The polar diagram plots boat speed (BSP) against true wind angle (TWA) for different wind speeds. Each curve shows how fast the boat goes at each angle — the further from center, the faster.
Wind angle/speed cells with fewer than 5 raw readings are dropped from the curve entirely rather than drawn — a single instrument or GPS glitch in an otherwise near-empty cell would otherwise render as a full-weight spike far outside the smooth, well-sampled shape around it.
This curve bins every qualifying 5-second sample by its raw TWA at that instant, with no regard for how the boat got there or how long it stayed — see the Steady-State Overlay topic below for why that matters at deep angles specifically.
Each wind angle/speed cell contains many individual BSP readings from past sessions. The slider controls which value from that distribution is used to draw the curve:
Avg — the mean of all readings in the cell. Reflects typical speed, pulled down by slow moments.
P75 — 75% of readings were at or below this speed. A conservative but reliable target.
P85 — the default. A good balance between achievable and realistic peak performance.
P95 — near the top of what's been recorded. Represents a very good sail.
Max — the single fastest reading ever recorded in that cell. May include gusts or instrument spikes.
Curves — connects the BSP data points for each wind speed bin into a smooth line.
Smooth — applies a 3-point weighted average across adjacent wind angles to reduce noise.
Scatter — plots every individual raw data point instead of a summarized curve.
Each colored button toggles a wind speed band on or off. Click any button to isolate specific conditions — useful for comparing performance at different wind strengths.
Three theoretical/VPP reference polars can be overlaid on whatever view you're currently looking at, so you can compare directly instead of switching sources back and forth: MM reference (the same table used by the "MM Reference Polar" source in the dropdown above), WCE (Main+A5O, 308m² downwind sail), and WCN (Main+VMG Spin, 344m² downwind sail) — WCE and WCN are two different downwind sail configurations from the same VPP revision, sharing identical upwind numbers but diverging past TWA 130° or so.
Each active TWS band gets one dashed target line, matched to the nearest discrete wind speed that target's table actually has data for (each is a sparse grid at fixed TWS values like 6, 8, 10, 12… — not every band has an exact match). The line color always matches its TWS band; the marker shape tells the three targets apart when more than one is switched on — diamond (MM), triangle (WCE), square (WCN). Where the empirical curve sits inside a target line (closer to center), that's a real speed gap at that angle.
A gybe is detected as a Twa sign-flip while sailing broader than 100° on both sides; a tack is the mirror image — a sign-flip while sailing narrower than 80° on both sides. Either way the flip must complete within 3 minutes, so two unrelated same-side periods hours apart aren't mistaken for one continuous maneuver. "Omit" excludes every point within ±2 minutes of each detected event, both gybes and tacks together (this page shows the whole polar at once, not a single point-of-sail view).
Why it might matter: mid-maneuver, boat speed stalls then surges (or the boat luffs near head-to-wind) — readings that aren't representative of steady-state sailing at that angle. Those transient points land disproportionately in already-thin cells, where a couple of stalled or transitional readings can swing the average and spike the curve. First caught this on the 25° TWA / 10-15kt cell — two near-stalled TransPac 2017 readings (0.6, 2.66kt) mixed into a small sample was enough to distort the percentile curve.
Real telemetry shows a repeating helm pattern nicknamed the hookup maneuver: heat up to a hotter angle to build apparent wind and boat speed, then dive to a deeper angle while the boat is still carrying that speed. It's a real, universal technique (confirmed across 6 races) — but it means many of the raw curve's deep-angle samples are brief momentum-carrying bursts, not sailing that could actually be held. On TransPac '15, 80% of every TWA ≥ 145° sample came from an excursion lasting 30 seconds or less, and the genuinely-held (60s+) samples ran up to 38% slower than the raw curve's bin average.
The steady-state overlay (dotted line, triangle markers) only counts samples that had already held a similar angle continuously for ≥60 seconds before being counted — the first 60 seconds of any qualifying stretch are excluded too, as still-ramping-up. Where the two curves diverge at deep angles, the raw curve is showing you the hookup maneuver, not a sustainable point of sail.
Only available for races (needs per-point timestamps that non-race sessions don't get this precompute pass on). Precomputed per race in race_polar_cells_steady — rebuilt in the chimchim_26 repo (scripts/rebuild_race_derived.php) and synced to this site's production DB separately from code deploys.