Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Sailboat instrument dashboard from an NMEA 0183 log

# Sailboat instrument dashboard from an NMEA 0183 log `log.nmea` is a **synthetic** instrument log from a sailboat; `input.json` names it and gives alarm thresholds. Write `solve.py` (Python 3 standard library only) so that `python solve.py input.json answer.json` parses the log, computes true wind and VMG, finds alarm periods, writes `answer.json`, **and** writes a static display `dashboard.html` in the working directory. We re-run it on fresh logs, so do not hard-code anything. ## Log format Each line is `<t>,<sentence>` where `<t>` is an integer second and `<sentence>` is an NMEA 0183 sentence `$<body>*<hh>`. The checksum hh is two hex digits (either case) equal to the XOR of all characters of `<body>` (between `$` and `*`). ## Rules 1. Blank lines are ignored and not counted. A line is **malformed** if it has no comma before the sentence, `<t>` is not an integer, the sentence does not start with `$`, has no `*`, the checksum is not exactly 2 hex digits, or (after the checksum passes) the sentence type is not one of those below or a needed field is missing or not a number. A line whose checksum does not match is **bad_checksum**. Every other line is **valid** (this includes MWV lines that are skipped by rule 3). 2. The sentence type is the last 3 characters of the first field. Fields are split on commas (field 0 is the talker+type): - `HDT`: field 1 = true heading (degrees). - `VHW`: field 5 = boat speed through water (knots). - `DBT`: field 3 = depth (metres). - `VTG`: field 5 = speed over ground (knots). - `MWV`: field 1 = apparent wind angle (degrees clockwise from the bow, 0-360), field 2 = reference, field 3 = apparent wind speed (knots), field 5 = status. 3. Keep the latest heading and boat speed. For an MWV with reference `R` and status `A`, if a heading and a boat speed have already been seen, make a **true wind sample** at its time: x = AWS*cos(AWA) - BSP, y = AWS*sin(AWA); TWS = hypot(x, y); TWA = atan2(y, x) in degrees mod 360; TWD = (heading + TWA) mod 360; VMG = BSP*cos(TWA). MWV lines with another reference or status are valid but make no sample. 4. Alarms: depth alarm on DBT readings with depth < `depth_min`; wind alarm on true wind samples with TWS > `tws_max`; speed alarm on VTG readings with SOG > `sog_max`. For each type, an alarm is a maximal run of consecutive readings of that type (in file order) that break the limit; `start`/`end` are the times of its first and last breaking reading. Sort all alarms by start, then type name. 5. `last` holds the last sample's tws, twa, twd, vmg (2 decimals) and the last boat speed and depth values as read. ## answer.json `{"counts": {"valid", "bad_checksum", "malformed"}, "true_wind_samples": n, "last": {"tws", "twa", "twd", "vmg", "bsp", "depth"}, "tws_mean": .., "tws_max": .., "alarms": [{"type": "depth"|"wind"|"speed", "start", "end"}, ...]}` (mean and max of TWS over all samples, 2 decimals). ## dashboard.html A self-contained static page (no scripts or network needed) with these element ids: `tws`, `twa`, `twd`, `vmg`, `bsp`, `depth` (each element's text is only the value from `last` with 1 decimal), an `<svg id="wind-rose">` containing a `<line id="twd-needle">` pointing toward TWD, and `<ul id="alarm-list">` with exactly one `<li>` per alarm. Styling is up to you. It must be static and self-contained so it can be screenshotted headless: no `<script>`, no `<img>`/`<link>`/`<iframe>`/`<object>`/`<embed>`, no `on*` handlers, no `href`/`src` except `#...` or `data:`, no CSS `url(http...)` or `@import`, no fetches. ## Scoring Hidden checks (partial credit = fraction passed; all must pass): `sentence_counts` (exact), `true_wind_and_vmg` (within 0.011, sample count exact), `alarm_intervals` (exact), `dashboard_elements` (file exists, static rules above, ids present, `wind-rose` is `<svg>`, `twd-needle` is `<line>`, `alarm-list` is `<ul>`), `dashboard_values` (each value within 0.051 of the answer, li count) on the given log; `fresh_log_1` (all answer keys plus dashboard values) and `fresh_log_2` (all answer keys) on fresh logs. Each run has a 60-second limit.