runtime/agent/orchestrator/session/manager.mjs:1943-1957`\n\n- **Fire condition**: auto compaction is checked before provider send after queued steering is drained; it fires when reactive retry is pending, token estimate is unknown, or `messages + reserve >= trigger`. `src/runtime/agent/orchestrator/session/loop.mjs:324-364`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:151-174`\n\n- **Policy inputs**: boundary comes from explicit boundary/context/auto limit, trigger is auto-limit or boundary minus buffer, reserve includes request reserve plus configured reserve, and agent-owned sessions force semantic. `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:39-45`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:89-148`\n\n- **Semantic flow**: sanitize first, no-op if already within effective budget unless forced, otherwise redact then select window, summarize head, emit `system + summary + tail`. `src/runtime/agent/orchestrator/session/compact/engine.mjs:278-314`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:367-392`\n\n- **Semantic tail**: context is split into protected system prefix, previous summary, and live non-summary messages; tail uses user-turn selection when users exist, otherwise non-tool cut points. `src/runtime/agent/orchestrator/session/compact/engine.mjs:100-111`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:235-250`\n\n- **User-tail rule**: semantic preserves up to `tailTurns` recent user-anchored turns within recent budget, splitting the newest turn suffix if needed. `src/runtime/agent/orchestrator/session/compact/engine.mjs:130-185`\n\n- **No-user tail rule**: semantic cut points cannot start on `tool`, grow newest-to-oldest while fitting, and fall back to minimal/empty tail cases instead of throwing. `src/runtime/agent/orchestrator/session/compact/engine.mjs:114-128`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:188-233`\n\n- **Recall-fasttrack tail**: recall preserves the newest structured turns, default max two user turns and 8k cap, drops older kept turns first, then truncates the surviving turn. `src/runtime/agent/orchestrator/session/compact/engine.mjs:449-468`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:593-619`\n\n- **Redaction hook**: compaction redacts tool-call secret arguments before window selection so preserved system/tail are measured and emitted redacted; no-op fast path intentionally skips this. `src/runtime/agent/orchestrator/session/compact/engine.mjs:286-314`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:389-392`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:644-646`\n\n- **Redactor shape**: redaction only changes sensitive argument values or raw secret substrings, preserving message/tool-call structure for provider validity. `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:219-229`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:291-304`\n\n- **Budget fallback**: both semantic and recall raise the working budget to fit mandatory `system + tail` plus summary room instead of failing on mandatory cost. `src/runtime/agent/orchestrator/session/compact/engine.mjs:319-330`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:658-663`\n\n- **Failure points**: semantic throws on missing provider, no compactable head, prompt/summary fit failure, empty summary, or final over-budget result. `src/runtime/agent/orchestrator/session/compact/engine.mjs:278-280`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:315-337`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:371-397`\n\n- **Recall failure points**: recall throws on no compactable prior context, empty recall text unless allowed, summary fit failure, or final over-budget result. `src/runtime/agent/orchestrator/session/compact/engine.mjs:653-668`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:691-699`\n\n- **Runtime fallback**: loop failures try one non-recall unanchored tool-output prune; recall failures skip that and escalate as overflow telemetry/error. `src/runtime/agent/orchestrator/session/loop.mjs:500-538`, `src/runtime/agent/orchestrator/session/loop.mjs:546-626`\n\n- **Manager fallback**: manual/clear recall-fasttrack failure can degrade to semantic; non-recall failures then try prune, otherwise record `manual_failed`/`post_turn_failed`. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:231-270`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:322-397`",
-      "card": {
-        "session": "sess_39880_1783063353281_1783063353525_1d1b13112d7adc7b0cf1e8742ffe6a5f",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 70341,
-        "turns": 7,
-        "tool_calls": 8,
-        "tools_per_turn": 1.1,
-        "total_tool_ms": 678,
-        "llm_stream_ms": 69228,
-        "cache_ratio": 0.6953839424963888,
-        "cached_tokens": 120832,
-        "prompt_tokens": 173763,
-        "uncached_tokens": 52931,
-        "output_tokens": 6592,
-        "cache_weighted_input_10": 65014.2,
-        "cache_weighted_input_25": 83139,
-        "prompt_growth": 42359,
-        "antipatterns": 1,
-        "issues": 12,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "broad_tool_result",
-          "service_tier",
-          "missed_parallelism",
-          "tool_churn_cluster"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-debug-forwarding-gap#2",
-      "agent": "reviewer",
-      "ok": true,
-      "ms": 92386,
-      "sessionId": "sess_33512_1783063371729_1783063371967_9a4241862453a1eea2e17976f6798c12",
-      "output": "sessionId: sess_33512_1783063371729_1783063371967_9a4241862453a1eea2e17976f6798c12\n요청하신 경로를 직접 추적해 실제 유실/지연 창과 backlog 복구 여부를 확인하겠습니다.조사 완료했습니다. `bridge-state.json` 기록은 실제 전달 준비보다 먼저 일어나며, 일부 창은 backlog로 복구되지 않습니다.\n\n- **최대 유실:** 기존 transcript를 새로 바인딩할 때 persisted cursor가 없으면 `setContext()`가 EOF부터 시작해, 활성화 후 바인딩 전 기록된 assistant 출력은 건너뜁니다. Backlog 복구 없음. `src/runtime/channels/lib/output-forwarder.mjs:117`\n\n- **조건부 복구:** transcript 파일이 아직 없으면 `fs.watch`는 실패하지만 re-arm poll이 파일 생성 후 `forwardNewText()`로 backlog를 읽습니다. 60초 초과·소유권 상실이면 다음 rebind 전까지 지연됩니다. `src/runtime/channels/index.mjs:469`\n\n- **지연만 발생:** `bridge-state.json` 직후부터 `backend.connect()` 완료 전까지 `_isOwner()`가 false라 watch flush는 보내지 않습니다. post-connect `forwardNewText()`가 재시도하므로 cursor가 살아 있으면 복구됩니다. `src/runtime/channels/index.mjs:1735`, `src/runtime/channels/index.mjs:801`, `src/runtime/channels/lib/output-forwarder.mjs:351`\n\n- **잘못된 바인딩 위험:** session discovery가 아직 current session을 못 찾으면 previous/latest transcript로 붙을 수 있고, 이후 올바른 transcript를 EOF로 재바인딩하면 초기 출력이 유실됩니다. pending self transcript 경로를 잡은 경우만 re-arm으로 복구됩니다. `src/runtime/channels/index.mjs:1533`, `src/runtime/channels/index.mjs:1597`\n\n- **watch 설치 경계:** stat 이후 watch 설치 전 기록은 `scheduleWatchFlush()`가 커버합니다. Backlog 복구됨. `src/runtime/channels/lib/output-forwarder.mjs:665`\n\n판정: 활성화 직후 출력은 “파일 없음” 창은 대체로 복구되지만, “이미 존재하는 transcript를 EOF로 바인딩”하는 창은 실제 silent drop 가능성이 있습니다.",
-      "card": {
-        "session": "sess_33512_1783063371729_1783063371967_9a4241862453a1eea2e17976f6798c12",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 81878,
-        "turns": 11,
-        "tool_calls": 10,
-        "tools_per_turn": 0.9,
-        "total_tool_ms": 760,
-        "llm_stream_ms": 80469,
-        "cache_ratio": 0.5901015721947442,
-        "cached_tokens": 166912,
-        "prompt_tokens": 282853,
-        "uncached_tokens": 115941,
-        "output_tokens": 5611,
-        "cache_weighted_input_10": 132632.2,
-        "cache_weighted_input_25": 157669,
-        "prompt_growth": 35859,
-        "antipatterns": 1,
-        "issues": 17,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "readonly_stall",
-          "service_tier",
-          "missed_parallelism",
-          "tool_churn_cluster"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-impl-usage-summary#2",
-      "agent": "worker",
-      "ok": true,
-      "ms": 53079,
-      "sessionId": "sess_46584_1783063434132_1783063434375_2f5bc8a596d80965dc2b007018f607cd",
-      "output": "sessionId: sess_46584_1783063434132_1783063434375_2f5bc8a596d80965dc2b007018f607cd\n요청 범위가 명확해 바로 구현과 검증까지 진행하겠습니다.이미 `usage:today`와 `scripts/usage-today.mjs`가 존재해 현재 구현 상태를 확인하고 검증하겠습니다.구현 파일과 npm script가 이미 반영된 상태라 검증 명령을 실행합니다.검증은 통과했고, 요청한 diff 요약을 위해 변경분을 확인합니다.`git diff`에는 npm script만 잡혀서, 새 스크립트 파일의 추적 상태까지 확인합니다.완료됐습니다. `usage:today` npm script와 새 `scripts/usage-today.mjs`가 준비되어 있습니다.\n\n- **Diff 요약**: `package.json`에 `usage:today` 추가, 새 스크립트는 `MIXDOG_DATA_DIR`/기본 trace 경로를 읽고 오늘 자정 이후 usage를 모델별·총합으로 집계합니다.\n\n- **검증**: `node --check scripts/usage-today.mjs` 통과, `npm run usage:today` 실행 완료.\n\n- **실행 결과**: 오늘 기준 `gpt-5.5` 78 calls, input `3,494,236`, cached `2,489,344`, output `70,355`, cost `$4.480805`.",
-      "card": {
-        "session": "sess_46584_1783063434132_1783063434375_2f5bc8a596d80965dc2b007018f607cd",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 42587,
-        "turns": 9,
-        "tool_calls": 11,
-        "tools_per_turn": 1.2,
-        "total_tool_ms": 2585,
-        "llm_stream_ms": 39637,
-        "cache_ratio": 0.6918918918918919,
-        "cached_tokens": 79872,
-        "prompt_tokens": 115440,
-        "uncached_tokens": 35568,
-        "output_tokens": 2162,
-        "cache_weighted_input_10": 43555.2,
-        "cache_weighted_input_25": 55536,
-        "prompt_growth": 7625,
-        "antipatterns": 1,
-        "issues": 14,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "service_tier",
-          "missed_parallelism",
-          "tool_churn_cluster"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-explain-compact-engine#2",
-      "agent": "explore",
-      "ok": true,
-      "ms": 95208,
-      "sessionId": "sess_8156_1783063464105_1783063464349_363e58bf8ffe09e0bd16e479ab77bef3",
-      "output": "sessionId: sess_8156_1783063464105_1783063464349_363e58bf8ffe09e0bd16e479ab77bef3\nDone — compact subsystem map, anchored to code.\n\n- **Public surface:** `compact.mjs` re-exports `semanticCompactMessages` and `recallFastTrackCompactMessages` from `engine.mjs`; the real entry points live at `engine.mjs:278` and `engine.mjs:445` (`src/runtime/agent/orchestrator/session/compact.mjs:52`).\n- **Callers:** the live loop compacts pre-send via `runRecallFastTrackCompact` or `semanticCompactMessages` (`src/runtime/agent/orchestrator/session/loop.mjs:414`, `src/runtime/agent/orchestrator/session/loop.mjs:441`); manual/auto-clear uses `runSessionCompaction` (`src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:143`).\n- **Fire condition:** loop drains queued messages, resolves policy, estimates tokens, then compacts when `shouldCompactForSession` says pressure crossed trigger or a reactive overflow retry is pending (`src/runtime/agent/orchestrator/session/loop.mjs:324`, `src/runtime/agent/orchestrator/session/loop.mjs:329`, `src/runtime/agent/orchestrator/session/loop.mjs:351`, `src/runtime/agent/orchestrator/session/loop.mjs:353`).\n- **Policy math:** pressure is `messageTokens + reserveTokens`, and compaction fires when pressure is `>= triggerTokens`, or immediately on `forceReactive` / null estimate (`src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:151`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:170`).\n- **Trigger budget:** boundary comes from session context/limits, trigger is explicit sub-boundary auto limit or `boundary - buffer`, and default buffer is 10% capped at 25% (`src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:94`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:107`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:113`, `src/runtime/agent/orchestrator/session/context-utils.mjs:94`).\n- **Compaction target:** compact budget targets a small post-compact window plus reserve, capped by boundary (`src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:163`).\n- **Semantic windowing:** `splitLiveCompactionContext` sanitizes tool pairs, separates protected system/reminder prefix, finds previous summary, and excludes summaries from live history (`src/runtime/agent/orchestrator/session/compact/engine.mjs:100`).\n- **Semantic tail:** `selectCompactionWindow` uses `preserveRecentBudget`, keeps recent user-anchored turns when users exist, otherwise uses non-tool cut points, then repairs tail tool pairing (`src/runtime/agent/orchestrator/session/compact/engine.mjs:235`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:240`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:250`).\n- **Tail budget:** preserved recent budget is explicit `preserveRecentTokens/keepTokens` or min/default bounded to 80% of compact budget (`src/runtime/agent/orchestrator/session/compact/budget.mjs:274`).\n- **User-turn tail:** semantic keeps up to `tailTurns` newest turns within budget and can split the newest oversized turn suffix (`src/runtime/agent/orchestrator/session/compact/engine.mjs:134`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:151`).\n- **No-user tail:** semantic cuts only at non-tool messages so tool results never start the preserved tail, and falls back to minimal/newer cuts instead of throwing on assistant/tool-only sessions (`src/runtime/agent/orchestrator/session/compact/engine.mjs:121`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:193`).\n- **Recall tail:** recall fast-track keeps newest structured turns, drops oldest kept turns until under cap, truncates the last surviving turn if needed, and allows no-user tails (`src/runtime/agent/orchestrator/session/compact/engine.mjs:593`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:601`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:610`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:613`).\n- **Redaction hook:** no-op semantic/recall returns sanitized messages unredacted, but actual compaction redacts tool-call secret values before window selection so preserved system/tail are measured and emitted redacted (`src/runtime/agent/orchestrator/session/compact/engine.mjs:286`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:306`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:644`).\n- **Redaction scope:** preserved-tail redaction only changes sensitive tool-call argument values / embedded raw secret pairs while preserving shape, order, ids, roles, and pairing (`src/runtime/agent/orchestrator/session/compact/text-utils.mjs:219`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:261`, `src/runtime/agent/orchestrator/session/compact/text-utils.mjs:291`).\n- **Semantic summary path:** provider prompt is fitted under headroom, provider output is schema-repaired if malformed, summary is fitted into remaining budget, then `[system, summary, tail]` is repaired and checked against budget (`src/runtime/agent/orchestrator/session/compact/engine.mjs:333`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:376`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:384`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:392`).\n- **Recall fast-track path:** recall text/prior summary are wrapped into a synthetic summary, recall injection room may be capped separately, then `[system, recallSummary, recallTail]` is repaired and checked (`src/runtime/agent/orchestrator/session/compact/engine.mjs:652`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:679`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:684`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:695`).\n- **Loop fallback:** semantic failures can recover with anchor-independent old tool-output pruning; recall-fasttrack failures skip that prune and surface overflow (`src/runtime/agent/orchestrator/session/loop.mjs:500`, `src/runtime/agent/orchestrator/session/loop.mjs:517`, `src/runtime/agent/orchestrator/session/loop.mjs:546`).\n- **Manager fallback:** recall-fasttrack failure may degrade to semantic compaction; if still uncompacted, non-recall paths try the same unanchored prune, else record `*_failed` metadata (`src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:237`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:245`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:322`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:348`).\n- **Reactive overflow:** provider context overflow triggers one retry by setting `reactiveOverflowRetryPending`, causing the next loop iteration to force pre-send compaction before retrying send (`src/runtime/agent/orchestrator/session/loop.mjs:982`, `src/runtime/agent/orchestrator/session/loop.mjs:995`, `src/runtime/agent/orchestrator/session/loop.mjs:1002`).",
-      "card": {
-        "session": "sess_8156_1783063464105_1783063464349_363e58bf8ffe09e0bd16e479ab77bef3",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 84705,
-        "turns": 11,
-        "tool_calls": 10,
-        "tools_per_turn": 0.9,
-        "total_tool_ms": 772,
-        "llm_stream_ms": 83249,
-        "cache_ratio": 0.586002007841701,
-        "cached_tokens": 179200,
-        "prompt_tokens": 305801,
-        "uncached_tokens": 126601,
-        "output_tokens": 5651,
-        "cache_weighted_input_10": 144521,
-        "cache_weighted_input_25": 171401,
-        "prompt_growth": 54901,
-        "antipatterns": 1,
-        "issues": 17,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "readonly_stall",
-          "broad_tool_result",
-          "service_tier",
-          "missed_parallelism",
-          "tool_churn_cluster"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-debug-forwarding-gap#3",
-      "agent": "reviewer",
-      "ok": true,
-      "ms": 125022,
-      "sessionId": "sess_50160_1783063467999_1783063468251_274caf327774153c94df94f9ffb0a9b9",
-      "output": "sessionId: sess_50160_1783063467999_1783063468251_274caf327774153c94df94f9ffb0a9b9\n조사 결과, 실제 손실 위험은 `fs.watch` 자체보다 “기존 transcript를 EOF로 바인딩하는 경로”가 더 큽니다.\n\n**경로 요약**: activation은 `bridge-state.json` 기록 후 ownership claim/refresh를 거쳐 backend connect, transcript bind, `OutputForwarder.startWatch()`, `forwardNewText()`로 이어집니다. 주요 앵커는 `src/runtime/channels/index.mjs:1735`, `src/runtime/channels/index.mjs:1744`, `src/runtime/channels/index.mjs:625`, `src/runtime/channels/lib/output-forwarder.mjs:93`, `src/runtime/channels/lib/output-forwarder.mjs:647`, `src/runtime/channels/lib/output-forwarder.mjs:351`입니다.\n\n- **가장 큰 손실: 기존 transcript를 EOF로 바인딩** — `catchUpFromPersisted`가 같은 transcript의 `lastFileSize`를 못 찾으면 `currentSize`로 커서를 잡아, activation 이후 bind 전 이미 쓰인 assistant output을 영구 스킵합니다. `src/runtime/channels/lib/output-forwarder.mjs:117`, `src/runtime/channels/lib/output-forwarder.mjs:122`, `src/runtime/channels/lib/output-forwarder.mjs:126`; backlog catch-up은 같은 persisted cursor가 있을 때만 복구됩니다.\n\n- **pending transcript가 poll 중 생성되는 레이스** — not-yet-created 경로는 이후 `replayFromStart`로 보호되지만, 파일이 `rebindTranscriptContext()` polling 중 “이미 존재”로 잡히면 `replayFromStart`가 꺼진 채 바인딩될 수 있습니다. `src/runtime/channels/index.mjs:438`, `src/runtime/channels/index.mjs:441`, `src/runtime/channels/index.mjs:447`; backlog catch-up은 이 분기에서는 복구하지 못합니다.\n\n- **fs.watch 대상 파일 미존재** — `startWatch()`는 없는 파일에서 실패하고 닫히지만, pending re-arm poll이 파일 생성 후 watch를 설치하고 `forwardNewText()`를 호출합니다. `src/runtime/channels/index.mjs:339`, `src/runtime/channels/index.mjs:376`, `src/runtime/channels/index.mjs:379`, `src/runtime/channels/lib/output-forwarder.mjs:657`; backlog catch-up은 60초 내 생성·동일 바인딩·owner 유지 조건에서 복구됩니다.\n\n- **bridge-state 기록 후 backend/owner 미준비** — `bridge-state.json`은 connect 전에 쓰이고, 그동안 `_isOwner()`가 false라 `forwardNewText()`는 읽지 않고 no-op입니다. `src/runtime/channels/index.mjs:1735`, `src/runtime/channels/index.mjs:1744`, `src/runtime/channels/index.mjs:316`, `src/runtime/channels/lib/output-forwarder.mjs:351`; backlog catch-up은 post-connect `forwardNewText()`/re-arm으로 지연 복구됩니다.\n\n- **stat→watch 설치 사이의 짧은 창** — `setContext()`가 현재 크기를 잡은 뒤 `startWatch()`가 즉시 flush를 예약해 그 사이에 쓰인 bytes를 읽습니다. `src/runtime/channels/lib/output-forwarder.mjs:112`, `src/runtime/channels/lib/output-forwarder.mjs:665`, `src/runtime/channels/lib/output-forwarder.mjs:724`; backlog catch-up은 정상 복구됩니다.\n\nVerdict: 현재 코드는 대부분의 activation 지연은 복구하지만, “이미 존재하는 transcript를 persisted cursor 없이 바인딩”하는 경로는 첫 assistant output을 조용히 잃을 수 있습니다.",
-      "card": {
-        "session": "sess_50160_1783063467999_1783063468251_274caf327774153c94df94f9ffb0a9b9",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 114511,
-        "turns": 6,
-        "tool_calls": 10,
-        "tools_per_turn": 1.7,
-        "total_tool_ms": 903,
-        "llm_stream_ms": 113542,
-        "cache_ratio": 0.7140668668976461,
-        "cached_tokens": 130048,
-        "prompt_tokens": 182123,
-        "uncached_tokens": 52075,
-        "output_tokens": 11313,
-        "cache_weighted_input_10": 65079.8,
-        "cache_weighted_input_25": 84587,
-        "prompt_growth": 41209,
-        "antipatterns": 0,
-        "issues": 9,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "broad_tool_result",
-          "service_tier"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-impl-usage-summary#3",
-      "agent": "worker",
-      "ok": true,
-      "ms": 47382,
-      "sessionId": "sess_35916_1783063487207_1783063487441_71e4fa81bf39a50301d9f3fe5118e29f",
-      "output": "sessionId: sess_35916_1783063487207_1783063487441_71e4fa81bf39a50301d9f3fe5118e29f\n요청 범위가 작고 구현 지시가 명확해 바로 적용하겠습니다.이미 `usage:today` 항목과 대상 파일이 있어, 기존 상태를 확인한 뒤 요구사항에 맞게 보정하겠습니다.구현 파일이 이미 존재해 요구사항과 맞는지 검증까지 진행하겠습니다.검증은 통과했고, 새 파일의 추적 상태만 확인해 최종 요약하겠습니다.완료됐습니다, 재영님. `usage:today` 스크립트와 `scripts/usage-today.mjs`가 추가되어 오늘 자정 이후 사용량을 모델별/총합으로 출력합니다.\n\n- **변경 요약**: `package.json`에 `usage:today` 추가, 새 스크립트는 `MIXDOG_DATA_DIR` 또는 기본 `~/.mixdog/data/history/agent-trace.jsonl`을 읽습니다.\n\n- **출력 내용**: 모델별 `calls`, `input`, `cached`, `output`, `costUsd`와 `GRAND TOTAL` 라인을 표시합니다.\n\n- **검증**: `node --check scripts/usage-today.mjs` 통과, `npm run usage:today` 실행 성공.",
-      "card": {
-        "session": "sess_35916_1783063487207_1783063487441_71e4fa81bf39a50301d9f3fe5118e29f",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 36892,
-        "turns": 7,
-        "tool_calls": 7,
-        "tools_per_turn": 1,
-        "total_tool_ms": 1780,
-        "llm_stream_ms": 34501,
-        "cache_ratio": 0.770208686675024,
-        "cached_tokens": 52224,
-        "prompt_tokens": 67805,
-        "uncached_tokens": 15581,
-        "output_tokens": 2014,
-        "cache_weighted_input_10": 20803.4,
-        "cache_weighted_input_25": 28637,
-        "prompt_growth": 2459,
-        "antipatterns": 1,
-        "issues": 11,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "service_tier",
-          "missed_parallelism",
-          "tool_churn_cluster"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-explain-compact-engine#3",
-      "agent": "explore",
-      "ok": true,
-      "ms": 71997,
-      "sessionId": "sess_59820_1783063534589_1783063534826_5b1388cb0dce5fcf4952797e321186d1",
-      "output": "sessionId: sess_59820_1783063534589_1783063534826_5b1388cb0dce5fcf4952797e321186d1\n완료했습니다 — compaction subsystem architecture brief입니다.\n\n- **Entry points**: public API is `semanticCompactMessages` and `recallFastTrackCompactMessages` re-exported by `compact.mjs`; implementations live in `engine.mjs`. `src/runtime/agent/orchestrator/session/compact.mjs:52`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:278`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:445`\n\n- **Callers**: live turns enter through `agentLoop`, manual/clear paths enter through `runSessionCompaction`; `askSession` calls `agentLoop`, `clearSessionMessages` and `compactSessionMessages` call the runner. `src/runtime/agent/orchestrator/session/loop.mjs:414`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:207`, `src/runtime/agent/orchestrator/session/manager.mjs:1379`, `src/runtime/agent/orchestrator/session/manager.mjs:1849`, `src/runtime/agent/orchestrator/session/manager.mjs:1950`\n\n- **When it fires**: pre-send drains queued input first, resolves policy, estimates message tokens, then compacts when pressure reaches trigger or reactive overflow is pending. `src/runtime/agent/orchestrator/session/loop.mjs:324`, `src/runtime/agent/orchestrator/session/loop.mjs:329`, `src/runtime/agent/orchestrator/session/loop.mjs:351`, `src/runtime/agent/orchestrator/session/loop.mjs:353`\n\n- **Policy**: auto can be disabled; boundary comes from explicit/context window/auto limit, trigger is boundary minus buffer, reserve includes tool request reserve, and compact type selects semantic vs recall-fasttrack. `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:89`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:94`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:117`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:121`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:123`\n\n- **Execution branch**: recall-fasttrack runs first when selected; otherwise semantic runs with `force:true`, compact budget, reserve, timeout, and tail settings. `src/runtime/agent/orchestrator/session/loop.mjs:414`, `src/runtime/agent/orchestrator/session/loop.mjs:439`, `src/runtime/agent/orchestrator/session/loop.mjs:441`, `src/runtime/agent/orchestrator/session/loop.mjs:458`\n\n- **Semantic tail**: messages are sanitized/deduped, protected prefix and previous summary are separated, live summary messages are removed, then tail is chosen by recent budget and `tailTurns`. `src/runtime/agent/orchestrator/session/compact/engine.mjs:100`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:102`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:110`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:237`\n\n- **Semantic cut rules**: if live has user turns, preserve newest turns within budget; if not, cut only at non-`tool` indices so assistant/tool pairs stay valid. `src/runtime/agent/orchestrator/session/compact/engine.mjs:134`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:151`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:121`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:193`\n\n- **Window output**: selected head becomes summary input, tail is preserved verbatim after tool-pair repair, and only an empty live window is unrecoverable. `src/runtime/agent/orchestrator/session/compact/engine.mjs:244`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:250`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:254`\n\n- **Recall tail**: recall-fasttrack preserves newest structured turns, defaulting to 2 user turns and 8k tokens; it drops oldest turns first, then truncates oldest surviving content. `src/runtime/agent/orchestrator/session/compact/engine.mjs:449`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:467`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:601`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:609`\n\n- **Redaction**: no-op fast path returns sanitized messages unchanged; real compaction redacts tool-call secret values before window selection, so preserved system/tail are measured and emitted redacted. `src/runtime/agent/orchestrator/session/compact/engine.mjs:286`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:289`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:306`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:312`\n\n- **Budget behavior**: mandatory preserved system+tail cost can raise budget to include minimum summary room; final result must fit or throws. `src/runtime/agent/orchestrator/session/compact/engine.mjs:319`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:328`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:395`\n\n- **Failure/fallbacks**: semantic validates provider/prompt/summary/final fit and throws on failures; live loop catches and, only for non-recall failures, tries unanchored tool-output prune before raising overflow. `src/runtime/agent/orchestrator/session/compact/engine.mjs:279`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:335`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:385`, `src/runtime/agent/orchestrator/session/loop.mjs:500`, `src/runtime/agent/orchestrator/session/loop.mjs:517`\n\n- **Runner fallbacks**: manual/clear recall-fasttrack failure can degrade to semantic; if still uncompacted, non-recall failures try unanchored prune, otherwise failure is recorded without replacing history. `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:237`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:245`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:327`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:348`",
-      "card": {
-        "session": "sess_59820_1783063534589_1783063534826_5b1388cb0dce5fcf4952797e321186d1",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 61497,
-        "turns": 7,
-        "tool_calls": 10,
-        "tools_per_turn": 1.4,
-        "total_tool_ms": 949,
-        "llm_stream_ms": 60317,
-        "cache_ratio": 0.6540460374630791,
-        "cached_tokens": 104960,
-        "prompt_tokens": 160478,
-        "uncached_tokens": 55518,
-        "output_tokens": 4207,
-        "cache_weighted_input_10": 66014,
-        "cache_weighted_input_25": 81758,
-        "prompt_growth": 34422,
-        "antipatterns": 0,
-        "issues": 10,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "broad_tool_result",
-          "service_tier"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    }
-  ],
-  "task_complete": true,
-  "task_errors": [],
-  "score_complete": true,
-  "score_errors": [],
-  "group": {
-    "n": 9,
-    "wall_ms": 67179.2,
-    "turns": 7.4,
-    "tool_calls": 8.1,
-    "tools_per_turn": 1,
-    "total_tool_ms": 993.7,
-    "llm_stream_ms": 65607.2,
-    "cache_ratio": 0.6,
-    "cached_tokens": 115484.4,
-    "prompt_tokens": 172987.2,
-    "uncached_tokens": 57502.8,
-    "output_tokens": 5422.1,
-    "cache_weighted_input_10": 69051.2,
-    "cache_weighted_input_25": 86373.9,
-    "prompt_growth": 28766.9,
-    "antipatterns": 0.7,
-    "issues": 11.9,
-    "grep_sweep_count": 0
-  },
-  "cards": [
-    {
-      "session": "sess_60408_1783063353252_1783063353494_ffb01c576742694d462d621830cf0e1c",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 104242,
-      "turns": 8,
-      "tool_calls": 7,
-      "tools_per_turn": 0.9,
-      "total_tool_ms": 516,
-      "llm_stream_ms": 102684,
-      "cache_ratio": 0.7824456056127135,
-      "cached_tokens": 202752,
-      "prompt_tokens": 259126,
-      "uncached_tokens": 56374,
-      "output_tokens": 10681,
-      "cache_weighted_input_10": 76649.2,
-      "cache_weighted_input_25": 107062,
-      "prompt_growth": 40068,
-      "antipatterns": 1,
-      "issues": 14,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "broad_tool_result",
-        "service_tier",
-        "missed_parallelism",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_4988_1783063353262_1783063353504_2e497c9ba1e6a87f33f6b7f1ee20672a",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 7960,
-      "turns": 1,
-      "tool_calls": 0,
-      "tools_per_turn": 0,
-      "total_tool_ms": 0,
-      "llm_stream_ms": 6838,
-      "cache_ratio": 0.2695871946082561,
-      "cached_tokens": 2560,
-      "prompt_tokens": 9496,
-      "uncached_tokens": 6936,
-      "output_tokens": 568,
-      "cache_weighted_input_10": 7192,
-      "cache_weighted_input_25": 7576,
-      "prompt_growth": 0,
-      "antipatterns": 0,
-      "issues": 3,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "service_tier"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_39880_1783063353281_1783063353525_1d1b13112d7adc7b0cf1e8742ffe6a5f",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 70341,
-      "turns": 7,
-      "tool_calls": 8,
-      "tools_per_turn": 1.1,
-      "total_tool_ms": 678,
-      "llm_stream_ms": 69228,
-      "cache_ratio": 0.6953839424963888,
-      "cached_tokens": 120832,
-      "prompt_tokens": 173763,
-      "uncached_tokens": 52931,
-      "output_tokens": 6592,
-      "cache_weighted_input_10": 65014.2,
-      "cache_weighted_input_25": 83139,
-      "prompt_growth": 42359,
-      "antipatterns": 1,
-      "issues": 12,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "broad_tool_result",
-        "service_tier",
-        "missed_parallelism",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_33512_1783063371729_1783063371967_9a4241862453a1eea2e17976f6798c12",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 81878,
-      "turns": 11,
-      "tool_calls": 10,
-      "tools_per_turn": 0.9,
-      "total_tool_ms": 760,
-      "llm_stream_ms": 80469,
-      "cache_ratio": 0.5901015721947442,
-      "cached_tokens": 166912,
-      "prompt_tokens": 282853,
-      "uncached_tokens": 115941,
-      "output_tokens": 5611,
-      "cache_weighted_input_10": 132632.2,
-      "cache_weighted_input_25": 157669,
-      "prompt_growth": 35859,
-      "antipatterns": 1,
-      "issues": 17,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "readonly_stall",
-        "service_tier",
-        "missed_parallelism",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_46584_1783063434132_1783063434375_2f5bc8a596d80965dc2b007018f607cd",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 42587,
-      "turns": 9,
-      "tool_calls": 11,
-      "tools_per_turn": 1.2,
-      "total_tool_ms": 2585,
-      "llm_stream_ms": 39637,
-      "cache_ratio": 0.6918918918918919,
-      "cached_tokens": 79872,
-      "prompt_tokens": 115440,
-      "uncached_tokens": 35568,
-      "output_tokens": 2162,
-      "cache_weighted_input_10": 43555.2,
-      "cache_weighted_input_25": 55536,
-      "prompt_growth": 7625,
-      "antipatterns": 1,
-      "issues": 14,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "service_tier",
-        "missed_parallelism",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_8156_1783063464105_1783063464349_363e58bf8ffe09e0bd16e479ab77bef3",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 84705,
-      "turns": 11,
-      "tool_calls": 10,
-      "tools_per_turn": 0.9,
-      "total_tool_ms": 772,
-      "llm_stream_ms": 83249,
-      "cache_ratio": 0.586002007841701,
-      "cached_tokens": 179200,
-      "prompt_tokens": 305801,
-      "uncached_tokens": 126601,
-      "output_tokens": 5651,
-      "cache_weighted_input_10": 144521,
-      "cache_weighted_input_25": 171401,
-      "prompt_growth": 54901,
-      "antipatterns": 1,
-      "issues": 17,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "readonly_stall",
-        "broad_tool_result",
-        "service_tier",
-        "missed_parallelism",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_50160_1783063467999_1783063468251_274caf327774153c94df94f9ffb0a9b9",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 114511,
-      "turns": 6,
-      "tool_calls": 10,
-      "tools_per_turn": 1.7,
-      "total_tool_ms": 903,
-      "llm_stream_ms": 113542,
-      "cache_ratio": 0.7140668668976461,
-      "cached_tokens": 130048,
-      "prompt_tokens": 182123,
-      "uncached_tokens": 52075,
-      "output_tokens": 11313,
-      "cache_weighted_input_10": 65079.8,
-      "cache_weighted_input_25": 84587,
-      "prompt_growth": 41209,
-      "antipatterns": 0,
-      "issues": 9,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "broad_tool_result",
-        "service_tier"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_35916_1783063487207_1783063487441_71e4fa81bf39a50301d9f3fe5118e29f",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 36892,
-      "turns": 7,
-      "tool_calls": 7,
-      "tools_per_turn": 1,
-      "total_tool_ms": 1780,
-      "llm_stream_ms": 34501,
-      "cache_ratio": 0.770208686675024,
-      "cached_tokens": 52224,
-      "prompt_tokens": 67805,
-      "uncached_tokens": 15581,
-      "output_tokens": 2014,
-      "cache_weighted_input_10": 20803.4,
-      "cache_weighted_input_25": 28637,
-      "prompt_growth": 2459,
-      "antipatterns": 1,
-      "issues": 11,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "service_tier",
-        "missed_parallelism",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_59820_1783063534589_1783063534826_5b1388cb0dce5fcf4952797e321186d1",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 61497,
-      "turns": 7,
-      "tool_calls": 10,
-      "tools_per_turn": 1.4,
-      "total_tool_ms": 949,
-      "llm_stream_ms": 60317,
-      "cache_ratio": 0.6540460374630791,
-      "cached_tokens": 104960,
-      "prompt_tokens": 160478,
-      "uncached_tokens": 55518,
-      "output_tokens": 4207,
-      "cache_weighted_input_10": 66014,
-      "cache_weighted_input_25": 81758,
-      "prompt_growth": 34422,
-      "antipatterns": 0,
-      "issues": 10,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "broad_tool_result",
-        "service_tier"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    }
-  ]
-}