Model → Tactic Synthesis — Newport Bermuda Race
Rev: v2.0 · Updated: 2026-05-29
Take a specific model read and produce a ranked, actionable tactic shortlist with Expedition setup notes. Use from T-10 through start. All scenarios at S3 (10% polar degradation) — Lupo's planning default.
TL;DR
- Flow: weather pattern → tactic shortlist (Step 1) → model + Stream gates (Steps 2–3) → Expedition setup (Step 4) → ranked hypotheses (Step 5).
- T-RH is always the baseline. Every off-rhumb tactic is only meaningful as a net delta from the rhumb after maneuver cost.
- T-CCR has zero wins and one decisive failure (2016) — never commit a ring exploit on RTOFS alone.
Step 1 — Pattern → tactic shortlist
| Analog | What the model shows | Evaluate | Ruled out | Default if uncertain |
|---|---|---|---|---|
| W-BH | Stable Bermuda High; SW–WSW 10–20 kt; no front | T-RH, T-PF, T-RR, T-CF (Stream feature confirmed) | T-CCR w/o SST+altimetry; T-FR | T-RH + T-CL |
| W-PF | Cold front crossing fleet; pre-frontal S/SW, post-frontal NW–W | T-FR, T-PF, T-EC, T-CL | T-CCR w/o confirmation; T-RR2 | T-FR + T-PF |
| W-LA | Light-air start; <8 kt first 12–24 h | T-CF, T-LA, T-CL, T-PF | T-CCR (maneuver cost too high); T-FR | T-CF + T-CL |
| W-CF | Strong active front; pre-frontal 20–30+ kt; post-frontal NW 25–35 kt | T-FR (dominant), T-WR or T-ER | Nearly everything else | T-FR first; hold others |
| W-SQ | Benign gradient, embedded convection; CAPE >500 J/kg | T-RH, T-CL | None — conditions management, not routing | T-RH + T-CL; squall protocol |
| W-RC | North wall near rhumb; favorable NE current; meander present | T-RH, T-CF, T-ME (meander confirmed) | T-CCR (meander ≠ ring) | T-ME if confirmed; T-RH if not |
| W-AC | Adverse current on rhumb (>2 kt for 60+ nm) | T-WR or T-ER, T-CF | T-RH (this is when rhumb fails); T-CCR | T-ER or T-WR per routing comparison |
| W-CC | CCR within 40 nm of rhumb crossing latitude | T-CCR (if SST+altimetry confirm), T-RH (if not — 2016) | T-RH alone if ring confirmed adverse on rhumb | T-RH unless confirmation is airtight |
| W-WE | WCE adverse limb across post-Stream corridor (31–34°N) | T-WCE, T-RH or slight west offset | T-ER (east increases WCE exposure) | T-WCE + T-RH |
| W-BS | Ridge building near Bermuda; light south of 33°N final 200 nm | T-PF, T-ER or T-WR | None specific | T-PF toward least-light path |
| W-PL | High-pressure stall; wide light-air zone 31–35°N after Stream | T-LA, T-CL, T-CF | Heavy-air / high-maneuver tactics | T-CF + T-CL; accept the parking lot |
Transition analogs: most races carry a primary/secondary pair (e.g. W-PF/W-BH). Build a scenario per code — dominant drives the primary hypothesis, secondary is the contingency.
Step 2 — Model agreement filter
| GFS vs ECMWF Δ (elapsed to finish) | Agreement | Effect on shortlist |
|---|---|---|
| ≤6 h | HIGH | Full menu; confidence to commit off-rhumb |
| 6–12 h | MODERATE | Drop T-CCR and T-ME; limit to T-RH, T-PF, T-FR; T-CF for Stream baseline only |
| >12 h | LOW — flag HIGH UNCERTAINTY | T-RH only; no off-rhumb commitment; run two scenarios; brief skipper explicitly |
When models diverge, identify where (ridge position, front timing, Stream-crossing wind), run a scenario per model, and treat opposite tactic families as an uncertainty bracket — not a synthesis. High-commitment calls cannot be made under model disagreement; wait for the next cycle.
Step 3 — Gulf Stream feature filter
| Feature | Confirmation required | If confirmed | If not |
|---|---|---|---|
| Meander (NE-flowing eastern limb) | SST + RTOFS agree | T-ME, T-CF, crossing strategy | Remove T-ME; keep T-RH |
| Cold-core ring (CCR) | SST + altimetry + RTOFS (all three) | T-CCR, T-CF | Remove T-CCR — 2016 is the warning |
| CCR from RTOFS only | RTOFS alone | Nothing — mark H-HR; require SST + altimetry first | T-RH default |
| WCE in post-Stream path (WCE) | SST + altimetry | T-WCE; plan west offset post-Stream | Still check — frequently underweighted |
| Normal / clean | SST + RTOFS agree on position | T-RH baseline; T-CF for Stream core | — |
| Adverse crossing | SST + RTOFS show adverse axis | T-WR or T-ER pending routing comparison | Do not assume adverse without data |
Meander vs CCR: a meander is a connected bend in the main Stream (continuous axis, no isolated cold core); a CCR is detached (isolated SSH low, cold center in SST). See T-ME vs T-CCR and the crossing strategies.
Step 4 — Expedition setup by tactic
| Tactic | Waypoint structure | Ensemble / sensitivity | Net comparison gate (S3) |
|---|---|---|---|
| T-RH | Newport start → St. David's Head direct | 5–10 ensemble runs; note elapsed spread | This IS the baseline. Everything else compares to it. |
| T-PF | Toward higher-wind-speed corridor | GFS + ECMWF; flag if they point to different pressure axes | ≥6 h saving vs T-RH |
| T-ME | Meander NE-limb waypoint + pre-calculated exit-latitude waypoint | RTOFS ON and OFF; note current contribution separately | ≥4 h net after maneuver cost; pre-set exit latitude before running |
| T-CCR | 5–15 nm west of ring center on western limb | Ring exploit vs rhumb with RTOFS overlay | ≥6 h net after full maneuver cost; SST+altimetry gate cleared first |
| T-FR | Two scenarios: (a) pre-frontal S optimized; (b) post-frontal NW optimized | Run both cycles; models will disagree on front timing | Works in both models = robust; single-model = gamble |
| T-CF | Through confirmed current corridor | Quantify current contribution in hours | ≥6 h gate (as T-CCR); confirm SST-validated |
| T-WCE | Slight west offset post-Stream (31–34°N) | RTOFS current overlay | Avoidance often worth 1–2 h for minimal distance cost |
| T-WR / T-ER | West/east waypoint; 15–30 nm lateral divergence unless compelling | Compare west and east vs rhumb | Large offsets rarely pay; check current overlay both sides |
Never skip the T-RH baseline. Every routing call is only meaningful as a delta from the rhumb.
Step 5 — Ranking the shortlist
Rank by: (1) robustness — works under both GFS and ECMWF; (2) net saving over T-RH at S3 after maneuver cost; (3) execution confidence on an unfamiliar boat; (4) data confidence — SST + altimetry confirmed, not RTOFS-only.
Hypothesis [letter]: [tactic tag]
Pattern: [analog code]
Saving vs T-RH at S3: [h] (with current) / [h] (without)
Data gate: CONFIRMED / H-HR / NOT CONFIRMED
Robustness / Execution: HIGH / MEDIUM / LOW
Verdict: PRIMARY / CONTINGENCY / RULED OUT
Lupo's tiebreaker: when two tactics show similar modeled savings, T-CL picks the lower-maneuver option.
Historical pattern benchmarks
The per-year corridor medians (E / RH / W, with n) and per-year lessons live in
winner_route_summary.md and
historical_race_database.md — both sourced
from scripts/analysis/stream_corridor_analysis.py (strict 40–45 ft cohort, YB tracker
2012–2024). Medians are not cross-year speed-comparable (different wind regimes), and EAST
is n=1–2 in every year except 2022 (n=29), so only 2022's east figure is statistically
meaningful. (The duplicate benchmark table was archived 2026-05-30 — see archive/.)
Pattern across 6 years: T-CL is in every winner's stack. T-CF appears in 3 of 6 light-air years. T-CCR = zero wins, one decisive failure. The actionable read by situation is the Default column of Step 1 plus the per-year Lesson in winner_route_summary.md — work those, not a third table.
Sources
- Corridor medians (E/RH/W, n) —
scripts/analysis/stream_corridor_analysis.pyoverdata/yb_cache/stream_corridor_stats.json(strict 40–45 ft cohort, YB tracker 2012–2024). - Winner tactics, Stream features, 5 kt / ~7 h eddy / RTOFS-20–40 nm details — winner_route_summary.md.
- W-* taxonomy & per-year classification — weather_analog_framework.md.
- Tactic definitions (T-*) & H-HR rule — routing_tactic_catalog.md.
- Stream features, crossing strategies, CCR/WCE guides — gulf_stream_framework.md.
- Forecast → analog pipeline (replaces the hand-kept model-comparison log) —
scripts/daily_brief.py, forecast_analog_matcher.md.
PRE-RACE RESEARCH — not race-period routing advice. Updated 2026-05-29.