Go/no-go departure time from a gridded wind file
Exact prompt
# Go/no-go departure time from a gridded wind file `input.json` (synthetic; not a forecast) names `grid.json` and gives `from` and `to` [lat, lon], `passage_hours` D (integer), `latest_departure_h`, and `limits`. `grid.json` has `lats`, `lons`, `times_h` (ascending) and fields `u_kt`, `v_kt` (eastward/northward wind components, knots, the direction the air moves TOWARD) and `gust_kt`, each indexed `[time][lat][lon]`. Write `solve.py` (Python 3 standard library only); `python solve.py input.json answer.json` writes `answer.json` and `window.svg`. Fresh inputs are used too. ## Rules 1. Interpolation of each field: bilinear in (lat, lon) at the two bracketing grid times, then linear in time. Outside a grid axis, clamp to its edge. 2. Course = atan2((lon_to - lon_from) * cos(mean lat), lat_to - lat_from) in degrees mod 360. 3. For every integer departure hour t0 = 0 .. latest_departure_h: sample k = 0..D at time t0 + k and position from + (to - from) * k / D (straight in lat/lon). At each sample: wind speed = hypot(u, v); wind FROM direction = atan2(-u, -v) mod 360; off = smallest angle between it and the course. The sample fails, checked in this order, with reason `wind@<t>` if speed > max_wind_kt, `gust@<t>` if gust > max_gust_kt, `headwind@<t>` if off < min_wind_angle_deg and speed > headwind_max_kt. Stop at the first failing sample; that departure is not ok and its reason is recorded (t as an integer). Otherwise ok with reason null. 4. `earliest_ok_h`: first ok departure or null. `longest_ok_run`: [first, last] departure hours of the longest run of consecutive ok departures (earliest wins ties), or null. ## Visual output: window.svg Static, self-contained SVG (root `<svg>`, no scripts, images, links or external assets; renders headless) with an element `id="timeline"` (one mark per departure hour) and an element `id="best-departure"`. The drawing itself is not graded. ## answer.json `{"course_deg", "departures": [{"depart_h", "ok", "reason"}, ...], "earliest_ok_h", "longest_ok_run"}` ## Scoring Hidden checks (partial credit; all must pass): `course` (within 0.11), `departure_flags` (ok flags), `failure_reasons` (exact), `best_departure`, `window_svg_static` (window.svg exists, follows the static rules, has the ids above), `fresh_run_1`, `fresh_run_2`. 60-second limit per run.
Space-time interpolation of vector components (not speed and direction), the toward/from convention, ordered failure reasons per sample along a moving position, and run-length 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. course 2. departure_flags 3. failure_reasons 4. best_departure 5. window_svg_static 6. fresh_run_1 7. fresh_run_2
Best model for this task so far: Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) (success, $0.0065, 26s).
Latest output per model
All runs (10)
90% pass rate across 10 eligible runs: 9 pass, 0 partial, 1 fail. Partial and failed runs are listed below with everything else.ⓘ
| Model | Cost | Time | Step-insⓘ | Autonomy | Outcome | Trustⓘ | Date | Compare |
|---|---|---|---|---|---|---|---|---|
| MMistral Small 3.2 24B Instruct (Venice)thinking on | $0.02 | 13m | 0 | Fully autonomous | fail | Trust 80% | 2026-10-29 | |
| QQwen3 Coder 480B Turbo (Venice)thinking on | $0.0089 | 41s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-27 | |
| GGemma 4 31B Instruct (Venice)thinking on | $0.01 | 5m | 0 | Fully autonomous | pass | Trust 80% | 2026-10-24 | |
| GGemini 3.8 Flashmedium · thinking on | $0.33 | 3m | 0 | Fully autonomous | pass | Trust 80% | 2026-10-22 | |
| GGemini 4 Argonmedium · thinking on | $0.23 | 5m | 0 | Fully autonomous | pass | Trust 80% | 2026-10-22 | |
| OGPT-6.1 Soldefault · thinking off | $0.05 | 1m | 0 | Fully autonomous | pass | Trust 80% | 2026-10-22 | |
| DDeepSeek V4.1 Flash (Venice)thinking on | $0.02 | 58s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-21 | |
| AClaude Haiku 5.5default · thinking on | $0.0065 | 26s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-20 | |
| AClaude Opus 5.5default · thinking on | $0.19 | 35s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-20 | |
| AClaude Sonnet 5.5default · thinking on | $0.06 | 21s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-20 |
Embed this result
A live badge for this use case. It links back here and updates as new runs are published.
Get new runs and ranking changes weekly
Which models got better or cheaper on real tasks like this one.