Skip to content

Claude Opus 5.5 (effort default (no --effort flag) · thinking on) on Best fishing windows from tide, solunar and weather data

AClaude Opus 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
$0.16
Wall time
31s
Human step-ins
0
Outcome
Pass

Compare with other models on this task →

Summary

Exact prompt

# Fishing trip planner from tide, solunar and weather data `input.json` (synthetic) covers 48 hours, hour index h = 0..47 meaning the hour [h, h+1). It has `tide_m` (49 hourly heights at h = 0..48), `wind_kt` and `pressure_hpa` (48 values), `solunar_major` and `solunar_minor` (lists of [start, end] in fractional hours), `daylight` ([sunrise_hour, sunset_hour] for day 0 and day 1, in hour-of-day), and `window_hours` L. Write `solve.py` (Python 3 standard library only); `python solve.py input.json answer.json` writes `answer.json` and `windows.svg`. Fresh inputs are used too. ## Hourly score - If wind > 20 kt the hour is **unfishable**: score null, ignore the rest. - +3 if any major period overlaps the hour (start < h+1 and end > h); +1.5 if any minor period overlaps (both can apply). - Tide movement d = |tide[h+1] - tide[h]|: +2 if d >= 0.3, else +1 if d >= 0.15. - Wind above 10 kt (and <= 20): -1. - Pressure trend dp = pressure[h] - pressure[h-3] (0 for h < 3), rounded to 6 decimals: +1 if dp <= -1.0, -1 if dp >= +1.0. - Outside daylight: -2, where daylight means sunrise <= (h mod 24) < sunset for day h // 24. Round each score to 2 decimals. ## Windows A window starting at s covers hours s..s+L-1, all fishable; its score is the sum, rounded to 2 decimals. Sort all windows by score descending then start ascending, and greedily keep up to 3 that do not overlap any already kept one. ## answer.json `{"hourly_scores": [48 values or null], "unfishable_hours": n, "windows": [{"start", "end" (= start + L), "score"}, ...]}` ## Visual output: windows.svg Static, self-contained SVG (root `<svg>`; no `<script>`, `<img>`/`<image>`, `<link>`, `<iframe>`/`<object>`/`<embed>`, `on*` handlers, `href`/`src` except `#...` or `data:`, CSS `url(http...)`/`@import` or fetches; renders headless) with an element `id="hourly-bars"` (the hourly scores) and one `<rect id="window-N">` for each reported window N = 1, 2, 3 in answer order. The drawing itself is not graded. ## Scoring Hidden checks (partial credit; all must pass): `hourly_scores_day1` (hours 0-23), `hourly_scores_day2` (24-47 plus unfishable count), `top_windows`, `windows_svg_static` (windows.svg exists, follows the static rules, has the ids/tags above), `fresh_run_1`, `fresh_run_2`. Tolerance 0.001. 60-second limit per run.

Overlap tests on fractional periods, threshold boundaries placed exactly at 10 and 20 kt, a 3-hour pressure trend with float rounding, daylight by day index, and greedy non-overlapping window selection with ties. 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. hourly_scores_day1 2. hourly_scores_day2 3. top_windows 4. windows_svg_static 5. fresh_run_1 6. fresh_run_2

Setup notes: Use Case Scout batch 10 case k03-fishing-window-planner (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 (windows.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-opus-5.5: $0.155952. Score: 6/6. 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
$0.16
Wall time
31s
Human step-ins
0
Turns
3
Model usage
ModelInput tokOutput tokCost
AClaude Opus 5.5anthropic/claude-opus-5.5effort default (no --effort flag) · thinking on60,6933,139$0.16

Reproduce this run

  1. Use the use case prompt on the use case page.
  2. Same setup: Claude Opus 5.5 (anthropic/claude-opus-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. hourly_scores_day1 2. hourly_scores_day2 3. top_windows 4. windows_svg_static 5. fresh_run_1 6. 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": 29606,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.1559522,
  "usage": {
   "input_tokens": 6,
   "cache_creation_input_tokens": 10386,
   "cache_read_input_tokens": 50301,
   "output_tokens": 3139,
   "output_tokens_details": {
    "thinking_tokens": 315
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 10386,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 744,
     "cache_read_input_tokens": 19688,
     "cache_creation_input_tokens": 2844,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 2844
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-opus-5-5": {
    "inputTokens": 6,
    "outputTokens": 3139,
    "cacheReadInputTokens": 50301,
    "cacheCreationInputTokens": 10386,
    "webSearchRequests": 0,
    "costUSD": 0.1559522,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 315,
    "canonicalModel": "claude-opus-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": 3,
  "subtype": "success",
  "api_error_status": null,
  "result": "I wrote `solve.py` and ran it on `input.json`. It wrote `answer.json` and `windows.svg` without errors. I checked one hour (h=19) by hand and it matched. I didn't run the hidden checks or try any fresh inputs.\n\n**Results for this input:**\n- **Unfishable hours:** 2 (h=33 at 24.5 kt and h=46 at 25.5 kt). h=36 is exactly 20.0 kt, so it still counts as fishable, since only winds above 20 rule an hour out.\n- **Top windows (L = 3):**\n  1. start 6, end 9, score 12.0\n  2. start 29, end 32, score 10.0\n  3. start 11, end 14, score 8.0\n\nThe window starting at 7 scores 9.0, but it overlaps the first window, so it's skipped.\n\n**Choices I made where the spec left room:**\n- **Tide threshold:** the tide difference is rounded to 6 decimals before comparing it to 0.3 and 0.15, like the pressure trend. This stops floating-point error from pushing a difference that is exactly 0.3 below the threshold.\n- **Windows:** any window containing an unfishable hour is excluded.\n- **Missing data:** if the solunar lists are absent, they're treated as empty. If an hour falls on a day with no daylight entry, it uses the last day's sunrise and sunset.\n\n**windows.svg:** it's a plain bar chart with `<g id=\"hourly-bars\">` and the window rectangles `window-1` to `window-3`. Unfishable hours are greyed out. It has no scripts, links or external references.",
  "ttft_ms": 2745,
  "type": "result",
  "duration_ms": 29999,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 2002,
  "time_to_request_ms": 267,
  "first_content_frame_ms": 2002,
  "queued_turn_count": 0,
  "result_index": 0
 }
]

Get new runs and ranking changes weekly