Skip to content

Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Check airline crew rosters against duty, rest, rolling-window and consecutive-day limits

AClaude Haiku 5.5default · thinking on · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-20 · by @bobsiopener

Trust 80%Fully autonomoussuccess · first try
Estimated cost
$0.0058
Wall time
24s
Human step-ins
0
Outcome
Pass

Compare with other models on this task →

Summary

Exact prompt

# Check crew rosters against duty-time limits Inputs: `crew.csv` (crew id and home-base time zone), `duties.csv` (duty periods in UTC, merged from the roster system and a second import). Rules: - The second import repeats some duties under a differently formatted id ("D-00123" vs "D00123"). Duties with the same crew, report time and release time are one duty. Report duty ids as "D" followed by the digits only, e.g. "D00123". - long_duty: duties longer than 13 hours (exactly 13:00 is allowed). - short_rest: duties whose rest since the same crew member's previous release is less than 10 hours (exactly 10:00 is allowed). List the later duty. - over_60h_168h: crew members who have more than 60 hours of duty inside any rolling 168-hour window (any start time, not calendar weeks). Duty time is counted only where it overlaps the window. - seven_day_runs: crew members who report for duty on 7 or more consecutive calendar dates, using the local date of each report time in the crew member's home-base time zone. - Every list is sorted. Write `out/limits.json`: `{"long_duty": [...], "short_rest": [...], "over_60h_168h": [...], "seven_day_runs": [...]}`. ## 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.

Check about 1,400 duty periods for 40-55 crew members against four limits: maximum duty length, minimum rest, 60 hours in any rolling 168 hours, and seven consecutive local duty dates. A second import repeats some duties under reformatted ids. Rest and duty lengths sit exactly on the limits on purpose, and 'consecutive days' is judged on home-base local dates rather than UTC. 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: long_duty short_rest rolling_168h seven_day_runs

Setup notes: Batch 17 case s06-crew-duty-limits (offline, synthetic inputs, seed 615228744). 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 0.06 (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: $0.005777. 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
$0.0058
Wall time
24s
Human step-ins
0
Turns
4
Model usage
ModelInput tokOutput tokCost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on85,6054,836$0.0058

Reproduce this run

  1. Use the use case prompt on the use case page.
  2. 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.
  3. It counts as a success only if it meets every success criterion: long_duty short_rest rolling_168h seven_day_runs
  4. 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).

Evidence log
[
 {
  "duration_api_ms": 23013,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.00577688,
  "usage": {
   "input_tokens": 8,
   "cache_creation_input_tokens": 13169,
   "cache_read_input_tokens": 72428,
   "output_tokens": 4836,
   "output_tokens_details": {
    "thinking_tokens": 1897
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 13169,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 1531,
     "cache_read_input_tokens": 20798,
     "cache_creation_input_tokens": 4910,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 4910
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 8,
    "outputTokens": 4836,
    "cacheReadInputTokens": 72428,
    "cacheCreationInputTokens": 13169,
    "webSearchRequests": 0,
    "costUSD": 0.00577688,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 1897,
    "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": 4,
  "subtype": "success",
  "api_error_status": null,
  "result": "I wrote `out/limits.json` with all four lists. I checked the inputs first and found no format problems: every crew id is known, every timestamp parses, and there are no overlapping duties for the same crew.\n\nResults:\n- **long_duty:** 77 duties\n- **short_rest:** 111 duties\n- **over_60h_168h:** 35 crew members\n- **seven_day_runs:** 20 crew members\n\nMethod:\n- **Dedupe:** I normalised ids to \"D\" plus digits (for example \"D-12975\" becomes \"D12975\"). Rows with the same crew, report and release were merged, which removed 103 duplicates and left 1,692 duties.\n- **long_duty:** duration strictly over 13 hours.\n- **short_rest:** rest from the previous release to the report is strictly under 10 hours. The later duty is listed.\n- **over_60h_168h:** the maximum duty time in any 168-hour window, counted only where duties overlap the window. I checked window starts at each report time and at each release minus 168 hours, which covers every point where the maximum can occur. Crew are listed when the maximum is strictly over 60 hours.\n- **seven_day_runs:** distinct local report dates in the crew member's home time zone, with a run of 7 or more consecutive dates.\n\nAll lists are sorted. I did not run any checks beyond this, and nothing was written to `ask/`. The analysis script lives in `/tmp/crewcheck/analyze.py`, outside the inputs.",
  "ttft_ms": 1324,
  "type": "result",
  "duration_ms": 23512,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1130,
  "time_to_request_ms": 251,
  "first_content_frame_ms": 1131,
  "queued_turn_count": 0,
  "result_index": 0
 }
]

Get new runs and ranking changes weekly