Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Schedule a 40-task build DAG on 3 workers with exclusive resources

AClaude Haiku 5.5default · thinking on · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-16 · by @bobsiopener

Trust 80%Fully autonomoussuccess · first tryCompare

Estimated cost: $0.0062
Wall time: 2m
Human step-ins: 0
Outcome: Pass

Compare with other models on this task →

Exact prompt

# Task: schedule a build `tasks.json` describes 40 build tasks: duration (minutes), dependencies, and resources each task holds while running. There are `workers` identical workers (each runs one task at a time) and resource limits (at most `db: 1` and `net: 2` tasks holding that resource concurrently). Tasks can't be preempted; a task may start only when all its deps have finished (start ≥ max dep end time). Produce `schedule.json`: {"t00": {"start": <int minute>, "worker": <0..workers-1>}, ...} for ALL tasks, minimising the makespan (time the last task finishes). Also write `plan_notes.md` with your makespan and the lower bound you computed (max of critical path and total work / workers). You may write code. No network.

Validator checks deps, worker overlap, resource limits; must match or beat critical-path list scheduling (makespan 98; lower bound 91). 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: 1. schedule.json must contain exactly the 40 tasks, each with a non-negative integer start and a valid worker index. 2. No task starts before all its dependencies finish. 3. No two tasks overlap on the same worker. 4. At no start time do concurrent tasks exceed the db (1) or net (2) resource limits. 5. Makespan <= the reference critical-path list-scheduling makespan.

Setup notes: Use Case Scout batch 4 case d12-build-scheduler (offline fixture). Prepared with batch4/prepare.sh; hidden key/tests kept outside the agent's work dir. Prompt = TASK.md verbatim. Scored with the scout's scorer: `python score.py WORKDIR KEYDIR/key.json`. Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. Claude Code's own reported cost for claude-haiku-5.5: $0.006229. Score: pass. 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: $0.0062
Wall time: 2m
Human step-ins: 0
Turns: 6

Model usage
Model	Input tok	Output tok	Cost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on	138,280	4,491	$0.0062

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: 1. schedule.json must contain exactly the 40 tasks, each with a non-negative integer start and a valid worker index. 2. No task starts before all its dependencies finish. 3. No two tasks overlap on the same worker. 4. At no start time do concurrent tasks exceed the db (1) or net (2) resource limits. 5. Makespan <= the reference critical-path list-scheduling makespan.
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": 24464,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.006228959999999999,
  "usage": {
   "input_tokens": 12,
   "cache_creation_input_tokens": 13682,
   "cache_read_input_tokens": 124586,
   "output_tokens": 4491,
   "output_tokens_details": {
    "thinking_tokens": 731
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 13682,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 431,
     "cache_read_input_tokens": 25476,
     "cache_creation_input_tokens": 745,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 745
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 12,
    "outputTokens": 4491,
    "cacheReadInputTokens": 124586,
    "cacheCreationInputTokens": 13682,
    "webSearchRequests": 0,
    "costUSD": 0.006228959999999999,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 731,
    "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": 6,
  "subtype": "success",
  "api_error_status": null,
  "result": "I wrote `schedule.json` and `plan_notes.md` in the work directory. The schedule has a makespan of **95 minutes**, against a lower bound of **91**. It is not proven optimal.\n\n**Lower bound (91):**\n- Critical path: 81 minutes.\n- Total work / workers: 273 / 3 = 91.\n- Resource bounds: db 88 (sum of db durations / 1), net 63 (sum of net durations / 2). These are weaker than 91, so the bound is 91.\n\n**Method:** I wrote an event-driven list scheduler. A task starts only when its dependencies have finished, a worker is free, and the db/net limits allow it. Priority is the longest downstream path. I ran about 20,000 randomized priority perturbations, then a swap-based local search. Nothing got below 95, and a further randomized search found nothing either. The script is in `/tmp/sched/solve.py`, outside the work directory.\n\n**Validation:** I checked every schedule against the constraints: all 40 tasks are present, no task starts before its dependencies end, no two tasks overlap on the same worker, and db/net concurrency never exceeds 1 and 2.\n\nThe 4-minute gap may be real or may be a search limit. Closing it would need an exact solver such as CP-SAT, which I did not run.",
  "ttft_ms": 1637,
  "type": "result",
  "duration_ms": 96793,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1396,
  "time_to_request_ms": 205,
  "first_content_frame_ms": 1396,
  "queued_turn_count": 0,
  "result_index": 0
 }
]