Anchor drag alarm from a GPS track
Exact prompt
# Anchor drag alarm from a GPS track `input.json` (synthetic, not for real anchor watch): `anchor` [lat, lon], `rode_m`, `depth_m`, `bow_height_m`, `gps_offset_m`, `margin_m`, `consecutive` N, and `track` (CSV `t,lat,lon`, t in seconds). Write `solve.py` (Python 3 standard library only); `python solve.py input.json answer.json` writes `answer.json` and `anchor.svg`. Fresh inputs are used too. ## Rules 1. Alarm radius = sqrt(rode^2 - (depth + bow_height)^2) (0 if rode <= depth + bow_height) + gps_offset + margin. 2. Track rows: skip the header. A row is **bad** (counted, skipped) if it has fewer than 3 fields, t is not an integer, lat/lon is not a finite number in range [-90, 90] / [-180, 180], or t is not greater than the last *valid* row's t. 3. Distance from the anchor (metres): R = 6371000; m = mean of the anchor and point latitudes; x = radians(dlon) * cos(m) * R (east), y = radians(dlat) * R (north); distance = hypot(x, y). A point is **outside** if distance > radius. 4. An **event** is a maximal run of consecutive valid points that are outside, with length >= N. Report `start_t` (first point), `trigger_t` (Nth point), `end_t` (last point), `max_dist_m`, and `bearing_deg` = atan2(x, y) mod 360 from the anchor to the run's last point. 5. Round the radius to 2 decimals and distances and bearings to 1. ## Visual output: anchor.svg Static, self-contained SVG (root `<svg>`, no scripts, images, links or external assets; renders headless) with `<circle id="swing-circle">`, `<polyline id="track">` and an element `id="anchor"`. The drawing itself is not graded. ## answer.json `{"radius_m", "valid_points", "bad_rows", "events": [...], "max_dist_m", "outside_points"}` ## Scoring Hidden checks (partial credit; all must pass): `swing_radius` (within 0.011), `track_cleaning` (exact), `distances`, `alarm_events` (times exact, values within 0.11), `anchor_svg_static` (anchor.svg exists, follows the static rules, has the ids/tags above), `fresh_run_1`, `fresh_run_2`. 60-second limit per run.
Catenary-free swing radius with offsets, rejecting bad and out-of-order rows against the last valid time, short excursions that must not trigger, and a real drag run with trigger time and bearing. 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. swing_radius 2. track_cleaning 3. distances 4. alarm_events 5. anchor_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, $0.0081, 34s).
Latest output per model
All runs (8)
88% pass rate across 8 eligible runs: 7 pass, 1 partial, 0 fail. Partial and failed runs are listed below with everything else.ⓘ
| Model | Cost | Time | Step-insⓘ | Autonomy | Outcome | Trustⓘ | Date | Compare |
|---|---|---|---|---|---|---|---|---|
| DDeepSeek V4.1 Flash (Venice)thinking on | $0.01 | 46s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-28 | |
| MMistral Small 3.2 24B Instruct (Venice)thinking on | $0.0077 | 39s | 0 | Fully autonomous | partial | Trust 80% | 2026-10-23 | |
| GGemini 3.8 Flashmedium · thinking on | $0.46 | 4m | 0 | Fully autonomous | pass | Trust 80% | 2026-10-22 | |
| GGemini 4 Argonmedium · thinking on | $0.22 | 3m | 0 | Fully autonomous | pass | Trust 80% | 2026-10-22 | |
| OGPT-6.1 Soldefault · thinking on | $0.05 | 1m | 0 | Fully autonomous | pass | Trust 80% | 2026-10-22 | |
| AClaude Haiku 5.5default · thinking on | $0.0081 | 34s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-20 | |
| AClaude Opus 5.5default · thinking on | $0.18 | 35s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-20 | |
| AClaude Sonnet 5.5default · thinking on | $0.07 | 18s | 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.