Methodology

Every rider carries six discipline ratings (climbing, sprint, time trial, hills, GC, one-day) computed from their official UCI results: placings in the relevant classifications, weighted toward the recent season.

The overall rating (OVR) blends two signals: the rider's discipline profile and their UCI world rank, which acts as the anchor — the sport's own tier-aware measure, where raw placing data cannot tell a national-race KOM from a WorldTour one. A calendar-level correction further deflates minor-race specialists.

No data, no rating. When a rider has neither computed sub-ratings nor a world rank, the card shows "—" and the rider is excluded from the homepage editorial picks. A dash means missing data, never zero.

The W/kg shown is an estimate from public results, never power-meter telemetry. Two methods, always labelled:

1. Real ascent times. On uphill finishes with official route geometry (organiser GPX), we reconstruct the rider's ascent time from official gaps, attributing to the decisive climb a share of stage time proportional to its elevation gain. Climbing speed (VAM, m/h) converts to W/kg with the classic field model W/kg = VAM / (200 + 10 × gradient%). Per-peloton plausibility bounds (men's / women's) reject outliers, and the published value is the median of the best recent samples.

2. Profile approximation. Without a usable ascent time we show an approximation derived from the climbing rating, anchored to the world rank. It is always marked "low confidence".

Rider mass does not enter this computation — the VAM model is mass-independent. Absolute watts are shown only where a real public weight exists (Wikidata, CC0); otherwise the column shows "—". We never invent a weight.

Every estimate carries an index computed per rider, per sample — never a global label:

high confidence a directly timed ascent. Rare: it requires public intermediate splits, which we do not yet have from a reliable source — the band exists in the engine and is deliberately left unused rather than inflated.

medium confidence derived from an official stage time on a summit finish with verified GPX geometry (method 1).

low confidence profile approximation (method 2), or a stage time whose decisive climb does not end the stage — a gap at the line says less about the climb itself.

The public panel's estimates are recomputed continuously (within 24 h of new results).

• No P3 number is a sensor reading — no measured power, no heart rate, no private training data.
• Heights, weights and birth dates absent from public sources are not "filled in": they render as "—".
• The homepage race simulation is fictional and labelled as such — a physics engine driven by real ratings, not a broadcast.
• An estimate is an estimate: it places an order of magnitude, it does not settle a watts debate to one decimal.

Results and rankings: official UCI data. Route geometry: organiser GPX and digital elevation models. Morphology (weight/height, where shown): Wikidata (CC0). Stage weather: Open-Meteo (observed, ≤15-day forecast, climate normals — each labelled).