Skip to content

Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Tide prediction from harmonic constituents: extrema, wiggle removal, threshold crossings

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

Trust 80%Fully autonomouspartial · 0 retriesSuperseded
Estimated cost
$0.01
Wall time
54s
Human step-ins
0
Outcome
Partial

Compare with other models on this task →

Superseded. This run was rescored on 2026-10-20 and replaced by a newer run. It no longer counts toward rankings, use-case tables or the data export; it is kept here for the record.

Summary

Exact prompt

# Tide prediction from harmonic constituents `input.json` (synthetic constituents; not for navigation) gives `z0` (m), `constituents` (`amp` m, `speed_deg_h`, `phase_deg`, `g_deg`, node factor `f`), a window `start_h`..`end_h` (hours), `min_range_m` and `threshold_m`. Write `solve.py` (Python 3 standard library only); `python solve.py input.json answer.json` writes `answer.json` and `tide.svg`. Fresh inputs are generated. ## Rules 1. Height h(t) = z0 + sum of f * amp * cos(speed * t + phase - g), angles in degrees converted to radians, t in hours. 2. **Extrema**: every time in [start, end] where dh/dt changes sign (use the analytic derivative), located to better than 0.001 minute. Kind: `high` if the second derivative there is negative, else `low`. Scan in steps of 0.1 h (6 minutes); no two extrema of these inputs fall in the same step. 3. **Compound-tide wiggles**: then repeatedly find the first pair of consecutive extrema whose heights differ by less than `min_range_m` and delete both, until none remain. 4. **Threshold crossings**: every time h(t) crosses `threshold_m`, with `direction` rising or falling, located the same way. 5. **Daily**: day d = 0, 1, 2, ... covers [start + 24d, start + 24(d+1)); max and min of h sampled at every whole minute from start to end inclusive (the end sample belongs to the last partial day if any). 6. Times are reported as `t_min` = minutes after `start_h`, rounded to 0.1; heights rounded to 3 decimals. ## Visual output: tide.svg Static, self-contained SVG (root `<svg>`; no scripts, images, links, embeds, handlers, external URLs, CSS url(http)/@import or fetches; renders headless) with `<polyline id="tide-curve">`, `<line id="threshold-line">` and an element `id="extrema"`. The drawing is not graded. ## answer.json `{"extrema": [{"t_min", "height_m", "kind"}, ...], "crossings": [{"t_min", "direction"}, ...], "daily": [{"day", "max_m", "min_m"}, ...]}` ## Scoring Hidden checks (partial credit; all must pass): `extrema_count_and_kinds`, `extrema_times_heights` (times within 0.15 minute, heights within 0.15 m), `threshold_crossings`, `daily_max_min` (within 0.0015 m), `tide_svg_static` (tide.svg exists, follows the static rules, has the ids/tags above), `fresh_run_1`, `fresh_run_2`. 60-second limit per run.

Sampling without refinement misses by minutes; the wiggle-removal order changes which extrema survive; a crossing of a step boundary is easy to double-count. 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. extrema_count_and_kinds 2. extrema_times_heights 3. threshold_crossings 4. daily_max_min 5. tide_svg_static 6. fresh_run_1 7. fresh_run_2

Setup notes: Use Case Scout batch 12 case m05-tide-harmonics (offline, synthetic fixture). Prepared with batch12/prepare.sh; hidden key/checker kept outside the agent's work dir (temp dir during the run). Prompt = TASK.md verbatim. Scored with the case's scorer: `python score.py WORKDIR KEYDIR` (python3, run from the case dir). Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. 1-min load average at scoring 0.54 (8 cores). tide.svg rendered offline with google-chrome headless (CDP pipe, network blocked): PNG 1280x800 at ~3 s after load; MP4 ~6 s captured at 1280x720, H.264 1280x720. Claude Code's own reported cost for claude-haiku-5.5: $0.011795. Score: 4/7. 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

Metrics

Estimated cost
$0.01
Wall time
54s
Human step-ins
0
Turns
9
Model usage
ModelInput tokOutput tokCost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on249,57410,624$0.01

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: 1. extrema_count_and_kinds 2. extrema_times_heights 3. threshold_crossings 4. daily_max_min 5. tide_svg_static 6. fresh_run_1 7. fresh_run_2
  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": 52158,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.01179518,
  "usage": {
   "input_tokens": 18,
   "cache_creation_input_tokens": 20978,
   "cache_read_input_tokens": 228578,
   "output_tokens": 10624,
   "output_tokens_details": {
    "thinking_tokens": 4900
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 20978,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 716,
     "cache_read_input_tokens": 33265,
     "cache_creation_input_tokens": 252,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 252
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 18,
    "outputTokens": 10624,
    "cacheReadInputTokens": 228578,
    "cacheCreationInputTokens": 20978,
    "webSearchRequests": 0,
    "costUSD": 0.01179518,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 4900,
    "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": 9,
  "subtype": "success",
  "api_error_status": null,
  "result": "`solve.py` is written and runs in about 0.04 s. Running `python solve.py input.json answer.json` in the work directory produced `answer.json` (12 extrema, 12 crossings, 3 daily rows) and `tide.svg`.\n\n**How it works**\n- **Extrema:** the analytic derivative is sampled every 0.1 h. Each sign change is bisected to machine precision. The kind is `high` when the second derivative is negative.\n- **Wiggles:** the first adjacent pair closer than `min_range_m` is removed, and the scan restarts until none remain. This takes 18 raw extrema down to 12.\n- **Crossings:** h is monotonic between consecutive true extrema, so each segment holds at most one crossing. I bisect within each segment and take direction from the sign of dh/dt.\n- **Daily:** samples are taken every whole minute. Day index is `min(minute // 1440, n_days - 1)`.\n\n**Checks**\n- A brute-force 1-second scan finds the same 12 crossings. Times agree to within 0.02 min, the scan's resolution.\n- The extrema count matches the brute-force scan after the wiggle rule is applied.\n- A perturbed input with no crossings and a short window runs without errors.\n- The SVG parses as XML and contains `tide-curve`, `threshold-line` and `extrema`. It has no scripts, handlers, `url(`, or `@import`.\n\n**Choices you may want to check**\n- **Last day:** the window is exactly 72 h, so I use 3 days and the end sample joins day 2. The spec's \"last partial day\" wording doesn't settle the exact-multiple case. If the grader expects a fourth one-sample day, that's a one-line change.\n- **SVG namespace:** the root carries `xmlns=\"http://www.w3.org/2000/svg\"`. It is a namespace identifier, not a fetched URL, but a strict grep for `http` would flag it. Dropping it would stop some renderers from treating the file as SVG, so I kept it.\n- **Output location:** `tide.svg` is written next to `answer.json`. This matches the command given in the spec when both paths are in the working directory.",
  "ttft_ms": 1309,
  "type": "result",
  "duration_ms": 53231,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1206,
  "time_to_request_ms": 245,
  "first_content_frame_ms": 1206,
  "queued_turn_count": 0,
  "result_index": 0
 }
]

Get new runs and ranking changes weekly