Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Diagnose why a PC crashes 30 to 60 seconds into a game from its logs, drivers, temperatures and power readings

AClaude Haiku 5.5default · thinking on · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-25 · by @bobsiopener

Trust 80%Fully autonomoussuccess · first tryCompare

Estimated cost: $0.0075
Wall time: 21s
Human step-ins: 0
Outcome: Pass

Compare with other models on this task →

Exact prompt

# Diagnose why a PC crashes 30 to 60 seconds into a game from its logs, drivers, temperatures and power readings A synthetic home PC keeps crashing 30 to 60 seconds after a game starts. Its exported diagnostics are in `logs/`: `events.log` (system and application events, one per line), `drivers.txt` (installed driver versions), `known-issues.txt` (the GPU vendor's advisory with affected driver ranges), `hardware.txt` (component limits), `sensors.csv` (1-second samples: time, gpu_temp_c, gpu_hotspot_c, cpu_temp_c), `psu.csv` (half-second samples: time, rail_12v), `memory.txt` (memory profile and test results) and `overlays.txt` (overlay programs and their module files). All of it is synthetic; nothing is fetched or changed. Where `friend-advice.txt` disagrees with these rules, the rules win. Rules: - Crashes: each crash is an event that says the game stopped or the system shut down unexpectedly; its crash time is the time the event gives for the stop (for an unexpected shutdown, the "previous system shutdown at" time). - Signals: gpu_driver = a display driver reset event, with the installed GPU driver version inside an affected range in known-issues.txt; thermal = gpu_hotspot_c at or above hardware.txt hotspot_limit_c; psu_transient = rail_12v below 11.40; memory_xmp = a WHEA memory error event, with memory.txt showing XMP enabled and the memory test failing at the XMP profile; overlay_conflict = a game crash event whose faulting module is an overlay module listed in overlays.txt. - Root cause: the one cause whose signal occurs in the 10 seconds up to and including every crash time (each crash must have it). A signal seen before only some crashes, or outside those windows, is a decoy and is not the cause. - Evidence: cite lines as `logs/<file>:<line number>` (1-based). For every crash, cite at least one line of the root cause's signal inside that crash's window; for gpu_driver also cite the drivers.txt line with the GPU driver version, and for memory_xmp the memory.txt line with the failed XMP test. You may also cite the crash event lines. Cite nothing else. - Fix: gpu_driver → `update_gpu_driver:<the fixed version named in known-issues.txt>`; thermal → `clean_gpu_cooling`; psu_transient → `replace_psu`; memory_xmp → `disable_xmp`; overlay_conflict → `disable_overlay:<overlay name from overlays.txt>`. - Write `out/diagnosis.json`: `{"root_cause": "<one of the five causes>", "evidence": ["logs/<file>:<line>", ...], "fix": "<fix>"}`. ## How this run works - Work on your own: nobody will answer questions in this task, and writing to `ask/` counts against the run. - Write only under `out/`. Create `out/` if it is missing. Inputs are synthetic; don't fetch anything from the network. - Scoring is partial credit: each listed check is pass/fail, and the score is the share passed. Numbers are compared to within 0.005 unless the task says otherwise.

Find the one root cause of a synthetic PC's repeated in-game crashes from event logs, driver versions and a vendor advisory, temperature and 12 V rail samples, memory test results and the overlay list. The cause is the signal seen in the 10 seconds before every crash; decoys appear before only some crashes or in other sessions. The answer names the cause, cites the supporting log lines and gives the matching fix. A friend's advice file points the wrong way. Original benchmark task written for usecaserank.com (not based on a specific public post). Headless, offline run on a reproducible fixture. Each run's notes say how its cost was measured (estimates are marked).

Success criteria: root_cause fix evidence_covers_crashes evidence_clean

Setup notes: Batch 37 case ty04-pc-game-crash-diagnosis (offline, synthetic inputs, seed 638318684). Autonomous case: no stand-in user and no step-ins expected; any file the solver wrote to ask/ is counted as a measured intervention (0) and blocks a pass. Prompt = TASK.md verbatim. Hidden key kept outside the solver's work dir. Scored with the case's scorer (python3 score.py WORKDIR KEYDIR, absolute paths); pass = every check true and no ask/ files. Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. 1-min load average at scoring 0.13 (8 cores). Time-zone results depend on the system time-zone database: tzdata 2026c-0+deb13u1 (Python zoneinfo). Claude Code's own reported cost for claude-haiku-5.5: $0.007482. Score: 4/4. Cost is an ESTIMATE: measured tokens x pricing.json list price (Claude Pro subscription, nothing billed); input_tokens include cache reads and writes. Single run.

Estimated cost List-price estimate; not actually billed.

Metrics

Estimated cost: $0.0075
Wall time: 21s
Human step-ins: 0
Turns: 12

Model usage
Model	Input tok	Output tok	Cost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on	143,812	4,199	$0.0075

Reproduce this run

Use the use case prompt on the use case page.
Same setup: Claude Haiku 5.5 (anthropic/claude-haiku-5.5, effort default (no --effort flag) · thinking on) via Claude Code 2.1.299 (headless -p), cloud.
It counts as a success only if it meets every success criterion: root_cause fix evidence_covers_crashes evidence_clean
Log model, tokens, wall time, human step-ins and the evidence log, then submit it as a reproduction. A different model? Submit a head-to-head rerun.

How runs are scored: Methodology.

Was this useful?

⇄ Rerun on another model

Automated checks

Trust 80% comes from these checks (how Trust % works).

Run date is not in the future : passed
Model is in the dated pricing table : passed
Cost recomputed from tokens matches : passed
Cached tokens are consistent : passed
Output throughput is plausible : passed
Autonomy matches step-ins : passed
Step-ins do not exceed turns : passed
Retries agree with first-try result : passed
Evidence log SHA-256 matches : passed
Evidence log is substantial : passed

Evidence log

[
 {
  "duration_api_ms": 20047,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.00748178,
  "usage": {
   "input_tokens": 10,
   "cache_creation_input_tokens": 20754,
   "cache_read_input_tokens": 123048,
   "output_tokens": 4199,
   "output_tokens_details": {
    "thinking_tokens": 2088
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 20754,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 589,
     "cache_read_input_tokens": 32422,
     "cache_creation_input_tokens": 871,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 871
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 10,
    "outputTokens": 4199,
    "cacheReadInputTokens": 123048,
    "cacheCreationInputTokens": 20754,
    "webSearchRequests": 0,
    "costUSD": 0.00748178,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 2088,
    "canonicalModel": "claude-haiku-5-5",
    "provider": "firstParty",
    "costBasis": "list"
   }
  },
  "permission_denials": [],
  "terminal_reason": "completed",
  "fast_mode_state": "off",
  "fast_mode_disabled_reason": "sdk_opt_in_required",
  "subagent_stats": {
   "spawned": 0,
   "requested": {
    "background": 0,
    "foreground": 0,
    "unset": 0
   },
   "started_in_background": 0,
   "max_depth": 0,
   "spawned_by_subagents": 0,
   "completed": 0,
   "failed": 0,
   "killed": {
    "parent": 0,
    "user": 0,
    "system": 0
   },
   "refused": {
    "depth_limit": 0,
    "concurrency_limit": 0,
    "budget": 0
   },
   "by_type": {}
  },
  "safety_stops": 0,
  "is_error": false,
  "num_turns": 12,
  "subtype": "success",
  "api_error_status": null,
  "result": "The root cause is `overlay_conflict`, and the fix is to disable the ClipKeeper overlay. I wrote the result to `out/diagnosis.json`:\n\n```json\n{\"root_cause\": \"overlay_conflict\", \"evidence\": [\"logs/events.log:5\", \"logs/events.log:7\", \"logs/events.log:9\"], \"fix\": \"disable_overlay:ClipKeeper\"}\n```\n\n- **Crashes:** three game.exe crash events, at 2026-10-21 19:16:27, 2026-10-22 19:23:48 and 2026-10-23 22:58:14 (`events.log` lines 5, 7, 9). Each game started 30 to 60 seconds before its crash.\n- **Overlay signal:** all three crash events name `ckoverlay64.dll` as the faulting module. `overlays.txt` lists that file under ClipKeeper. The crash event lines are the signal, and they fall inside their own windows.\n- **Ruled out:**\n  - **gpu_driver:** no display driver reset event in `events.log`. The installed driver 556.33 is also outside the affected range 556.22 to 556.27.\n  - **thermal:** hotspot reaches 110 \u00b0C or more only during the 2026-10-20 bench run, not in any crash window.\n  - **psu_transient:** rail_12v drops below 11.40 only once, at 2026-10-22 19:23:44 to 45, which is a single crash. The other two crashes have no low-rail samples.\n  - **memory_xmp:** there are no WHEA events, and the memory test passes at the XMP profile.\n\nI cited only the crash event lines, since the rules list no other evidence for this cause. The fixed-version rule for gpu_driver did not apply, and `friend-advice.txt` was not used.",
  "ttft_ms": 1212,
  "type": "result",
  "duration_ms": 20365,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 966,
  "time_to_request_ms": 191,
  "first_content_frame_ms": 966,
  "queued_turn_count": 0,
  "result_index": 0
 }
]