System Map

One page: how raw data becomes a tactical read, and where every number comes from.

The pipeline

forecast (Open-Meteo: GFS + ECMWF + ICON)
        │
        ▼
classify per period  →  W-* weather-type codes  (P1 start / P2 Stream / P3 approach)
        │                with model-agreement confidence
        ▼
analog matcher  →  closest historical race years  (period-aware, confidence-weighted)
        │
        ▼
tactics  →  corridor decision · sail calls · crossing plan
        │
        ▼
daily brief  →  data/daily-brief/YYYY-MM-DD.md   (scripts/daily_brief.py)

Live Gulf Stream state (north wall, rings, eddies) is tracked separately by the weekly monitor and folded into the corridor decision.

Data provenance

Layer	Source	Type	Where it enters
Wind / MSLP / gusts / CAPE forecast	Open-Meteo (GFS, ECMWF, ICON)	Forecast model	`scripts/forecast_monitor/` → classifier
Weather-type classification	Rule-based, thresholds from `weather_analog_framework.md`	Derived (auto)	`scripts/forecast_monitor/classify.py`
Historical analogs	NB Race results 2012–2024	Observed history	`analog_matcher.py` + `05_historical_races/`
Gulf Stream structure	6-source weekly monitor (altimetry, NCOM, RTOFS, Rutgers, SPoRT SST, Frank Bohlen notes)	Mixed observed + model	`scripts/gulfstream_monitor/` → corridor decision
Fleet tracks	YB Tracking	Observed	`historical_fleet_tracks.md`
Polars	ORC VPP + North Sails chart (Italia 12.98)	Modeled boat data	`02_polars/`

Forecast-horizon truth

No global model reaches race day with skill until roughly T-10 to T-14. Before that window, analogs and climatology carry more weight than any single forecast run. The classifier always prints raw numbers and model-agreement confidence so a human can override a low-confidence tag — divergence is flagged, never hidden.

Auto vs hand

Automatic: forecast fetch, weather-type classification, analog matching, brief generation.
Human-in-the-loop: reading the Gulf Stream imagery, confirming features across sources, and the final tactical call. The system narrows the choice; the navigator makes it.

Pre-race research only.