Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Water-quality sensor anomaly flags and daily means

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

Trust 80%Fully autonomoussuccess · first tryCompare

Estimated cost: $0.0094
Wall time: 42s
Human step-ins: 0
Outcome: Pass

Compare with other models on this task →

Exact prompt

# Water-quality sensor anomaly flags `input.json` (synthetic buoy data) names `readings.csv` (`ts` "YYYY-MM-DDTHH:MM", `sensor`, `value`; rows are shuffled) and gives `stuck_n`, `window`, `spike_k`, and per-sensor `config` (`min`, `max`, `spike_floor`, `max_step`). Write `solve.py` (Python 3 standard library only); `python solve.py input.json answer.json` writes `answer.json` and `chart.svg`. Fresh inputs are used too. ## Rules (per sensor, rows in ascending ts) 1. **missing**: value is blank or not a finite number (e.g. `err`, `nan`). It is skipped entirely (it does not break or extend a stuck run). 2. For other values, track the run of consecutive non-missing readings with exactly the same value (all flags included). Then the first match wins: - **range**: value < min or > max. - **stuck**: the current run length >= `stuck_n`. - **spike**: at least `window` ok values exist; with the last `window` ok values, med = median and MAD = median of |x - med|; flag if |v - med| > max(spike_k * MAD, spike_floor). - **rate**: |v - last ok value| > max_step. - else **ok**. Only ok values are used as "ok values", in daily means and in the chart. 3. `daily_means[sensor][date]` = mean of ok values that day (3 decimals); `counts[sensor]` has keys ok, missing, range, stuck, spike, rate; `flags` lists every non-ok reading `{"ts", "sensor", "flag"}` sorted by ts then sensor. ## Visual output: chart.svg Static, self-contained SVG (root `<svg>`, no scripts, images, links or external assets; renders headless) with `<polyline id="series-<sensor>">` for each sensor (`series-temp_c`, `series-do_mgl`, `series-ph`) and an element `id="flag-summary"`. The drawing itself is not graded. ## answer.json `{"counts": {...}, "daily_means": {...}, "flags": [...]}` ## Scoring Hidden checks (partial credit; all must pass): `missing_and_range`, `stuck_detection`, `spike_and_rate` (with ok counts), `flag_list`, `daily_means` (within 0.0011), `chart_svg_static` (chart.svg exists, follows the static rules, has the ids/tags above), `fresh_run_1`, `fresh_run_2`. 60-second limit per run.

Shuffled rows, a stuck run that ignores missing values but includes flagged ones, a robust median/MAD spike test with a floor that only uses ok history, a rate test against the last ok value, and flag precedence. 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).

Benchmark code for comparing AI models, not navigation or safety software.

Success criteria: 1. missing_and_range 2. stuck_detection 3. spike_and_rate 4. flag_list 5. daily_means 6. chart_svg_static 7. fresh_run_1 8. fresh_run_2

Setup notes: Use Case Scout batch 10 case k12-water-quality-anomalies (offline fixture). Prepared with batch10/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`. Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. Visual output (chart.svg) rendered offline to PNG with google-chrome headless, network blocked, 1280x800 at ~3 s after load. Claude Code's own reported cost for claude-haiku-5.5: $0.009391. Score: 8/8. 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.

Output

Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Water-quality sensor anomaly flags and daily meansSuccess

Metrics

Estimated cost: $0.0094
Wall time: 42s
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	120,798	9,465	$0.0094

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. missing_and_range 2. stuck_detection 3. spike_and_rate 4. flag_list 5. daily_means 6. chart_svg_static 7. fresh_run_1 8. fresh_run_2
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": 41378,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.00939121,
  "usage": {
   "input_tokens": 10,
   "cache_creation_input_tokens": 18157,
   "cache_read_input_tokens": 102631,
   "output_tokens": 9465,
   "output_tokens_details": {
    "thinking_tokens": 3660
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 18157,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 1155,
     "cache_read_input_tokens": 28673,
     "cache_creation_input_tokens": 2023,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 2023
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 10,
    "outputTokens": 9465,
    "cacheReadInputTokens": 102631,
    "cacheCreationInputTokens": 18157,
    "webSearchRequests": 0,
    "costUSD": 0.00939121,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 3660,
    "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 `solve.py` in the work directory. It runs in about 0.06 s on the provided data, and the output checks out.\n\nResults on `readings.csv`:\n- **Counts:** each sensor's counts sum to its 144 readings. `temp_c` has ok 134, missing 3, range 1, stuck 2, spike 4, rate 0. `do_mgl` has ok 131, missing 3, range 1, stuck 3, spike 3, rate 3. `ph` has ok 135, missing 2, range 1, stuck 3, spike 2, rate 1.\n- **Flags:** 32 entries, which equals the total non-ok count across the three sensors.\n- **chart.svg:** has `series-temp_c`, `series-do_mgl`, `series-ph` polylines and a `flag-summary` text element, with no scripts or external references.\n\nI haven't run it against a fresh input or checked it against the hidden scoring.\n\nChoices I made where the spec was open:\n- **Missing in flags:** missing readings appear in `flags` with `\"flag\": \"missing\"`, since the spec says the list holds every non-ok reading. If the grader expects only range/stuck/spike/rate, drop the `missing` entries from the list.\n- **Chart location:** `chart.svg` is written next to `answer.json`.\n- **Readings path:** resolved relative to the directory containing `input.json`, falling back to the path as given.\n- **Unknown sensors:** a sensor in the CSV with no config entry is ignored in the output.\n\nI haven't committed anything; this directory is not a git repository.",
  "ttft_ms": 1902,
  "type": "result",
  "duration_ms": 41894,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1662,
  "time_to_request_ms": 254,
  "first_content_frame_ms": 1662,
  "queued_turn_count": 0,
  "result_index": 0
 }
]