mixdog 0.9.18 → 0.9.20

package/scripts/bench/round-r26-parallel3.json

@@ -1,671 +0,0 @@
-{
-  "round": "26",
-  "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_36552_1783069830153_1783069830396_d13a12bb2156913c8a60e9442598d0c8",
-    "sess_57952_1783069830168_1783069830408_4ec1cd0f3324c0260f3de8ee49f265d7",
-    "sess_48676_1783069830186_1783069830428_0a5bfd5ec7bb592c4472884b507d14db",
-    "sess_36256_1783069847899_1783069848122_f54e2a821584c50ac4e979c50a601646",
-    "sess_18484_1783069914981_1783069915210_7f0733a8893fa79343fc8708019b0acd",
-    "sess_51948_1783069931009_1783069931236_da461bb4146fd67343d59e15f87c3b45",
-    "sess_46144_1783069938017_1783069938263_a5dae16078bdeb9673d932cc9b9574a5",
-    "sess_47516_1783070000924_1783070001163_6b4f4d171e302f017c024a040f0956a6",
-    "sess_39060_1783070024383_1783070024607_f04e5c1bfb6beb27bc3ac82d8dbc7a46"
-  ],
-  "results": [
-    {
-      "id": "r4-debug-forwarding-gap#1",
-      "agent": "reviewer",
-      "ok": true,
-      "ms": 100861,
-      "sessionId": "sess_36552_1783069830153_1783069830396_d13a12bb2156913c8a60e9442598d0c8",
-      "output": "sessionId: sess_36552_1783069830153_1783069830396_d13a12bb2156913c8a60e9442598d0c8\n관련 경로를 직접 추적해 첫 전달 전까지의 손실/지연 지점을 확인하겠습니다.조사 완료: 활성화 직후 첫 전달까지의 주된 위험은 “이미 존재하는 transcript를 EOF로 바인딩하는 경우”이며, 이 경우 backlog catch-up이 복구하지 못합니다.\n\n- **최대 영향 — 기존 transcript EOF 바인딩:** `bindPersistedTranscriptIfAny()`가 `catchUpFromPersisted`만 넘기고, persisted 경로가 아니면 `setContext()`가 현재 파일 끝으로 커서를 둡니다. 이미 기록된 assistant 출력은 파싱되지 않습니다. 복구 안 됨. `src/runtime/channels/index.mjs:625-629`, `src/runtime/channels/index.mjs:434-449`, `src/runtime/channels/lib/output-forwarder.mjs:117-127`\n\n- **pending → exists 전환 중 replay 누락:** activation bind 경로는 `options.catchUp`이 없어, poll 중 pending transcript가 생성되어도 `replayFromStart`가 false입니다. 생성 직후 이미 쌓인 첫 출력은 EOF 바인딩으로 건너뜁니다. 복구 안 됨. `src/runtime/channels/index.mjs:438-443`\n\n- **없는 transcript에 대한 `fs.watch` 실패:** `startWatch()`는 없는 파일에서 실패하고 닫히지만, re-arm poll이 파일 생성 후 watch를 설치하고 `forwardNewText()`로 backlog를 읽습니다. 60초 내 생성·소유 유지 시 복구됨, 그 뒤는 다음 rebind/trigger 전까지 지연. `src/runtime/channels/index.mjs:357-392`, `src/runtime/channels/index.mjs:492-499`, `src/runtime/channels/lib/output-forwarder.mjs:647-670`\n\n- **pre-connect ownership gate:** binding task가 `backend.connect()` 전 시작되므로 `_isOwner()`가 false여서 flush/rearm이 무시될 수 있습니다. 커서는 전진하지 않아 정상 start에서는 post-connect forward/rearm으로 복구됩니다. `src/runtime/channels/index.mjs:758-812`, `src/runtime/channels/index.mjs:316`, `src/runtime/channels/lib/output-forwarder.mjs:36-41`, `src/runtime/channels/lib/output-forwarder.mjs:351-354`\n\n- **채널 바인딩 부재:** persisted/main channel을 못 찾으면 transcript bind 자체가 없고, 나중에 기존 파일을 처음 bind하면 EOF 규칙으로 이전 출력이 빠질 수 있습니다. 복구 안 됨. `src/runtime/channels/index.mjs:610-624`, `src/runtime/channels/index.mjs:1597-1617`\n\n- **stat과 watch 설치 사이 쓰기:** `startWatch()` 직후 강제 flush가 있어 이 짧은 구간의 backlog는 복구됩니다. `src/runtime/channels/lib/output-forwarder.mjs:665-666`\n\n**Verdict:** not-yet-created transcript 창은 대체로 catch-up 되지만, activation 시 이미 존재하는 새 transcript를 EOF로 바인딩하는 경로는 첫 assistant 출력을 실제로 silently drop할 수 있습니다.",
-      "card": {
-        "session": "sess_36552_1783069830153_1783069830396_d13a12bb2156913c8a60e9442598d0c8",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 90358,
-        "turns": 7,
-        "tool_calls": 6,
-        "tools_per_turn": 0.9,
-        "total_tool_ms": 315,
-        "llm_stream_ms": 89636,
-        "cache_ratio": 0.722787919558858,
-        "cached_tokens": 141824,
-        "prompt_tokens": 196218,
-        "uncached_tokens": 54394,
-        "output_tokens": 8930,
-        "cache_weighted_input_10": 68576.4,
-        "cache_weighted_input_25": 89850,
-        "prompt_growth": 38506,
-        "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-impl-usage-summary#1",
-      "agent": "worker",
-      "ok": true,
-      "ms": 17725,
-      "sessionId": "sess_57952_1783069830168_1783069830408_4ec1cd0f3324c0260f3de8ee49f265d7",
-      "output": "sessionId: sess_57952_1783069830168_1783069830408_4ec1cd0f3324c0260f3de8ee49f265d7\n재영님, 진행 전 계획입니다. 승인해 주시면 바로 구현·검증하겠습니다.\n\n- **구현** `scripts/session-bench.mjs`의 arg parsing 스타일을 확인해 `scripts/usage-today.mjs`를 추가합니다.\n- **동작** `MIXDOG_DATA_DIR` 우선, 기본 `~/.mixdog/data/history/agent-trace.jsonl`에서 오늘 00:00 이후 usage row만 집계합니다.\n- **출력/검증** 모델별 calls/input/cached/output/costUsd와 grand total을 출력하고, `package.json`에 `usage:today`를 추가한 뒤 `node --check`와 1회 실행으로 확인합니다.",
-      "card": {
-        "session": "sess_57952_1783069830168_1783069830408_4ec1cd0f3324c0260f3de8ee49f265d7",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 7228,
-        "turns": 1,
-        "tool_calls": 0,
-        "tools_per_turn": 0,
-        "total_tool_ms": 0,
-        "llm_stream_ms": 6268,
-        "cache_ratio": 0.2695871946082561,
-        "cached_tokens": 2560,
-        "prompt_tokens": 9496,
-        "uncached_tokens": 6936,
-        "output_tokens": 495,
-        "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": 84789,
-      "sessionId": "sess_48676_1783069830186_1783069830428_0a5bfd5ec7bb592c4472884b507d14db",
-      "output": "sessionId: sess_48676_1783069830186_1783069830428_0a5bfd5ec7bb592c4472884b507d14db\n필요한 앵커만 한 번에 묶어서 확인하겠습니다.자동 경로의 호출부만 추가로 고정하겠습니다.recall-fasttrack 래퍼의 실제 호출 앵커만 마지막으로 확인합니다.완료했습니다 — compaction 서브시스템 요약은 아래와 같습니다.\n\n- **진입점**: `compact.mjs`가 `semanticCompactMessages`/`recallFastTrackCompactMessages`를 재수출하고, 실제 구현은 `engine.mjs`의 semantic/recall 엔트리다. `compact.mjs:52-55`, `engine.mjs:278`, `engine.mjs:445`\n\n- **호출부**: 자동 pre-send 경로는 `loop.mjs`가 policy를 계산한 뒤 recall 또는 semantic compact를 호출하고, 수동/API 경로는 `compactSessionMessages()`→`runSessionCompaction()`이다. `loop.mjs:329-353`, `loop.mjs:414-460`, `manager.mjs:1943-1957`, `compaction-runner.mjs:143-149`\n\n- **발화 조건**: 자동은 `auto`가 켜지고 boundary/trigger가 계산된 뒤 `messageTokens + reserve >= trigger`이거나 token 추정 실패/반응형 overflow 재시도면 실행된다. `compact-policy.mjs:89-117`, `compact-policy.mjs:151-175`, `loop.mjs:351-365`\n\n- **수동/러너 조건**: `runSessionCompaction()`은 closed/auto-off/짧은 메시지/무 boundary를 건너뛰고, force가 아니면 `pressureTokens < triggerTokens`에서 “below threshold”로 종료한다. `compaction-runner.mjs:143-176`\n\n- **타입 선택**: compact type은 env/config에서 정규화되며 기본은 semantic, agent-owned 세션은 semantic으로 고정된다. `constants.mjs:42-74`, `compact-policy.mjs:39-52`, `context-meta.mjs:222-230`\n\n- **semantic tail 선택**: semantic은 system/protected prefix와 이전 summary를 분리한 뒤, user turn이 있으면 최근 `tailTurns`를 `preserveRecentBudget` 안에서 보존하고 필요 시 최신 turn suffix로 자른다. `engine.mjs:100-111`, `engine.mjs:134-185`, `engine.mjs:235-245`\n\n- **no-user tail 선택**: user turn이 없으면 tool 결과로 시작하지 않는 cut point만 후보로 삼아 최신부터 budget에 맞는 suffix를 고르고, 비어 있거나 전체가 tail이 되는 경우 head를 남기는 보정을 한다. `engine.mjs:121-127`, `engine.mjs:193-232`\n\n- **recall tail 선택**: recall-fasttrack은 최신 최대 2 user-anchored turns를 보존하고 8k cap을 넘으면 오래된 turn을 버리거나 단일 turn을 truncation하며 no-user tail도 허용한다. `engine.mjs:449-468`, `engine.mjs:593-619`\n\n- **redaction 위치**: compaction이 실제 진행될 때만 base sanitized transcript에 `redactToolCallSecretsInMessages()`를 적용한 후 window selection을 하므로 preserved system/tail도 redacted 상태로 측정·출력된다. `engine.mjs:284-314`, `engine.mjs:389-392`, `engine.mjs:644-651`\n\n- **redaction 범위**: redaction은 tool call argument 값만 shape-preserving으로 scrub하며 secret key/raw secret pair를 `[redacted]`로 바꾸고 role/content/tool ids/order는 유지한다. `text-utils.mjs:219-229`, `text-utils.mjs:261-288`, `text-utils.mjs:291-325`\n\n- **semantic 실패 처리**: provider 없음, 빈 head, prompt/summary fit 실패, empty summary, 최종 budget 초과는 throw되고 runner/loop가 error telemetry로 기록한다. `engine.mjs:278-317`, `engine.mjs:333-386`, `engine.mjs:392-397`, `loop.mjs:486-560`\n\n- **fallback 경로**: recall-fasttrack 실패는 runner에서 semantic fallback을 한 번 시도하고, semantic/non-recall 실패는 `pruneToolOutputsUnanchored()`가 budget 안에 들어가면 복구하며 recall 실패에는 prune fallback을 건너뛴다. `compaction-runner.mjs:231-279`, `compaction-runner.mjs:322-347`, `loop.mjs:500-545`\n\n- **최종 실패/성공 기록**: 복구 실패 시 `post_turn_failed`/`manual_failed` 또는 loop `overflow_failed`를 기록하고, 성공 시 messages 교체·providerState 초기화·compaction counters/lastStage를 갱신한다. `compaction-runner.mjs:348-397`, `compaction-runner.mjs:410-447`, `loop.mjs:552-626`, `loop.mjs:630-656`",
-      "card": {
-        "session": "sess_48676_1783069830186_1783069830428_0a5bfd5ec7bb592c4472884b507d14db",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 74277,
-        "turns": 9,
-        "tool_calls": 14,
-        "tools_per_turn": 1.6,
-        "total_tool_ms": 2165,
-        "llm_stream_ms": 72331,
-        "cache_ratio": 0.7512216810492917,
-        "cached_tokens": 185856,
-        "prompt_tokens": 247405,
-        "uncached_tokens": 61549,
-        "output_tokens": 5622,
-        "cache_weighted_input_10": 80134.6,
-        "cache_weighted_input_25": 108013,
-        "prompt_growth": 44952,
-        "antipatterns": 1,
-        "issues": 15,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "readonly_stall",
-          "broad_tool_result",
-          "service_tier",
-          "tool_churn_cluster"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-debug-forwarding-gap#2",
-      "agent": "reviewer",
-      "ok": true,
-      "ms": 90118,
-      "sessionId": "sess_36256_1783069847899_1783069848122_f54e2a821584c50ac4e979c50a601646",
-      "output": "sessionId: sess_36256_1783069847899_1783069848122_f54e2a821584c50ac4e979c50a601646\n요청 범위가 읽기 전용 조사라서 바로 코드 경로를 추적해 확인하겠습니다.조사 완료: 활성화부터 첫 전달까지의 경로에는 “지연 후 회수”와 “EOF 바인딩으로 영구 누락”이 섞여 있습니다.\n\n**영향도 순 findings**\n- **영구 누락 가능**: 채널 ID가 없으면 `bindPersistedTranscriptIfAny()`가 반환하고, 이후 첫 inbound에서 이미 존재하는 transcript를 `applyTranscriptBinding()`으로 EOF 바인딩합니다. 그 사이 출력은 catch-up 없이 스킵됩니다. `src/runtime/channels/index.mjs:624`, `src/runtime/channels/index.mjs:1597`, `src/runtime/channels/lib/output-forwarder.mjs:125`\n- **영구 누락 가능**: 활성화 직후 session/transcript discovery가 늦어 stale binding을 쓰다가, 나중에 “이미 존재하는” self transcript로 steal하면 EOF 바인딩되어 이전 출력이 스킵됩니다. 500ms poll 밖이면 회수 안 됩니다. `src/runtime/channels/index.mjs:1526`, `src/runtime/channels/index.mjs:1571`, `src/runtime/channels/index.mjs:1597`\n- **대체로 회수됨**: transcript 파일이 아직 없으면 `fs.watch()` 설치가 실패하지만 pending re-arm이 60초 동안 polling 후 `startWatch()`와 `forwardNewText()`로 backlog를 읽습니다. 60초 초과 또는 재트리거 없음이면 지연/미회수입니다. `src/runtime/channels/index.mjs:340`, `src/runtime/channels/index.mjs:357`, `src/runtime/channels/index.mjs:376`\n- **회수됨**: backend connect 전에는 ownership probe가 false라 `forwardNewText()`가 no-op이지만, connect 후 `forwardNewText()` 또는 re-arm 재설정이 다시 실행되어 backlog를 읽습니다. `src/runtime/channels/lib/output-forwarder.mjs:351`, `src/runtime/channels/index.mjs:776`, `src/runtime/channels/index.mjs:801`\n- **회수됨**: `setContext()` stat과 `fs.watch()` 설치 사이에 쓰인 bytes는 `startWatch()`의 즉시 flush가 커버합니다. `src/runtime/channels/lib/output-forwarder.mjs:111`, `src/runtime/channels/lib/output-forwarder.mjs:665`\n\n**경로 요약**: `writeBridgeState(true)` → ownership claim/refresh → `startOwnedRuntime()` → `bindPersistedTranscriptIfAny()`/`rebindTranscriptContext()` → `OutputForwarder.setContext()` → `startWatch()` → `scheduleWatchFlush()` → `forwardNewText()` → `deliverQueueItem()`.\n\n**Verdict**: not-yet-created transcript와 pre-connect ownership gap은 대부분 catch-up되지만, “채널/세션 바인딩이 늦고 transcript가 이미 존재하는” 경우는 EOF 바인딩으로 첫 출력이 silently dropped 될 수 있습니다.",
-      "card": {
-        "session": "sess_36256_1783069847899_1783069848122_f54e2a821584c50ac4e979c50a601646",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 79636,
-        "turns": 7,
-        "tool_calls": 6,
-        "tools_per_turn": 0.9,
-        "total_tool_ms": 493,
-        "llm_stream_ms": 78523,
-        "cache_ratio": 0.6874013304367773,
-        "cached_tokens": 101888,
-        "prompt_tokens": 148222,
-        "uncached_tokens": 46334,
-        "output_tokens": 6629,
-        "cache_weighted_input_10": 56522.8,
-        "cache_weighted_input_25": 71806,
-        "prompt_growth": 30328,
-        "antipatterns": 1,
-        "issues": 13,
-        "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#2",
-      "agent": "worker",
-      "ok": true,
-      "ms": 118432,
-      "sessionId": "sess_18484_1783069914981_1783069915210_7f0733a8893fa79343fc8708019b0acd",
-      "output": "sessionId: sess_18484_1783069914981_1783069915210_7f0733a8893fa79343fc8708019b0acd\n요청 범위가 명확해 바로 구현하고 검증까지 진행하겠습니다.구현 파일이 이미 작업트리에 있어 요구사항 기준으로 점검한 뒤 필요한 수정과 검증을 진행합니다.검증은 통과했고, diff 요약만 확인합니다.완료했습니다. `usage:today` 스크립트와 `scripts/usage-today.mjs`가 준비되어 오늘 사용량을 모델별/총합으로 출력합니다.\n\n- **Diff 요약**: `package.json`에 `\"usage:today\": \"node scripts/usage-today.mjs\"` 추가, `scripts/usage-today.mjs` 신규 추가.\n- **검증**: `node --check scripts/usage-today.mjs` 통과, `npm run usage:today` 실행 성공.\n- **참고**: 작업 전부터 `src/runtime/channels/lib/event-queue.mjs`, `src/runtime/channels/lib/output-forwarder.mjs`에 별도 수정이 있었습니다.",
-      "card": {
-        "session": "sess_18484_1783069914981_1783069915210_7f0733a8893fa79343fc8708019b0acd",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 107937,
-        "turns": 14,
-        "tool_calls": 20,
-        "tools_per_turn": 1.4,
-        "total_tool_ms": 2983,
-        "llm_stream_ms": 105065,
-        "cache_ratio": 0.7753208965282127,
-        "cached_tokens": 293376,
-        "prompt_tokens": 378393,
-        "uncached_tokens": 85017,
-        "output_tokens": 9176,
-        "cache_weighted_input_10": 114354.6,
-        "cache_weighted_input_25": 158361,
-        "prompt_growth": 30327,
-        "antipatterns": 1,
-        "issues": 16,
-        "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-explain-compact-engine#2",
-      "agent": "explore",
-      "ok": true,
-      "ms": 69918,
-      "sessionId": "sess_51948_1783069931009_1783069931236_da461bb4146fd67343d59e15f87c3b45",
-      "output": "sessionId: sess_51948_1783069931009_1783069931236_da461bb4146fd67343d59e15f87c3b45\nMapped.\n\n- **Entry points:** `semanticCompactMessages` and `recallFastTrackCompactMessages` are exported from `engine.mjs` and re-exported by `compact.mjs`. `src/runtime/agent/orchestrator/session/compact/engine.mjs:278`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:445`, `src/runtime/agent/orchestrator/session/compact.mjs:52`\n\n- **Callers:** auto pre-send compaction runs from `agentLoop`; manual `/compact` and clear-before-clear run through `runSessionCompaction`. `src/runtime/agent/orchestrator/session/loop.mjs:329`, `src/runtime/agent/orchestrator/session/loop.mjs:416`, `src/runtime/agent/orchestrator/session/loop.mjs:441`, `src/runtime/agent/orchestrator/session/manager.mjs:1849`, `src/runtime/agent/orchestrator/session/manager.mjs:1950`\n\n- **Fire decision:** worker compaction is enabled by `resolveWorkerCompactPolicy`, then fires when pressure plus reserve reaches `triggerTokens`, token estimate is unavailable, or a reactive overflow retry forces it. `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:89`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:170`, `src/runtime/agent/orchestrator/session/loop.mjs:351`\n\n- **Reactive path:** provider context overflow sets `reactiveOverflowRetryPending`, loops back, and tags the next compact as `trigger:'reactive'`. `src/runtime/agent/orchestrator/session/loop.mjs:982`, `src/runtime/agent/orchestrator/session/loop.mjs:995`, `src/runtime/agent/orchestrator/session/loop.mjs:1002`\n\n- **Type choice:** policy chooses `semantic` or `recall-fasttrack`; agent-owned sessions are forced semantic, and loop dispatches recall first otherwise semantic. `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:39`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:123`, `src/runtime/agent/orchestrator/session/loop.mjs:414`, `src/runtime/agent/orchestrator/session/loop.mjs:439`\n\n- **Window prep:** engine first sanitizes tool pairs, dedups/reconciles tool results, splits protected system prefix, finds previous summary, and filters summaries out of live history. `src/runtime/agent/orchestrator/session/compact/engine.mjs:100`\n\n- **Semantic tail:** semantic keeps recent user-anchored turns within `preserveRecentBudget`; if no user turns exist, it cuts only at non-tool boundaries to keep tool pairs valid. `src/runtime/agent/orchestrator/session/compact/engine.mjs:134`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:193`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:235`\n\n- **Recall tail:** recall-fasttrack keeps the newest structured turns, drops oldest turns to fit cap, then truncates the oldest surviving content if needed. `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:606`\n\n- **Redaction hook:** actual compaction redacts tool-call argument secrets after the no-op fit check and 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\n- **Redaction behavior:** redaction only touches sensitive tool-call argument values or embedded raw secret pairs while preserving message/tool-call shape and order. `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:319`\n\n- **Semantic output:** semantic calls provider with a compact prompt, repairs malformed summaries, emits `[system, summary, tail]`, and rejects empty/no-fit/over-budget results. `src/runtime/agent/orchestrator/session/compact/engine.mjs:333`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:367`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:376`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:392`\n\n- **Recall output:** recall-fasttrack injects recall/prior text into a fitted summary, caps recall room separately, emits `[system, summary, tail]`, and rejects empty recall unless allowed. `src/runtime/agent/orchestrator/session/compact/engine.mjs:666`, `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\n- **Fallbacks:** loop semantic failures can fall back to unanchored tool-output pruning; recall-fasttrack failures do not use that prune fallback and escalate overflow. `src/runtime/agent/orchestrator/session/loop.mjs:500`, `src/runtime/agent/orchestrator/session/loop.mjs:517`, `src/runtime/agent/orchestrator/session/loop.mjs:618`\n\n- **Manual fallback:** session-runner recall-fasttrack failure can degrade to semantic once; non-recall failures can use the same unanchored prune fallback, otherwise error state is recorded. `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`",
-      "card": {
-        "session": "sess_51948_1783069931009_1783069931236_da461bb4146fd67343d59e15f87c3b45",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 59444,
-        "turns": 8,
-        "tool_calls": 11,
-        "tools_per_turn": 1.4,
-        "total_tool_ms": 917,
-        "llm_stream_ms": 58557,
-        "cache_ratio": 0.6344514338241921,
-        "cached_tokens": 140800,
-        "prompt_tokens": 221924,
-        "uncached_tokens": 81124,
-        "output_tokens": 4388,
-        "cache_weighted_input_10": 95204,
-        "cache_weighted_input_25": 116324,
-        "prompt_growth": 39319,
-        "antipatterns": 0,
-        "issues": 12,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "broad_tool_result",
-          "service_tier"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-debug-forwarding-gap#3",
-      "agent": "reviewer",
-      "ok": true,
-      "ms": 86359,
-      "sessionId": "sess_46144_1783069938017_1783069938263_a5dae16078bdeb9673d932cc9b9574a5",
-      "output": "sessionId: sess_46144_1783069938017_1783069938263_a5dae16078bdeb9673d932cc9b9574a5\n요청하신 경로를 기준으로 활성화부터 첫 전달까지의 흐름과 손실 구간을 바로 추적하겠습니다.조사 완료: `bridge-state.json`은 “활성 요청” 신호일 뿐, Discord 연결·소유권·transcript bind·watch 준비 전 창이 실제로 있습니다.\n\n- **기존 transcript EOF 스킵**: bind 시 `replayFromStart`가 아니면 현재 파일 크기로 cursor를 잡아 이미 쓰인 assistant 출력은 건너뜁니다. `src/runtime/channels/lib/output-forwarder.mjs:111` / `src/runtime/channels/index.mjs:624` — **backlog catch-up 불가**.\n\n- **미생성 transcript watch 실패**: `fs.watch`는 없는 파일에 실패하지만 re-arm poll이 생성 후 `startWatch()`와 `forwardNewText()`를 호출합니다. `src/runtime/channels/index.mjs:357` / `src/runtime/channels/lib/output-forwarder.mjs:647` — **60초 안 생성되면 복구, timeout 후에는 지연·추후 bind에서 드롭 가능**.\n\n- **소유권/연결 전 flush 무시**: `forwardNewText()`는 owner가 아니면 읽지 않고 반환하고, connect 후 재호출/재무장합니다. `src/runtime/channels/lib/output-forwarder.mjs:351` / `src/runtime/channels/index.mjs:801` — **보통 backlog catch-up 가능, 지연만 발생**.\n\n- **watch 설치 직전 write 경합**: watch 성공 시 즉시 flush를 예약해 stat 이후 쓰인 bytes를 회수합니다. `src/runtime/channels/lib/output-forwarder.mjs:665` — **backlog catch-up 가능**.\n\nVerdict: 가장 큰 실제 손실 위험은 “이미 존재하는 transcript를 activation 후 EOF로 bind하는 경로”이며, 미생성 transcript 창은 60초 re-arm 안에서는 복구됩니다.",
-      "card": {
-        "session": "sess_46144_1783069938017_1783069938263_a5dae16078bdeb9673d932cc9b9574a5",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 75840,
-        "turns": 5,
-        "tool_calls": 8,
-        "tools_per_turn": 1.6,
-        "total_tool_ms": 626,
-        "llm_stream_ms": 75046,
-        "cache_ratio": 0.4436709993912701,
-        "cached_tokens": 61952,
-        "prompt_tokens": 139635,
-        "uncached_tokens": 77683,
-        "output_tokens": 7952,
-        "cache_weighted_input_10": 83878.2,
-        "cache_weighted_input_25": 93171,
-        "prompt_growth": 35029,
-        "antipatterns": 0,
-        "issues": 8,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "service_tier"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-impl-usage-summary#3",
-      "agent": "worker",
-      "ok": true,
-      "ms": 119618,
-      "sessionId": "sess_47516_1783070000924_1783070001163_6b4f4d171e302f017c024a040f0956a6",
-      "output": "sessionId: sess_47516_1783070000924_1783070001163_6b4f4d171e302f017c024a040f0956a6\n요청 범위가 명확해서 바로 구현하고 검증까지 진행하겠습니다.구현 파일과 npm script가 이미 존재해 요구사항과 맞는지 확인한 뒤 검증을 실행하겠습니다.검증은 통과했고, diff 범위를 확인해 관련 변경만 요약하겠습니다.완료했습니다, 재영님. `usage:today` 스크립트와 `scripts/usage-today.mjs` 구현이 준비되어 검증까지 통과했습니다.\n\n- **Diff 요약**: `package.json`에 `usage:today` 추가, `scripts/usage-today.mjs` 신규 추가.\n\n- **검증**: `node --check scripts/usage-today.mjs` 통과, `npm run usage:today` 실행 성공.\n\n- **참고**: 작업 범위 밖 변경으로 `src/runtime/channels/lib/event-queue.mjs`, `src/runtime/channels/lib/output-forwarder.mjs` 수정분이 이미 보입니다.",
-      "card": {
-        "session": "sess_47516_1783070000924_1783070001163_6b4f4d171e302f017c024a040f0956a6",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 109123,
-        "turns": 14,
-        "tool_calls": 21,
-        "tools_per_turn": 1.5,
-        "total_tool_ms": 2822,
-        "llm_stream_ms": 106210,
-        "cache_ratio": 0.8669324056243677,
-        "cached_tokens": 328192,
-        "prompt_tokens": 378567,
-        "uncached_tokens": 50375,
-        "output_tokens": 10990,
-        "cache_weighted_input_10": 83194.20000000001,
-        "cache_weighted_input_25": 132423,
-        "prompt_growth": 30286,
-        "antipatterns": 0,
-        "issues": 14,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "cache_miss",
-          "readonly_stall",
-          "service_tier"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    },
-    {
-      "id": "r4-explain-compact-engine#3",
-      "agent": "explore",
-      "ok": true,
-      "ms": 71032,
-      "sessionId": "sess_39060_1783070024383_1783070024607_f04e5c1bfb6beb27bc3ac82d8dbc7a46",
-      "output": "runtime/agent/orchestrator/session/loop/compact-policy.mjs:89`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:92`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:100`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:123`).\n- **발화 조건**: pre-send loop는 `shouldCompactForSession()`가 true일 때만 compaction으로 진입한다 (`src/runtime/agent/orchestrator/session/loop.mjs:353`, `src/runtime/agent/orchestrator/session/loop.mjs:365`, `src/runtime/agent/orchestrator/session/loop.mjs:372`).\n- **임계값 로직**: reactive overflow retry는 무조건 compact, 토큰 추정 실패도 compact, 그 외에는 `messageTokens + reserve >= triggerTokens`가 조건이다 (`src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:170`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:172`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:173`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:174`).\n- **목표 예산**: compact target은 boundary 안에서 target effective + reserve로 계산된다 (`src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:163`, `src/runtime/agent/orchestrator/session/loop/compact-policy.mjs:168`).\n- **수동 runner 조건**: manual은 force로 간주되고, auto runner는 `pressureTokens < triggerTokens`면 “below threshold”로 종료한다 (`src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:145`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:146`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:176`).\n- **공통 전처리**: semantic/recall 모두 tool pair sanitize, tool result dedup, dedup stub reconcile을 먼저 수행한다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:284`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:625`).\n- **no-op 경로**: sanitized transcript가 budget 안에 있고 force가 아니면 원본 sanitize 결과를 그대로 반환한다 (`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:627`).\n- **redaction 위치**: 실제 compaction에 들어갈 때만 `redactToolCallSecretsInMessages()`를 적용하고, semantic은 window selection 전에 적용한다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:306`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:312`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:314`).\n- **redaction 보존**: semantic output은 redacted `system`/`tail`을 그대로 내보낸다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:389`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:392`).\n- **semantic window split**: protected/system prefix, previous summary, non-summary live transcript로 분리한다 (`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:103`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:110`).\n- **semantic tail 선택**: user turn이 있으면 최근 `tailTurns`를 budget 안에서 보존하고, 없으면 tool result가 아닌 cut point 기준으로 tail을 고른다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:240`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:241`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:242`).\n- **user-turn tail**: 최신 turn부터 거꾸로 확장하고, 최신 turn이 너무 크면 fitting suffix만 보존한다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:141`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:146`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:151`).\n- **no-user tail**: valid cut은 tool result로 시작하지 않으며, budget에 맞는 가장 오래된 suffix cut까지 확장한다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:121`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:124`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:197`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:200`).\n- **semantic 결과 조립**: `head`는 summary로 대체되고 `system + summary + tail`이 다시 sanitize/reconcile된다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:244`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:258`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:384`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:392`).\n- **recall tail 정책**: recall-fasttrack은 최신 structured turns를 verbatim 보존하고 오래된 history는 recall summary로 주입한다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:449`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:452`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:593`).\n- **recall tail 축소**: 최신 `maxUsers` turns를 고르고, cap 초과 시 오래된 turn을 버린 뒤 필요하면 단일 turn을 truncate한다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:596`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:601`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:602`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:610`).\n- **recall no-user 허용**: no-user tail도 유효하며 tool-pairing은 sanitize/reconcile로 복구된다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:613`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:617`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:618`).\n- **recall 입력**: loop path는 memory ingest/dump 결과와 cycle1 text를 합쳐 recall text를 만들고 token cap을 전달한다 (`src/runtime/agent/orchestrator/session/loop/recall-fasttrack.mjs:148`, `src/runtime/agent/orchestrator/session/loop/recall-fasttrack.mjs:157`, `src/runtime/agent/orchestrator/session/loop/recall-fasttrack.mjs:161`).\n- **semantic 실패 지점**: provider 부재, compactable head 부재, prompt fit 실패, empty summary, summary fit 실패, final budget 초과가 throw된다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:279`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:315`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:335`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:371`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:385`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:395`).\n- **recall 실패 지점**: compactable prior/recall text가 없거나 recall summary가 fit되지 않거나 final budget을 넘으면 throw된다 (`src/runtime/agent/orchestrator/session/compact/engine.mjs:653`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:666`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:691`, `src/runtime/agent/orchestrator/session/compact/engine.mjs:698`).\n- **loop fallback**: non-recall compact 실패는 unanchored tool-output prune으로 한 번 복구를 시도하고, recall 실패에는 이 prune fallback을 적용하지 않는다 (`src/runtime/agent/orchestrator/session/loop.mjs:501`, `src/runtime/agent/orchestrator/session/loop.mjs:511`, `src/runtime/agent/orchestrator/session/loop.mjs:517`, `src/runtime/agent/orchestrator/session/loop.mjs:527`).\n- **loop hard failure**: fallback도 없으면 `overflow_failed` telemetry를 남기고 `agentContextOverflowError()`를 throw한다 (`src/runtime/agent/orchestrator/session/loop.mjs:552`, `src/runtime/agent/orchestrator/session/loop.mjs:597`, `src/runtime/agent/orchestrator/session/loop.mjs:618`).\n- **runner degraded fallback**: manager runner에서 recall-fasttrack 실패 시 semantic이 enabled이고 provider가 있으면 semantic fallback을 한 번 시도한다 (`src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:237`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:242`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:245`).\n- **runner prune fallback**: non-recall 실패는 loop와 동일하게 unanchored prune으로 복구를 시도한다 (`src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:322`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:327`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:330`).\n- **성공 commit**: compacted transcript로 `session.messages`를 교체하고 providerState를 무효화한다 (`src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:410`, `src/runtime/agent/orchestrator/session/manager/compaction-runner.mjs:411`, `src/runtime/agent/orchestrator/session/loop.mjs:631`, `src/runtime/agent/orchestrator/session/loop.mjs:639`).",
-      "card": {
-        "session": "sess_39060_1783070024383_1783070024607_f04e5c1bfb6beb27bc3ac82d8dbc7a46",
-        "provider": "openai-oauth",
-        "model": "gpt-5.5",
-        "wall_ms": 60530,
-        "turns": 6,
-        "tool_calls": 7,
-        "tools_per_turn": 1.2,
-        "total_tool_ms": 741,
-        "llm_stream_ms": 59738,
-        "cache_ratio": 0.6792775828126337,
-        "cached_tokens": 79360,
-        "prompt_tokens": 116830,
-        "uncached_tokens": 37470,
-        "output_tokens": 4737,
-        "cache_weighted_input_10": 45406,
-        "cache_weighted_input_25": 57310,
-        "prompt_growth": 26262,
-        "antipatterns": 1,
-        "issues": 9,
-        "issue_types": [
-          "cache",
-          "cache_break",
-          "service_tier",
-          "tool_churn_cluster"
-        ],
-        "read_fragmentation_paths": [],
-        "grep_sweep_count": 0
-      }
-    }
-  ],
-  "task_complete": true,
-  "task_errors": [],
-  "score_complete": true,
-  "score_errors": [],
-  "group": {
-    "n": 9,
-    "wall_ms": 73819.2,
-    "turns": 7.9,
-    "tool_calls": 10.3,
-    "tools_per_turn": 1.2,
-    "total_tool_ms": 1229.1,
-    "llm_stream_ms": 72374.9,
-    "cache_ratio": 0.6,
-    "cached_tokens": 148423.1,
-    "prompt_tokens": 204076.7,
-    "uncached_tokens": 55653.6,
-    "output_tokens": 6546.6,
-    "cache_weighted_input_10": 70495.9,
-    "cache_weighted_input_25": 92759.3,
-    "prompt_growth": 30556.6,
-    "antipatterns": 0.6,
-    "issues": 11.2,
-    "grep_sweep_count": 0
-  },
-  "cards": [
-    {
-      "session": "sess_36552_1783069830153_1783069830396_d13a12bb2156913c8a60e9442598d0c8",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 90358,
-      "turns": 7,
-      "tool_calls": 6,
-      "tools_per_turn": 0.9,
-      "total_tool_ms": 315,
-      "llm_stream_ms": 89636,
-      "cache_ratio": 0.722787919558858,
-      "cached_tokens": 141824,
-      "prompt_tokens": 196218,
-      "uncached_tokens": 54394,
-      "output_tokens": 8930,
-      "cache_weighted_input_10": 68576.4,
-      "cache_weighted_input_25": 89850,
-      "prompt_growth": 38506,
-      "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_57952_1783069830168_1783069830408_4ec1cd0f3324c0260f3de8ee49f265d7",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 7228,
-      "turns": 1,
-      "tool_calls": 0,
-      "tools_per_turn": 0,
-      "total_tool_ms": 0,
-      "llm_stream_ms": 6268,
-      "cache_ratio": 0.2695871946082561,
-      "cached_tokens": 2560,
-      "prompt_tokens": 9496,
-      "uncached_tokens": 6936,
-      "output_tokens": 495,
-      "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_48676_1783069830186_1783069830428_0a5bfd5ec7bb592c4472884b507d14db",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 74277,
-      "turns": 9,
-      "tool_calls": 14,
-      "tools_per_turn": 1.6,
-      "total_tool_ms": 2165,
-      "llm_stream_ms": 72331,
-      "cache_ratio": 0.7512216810492917,
-      "cached_tokens": 185856,
-      "prompt_tokens": 247405,
-      "uncached_tokens": 61549,
-      "output_tokens": 5622,
-      "cache_weighted_input_10": 80134.6,
-      "cache_weighted_input_25": 108013,
-      "prompt_growth": 44952,
-      "antipatterns": 1,
-      "issues": 15,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "readonly_stall",
-        "broad_tool_result",
-        "service_tier",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_36256_1783069847899_1783069848122_f54e2a821584c50ac4e979c50a601646",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 79636,
-      "turns": 7,
-      "tool_calls": 6,
-      "tools_per_turn": 0.9,
-      "total_tool_ms": 493,
-      "llm_stream_ms": 78523,
-      "cache_ratio": 0.6874013304367773,
-      "cached_tokens": 101888,
-      "prompt_tokens": 148222,
-      "uncached_tokens": 46334,
-      "output_tokens": 6629,
-      "cache_weighted_input_10": 56522.8,
-      "cache_weighted_input_25": 71806,
-      "prompt_growth": 30328,
-      "antipatterns": 1,
-      "issues": 13,
-      "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_18484_1783069914981_1783069915210_7f0733a8893fa79343fc8708019b0acd",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 107937,
-      "turns": 14,
-      "tool_calls": 20,
-      "tools_per_turn": 1.4,
-      "total_tool_ms": 2983,
-      "llm_stream_ms": 105065,
-      "cache_ratio": 0.7753208965282127,
-      "cached_tokens": 293376,
-      "prompt_tokens": 378393,
-      "uncached_tokens": 85017,
-      "output_tokens": 9176,
-      "cache_weighted_input_10": 114354.6,
-      "cache_weighted_input_25": 158361,
-      "prompt_growth": 30327,
-      "antipatterns": 1,
-      "issues": 16,
-      "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_51948_1783069931009_1783069931236_da461bb4146fd67343d59e15f87c3b45",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 59444,
-      "turns": 8,
-      "tool_calls": 11,
-      "tools_per_turn": 1.4,
-      "total_tool_ms": 917,
-      "llm_stream_ms": 58557,
-      "cache_ratio": 0.6344514338241921,
-      "cached_tokens": 140800,
-      "prompt_tokens": 221924,
-      "uncached_tokens": 81124,
-      "output_tokens": 4388,
-      "cache_weighted_input_10": 95204,
-      "cache_weighted_input_25": 116324,
-      "prompt_growth": 39319,
-      "antipatterns": 0,
-      "issues": 12,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "broad_tool_result",
-        "service_tier"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_46144_1783069938017_1783069938263_a5dae16078bdeb9673d932cc9b9574a5",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 75840,
-      "turns": 5,
-      "tool_calls": 8,
-      "tools_per_turn": 1.6,
-      "total_tool_ms": 626,
-      "llm_stream_ms": 75046,
-      "cache_ratio": 0.4436709993912701,
-      "cached_tokens": 61952,
-      "prompt_tokens": 139635,
-      "uncached_tokens": 77683,
-      "output_tokens": 7952,
-      "cache_weighted_input_10": 83878.2,
-      "cache_weighted_input_25": 93171,
-      "prompt_growth": 35029,
-      "antipatterns": 0,
-      "issues": 8,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "service_tier"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_47516_1783070000924_1783070001163_6b4f4d171e302f017c024a040f0956a6",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 109123,
-      "turns": 14,
-      "tool_calls": 21,
-      "tools_per_turn": 1.5,
-      "total_tool_ms": 2822,
-      "llm_stream_ms": 106210,
-      "cache_ratio": 0.8669324056243677,
-      "cached_tokens": 328192,
-      "prompt_tokens": 378567,
-      "uncached_tokens": 50375,
-      "output_tokens": 10990,
-      "cache_weighted_input_10": 83194.20000000001,
-      "cache_weighted_input_25": 132423,
-      "prompt_growth": 30286,
-      "antipatterns": 0,
-      "issues": 14,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "cache_miss",
-        "readonly_stall",
-        "service_tier"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    },
-    {
-      "session": "sess_39060_1783070024383_1783070024607_f04e5c1bfb6beb27bc3ac82d8dbc7a46",
-      "provider": "openai-oauth",
-      "model": "gpt-5.5",
-      "wall_ms": 60530,
-      "turns": 6,
-      "tool_calls": 7,
-      "tools_per_turn": 1.2,
-      "total_tool_ms": 741,
-      "llm_stream_ms": 59738,
-      "cache_ratio": 0.6792775828126337,
-      "cached_tokens": 79360,
-      "prompt_tokens": 116830,
-      "uncached_tokens": 37470,
-      "output_tokens": 4737,
-      "cache_weighted_input_10": 45406,
-      "cache_weighted_input_25": 57310,
-      "prompt_growth": 26262,
-      "antipatterns": 1,
-      "issues": 9,
-      "issue_types": [
-        "cache",
-        "cache_break",
-        "service_tier",
-        "tool_churn_cluster"
-      ],
-      "read_fragmentation_paths": [],
-      "grep_sweep_count": 0
-    }
-  ]
-}