Methodology
Fifty-two dimensions. Thirty-one features. Four factors. Eleven archetypes. The complete analytical pipeline.
FleetEdge's Multi-dimensional Performance Archetype Engine combines three independent data authorities into a single analytical framework. The preceding pages — Hull Efficiency, Rating & Classification, The Physics — each describe one layer of input. This page shows what happens when they converge.
Ocean and atmosphere, sampled at the race.
ERA5 Atmosphere ERA5 is the ECMWF global atmospheric reanalysis — wind, wave, pressure, and boundary-layer state on a common hourly grid, retrospectively computed against the full observation network. FleetEdge samples ERA5 at race-leg centroids and venue bounding boxes, timed to each race's actual window, and associates the atmospheric state per boat via race participation. Atmospheric enrichment is corpus-wide: every scored race carries ERA5-derived dimensions, and weather-dependent features fold into the archetype engine uniformly.
HYCOM Ocean HYCOM GLBy 93.0 is the Hybrid Coordinate Ocean Model global reanalysis at 1/12° horizontal resolution — surface currents, temperature, and sea-surface height retrospectively computed against the global ocean-observation network. FleetEdge samples HYCOM at the same race-leg centroids and venue bounding boxes as ERA5, extracting surface-current magnitude and direction timed to each boat's actual race geometry rather than to a generic venue average.
Coverage posture. ERA5 atmospheric enrichment is corpus-wide across the scored race set. HYCOM ocean-current enrichment is flagship-scoped at launch — Cowes, Fastnet, Rolex Middle Sea, Sydney–Hobart, and RORC Caribbean 600, the five IRC offshore classics where tidal flow is a first-order variable rather than a rounding error. The operational ocean-current floor runs at approximately ten percent of corpus-wide observation coverage; broader coverage follows as the ocean-currents residue sweep closes. Both layers fold into the unified environmental input to the archetype engine, and current-derived dimensions are disclosed as flagship-scoped on every surface that cites them. See Ocean Intelligence for the race-by-race current treatment.
Thirty-one dimensional features of performance.
Each boat in the fleet is described by thirty-one dimensional features extracted from the three data authorities. These features are not arbitrary selections — they are the parameters that naval architects, rating engineers, and physicists use to characterise a sailing yacht's performance envelope.
| Family | Representative Features | Source Authority |
|---|---|---|
| Hull Efficiency | L/Disp1⁄3, SA/WS, Cp, RM@25°, BWL/TC, Δ/L | ORC Certificate |
| Rig Performance | SA/Disp2⁄3, foretriangle ratio, mast height ratio, J/P | ORC Certificate |
| Stability | RM@25°, AVS, GM, displacement ratio | ORC Certificate |
| Rating Position | GPH, TCC, upwind/reaching/downwind allowances, rating band | ORC Scoring |
| Physics Delta | Upwind delta, reaching delta, downwind delta, VMG deltas | SailEdge force-balance vs ORC baseline |
Finding the axes that matter.
Thirty-one features contain redundancy. Hull length correlates with displacement. Rig height correlates with sail area. Principal component analysis identifies the orthogonal directions of maximum variance in the fleet — the axes along which boats genuinely differ from one another, stripped of correlation and noise.
Four latent factors capture 87.14% of the total variance in the fleet. The first factor captures the hull-size-to-rig-power ratio — the single most important axis of differentiation. The second captures stability strategy — how the boat generates righting moment relative to its displacement.
Factors three and four resolve progressively finer distinctions in appendage configuration, rating position, and the physics delta patterns that separate boats within the same broad category.
31 features → PCA → 4 latent factors (87.14% variance explained)
Eleven natural groupings.
The four PCA factors define a performance space — a four-dimensional coordinate system in which every boat occupies a specific position. Boats that share structural strategies cluster together in this space. The question is where to draw the boundaries.
FleetEdge uses ensemble clustering — applying multiple independent clustering algorithms (k-means, hierarchical agglomerative, and DBSCAN) to the same data and taking the consensus. Only groupings that persist regardless of the clustering method survive. This eliminates method-dependent artefacts and produces stable, reproducible archetype assignments.
4 PCA factors → k-means + hierarchical + DBSCAN → consensus → 11 archetypes
Five structural families, eleven archetypes.
Aero-Driven
AEROMAX and AEROBLADE — boats whose identity is set by rig power, with sail-area efficiency and upwind VMG characteristics that dominate the structural signature.
Balance-Sensitive
KEELFLEX and BALANCECORE — designs that operate inside a narrow stability envelope where small changes in heel angle and trim produce large changes in speed.
Downwind-Optimized
GLIDEFORM and GRAVITYRUN — hull forms that trade upwind rig power for low-drag profiles and reaching-and-running leverage, with strong downwind VMG relative to upwind.
Platform-Rigid
IRONWIND, STEELFORM, and DEEPFRAME — stiff, high-righting-moment platforms with predictable behaviour under load and conservative rig-to-displacement ratios.
Mixed-Mode-Power
STORMLINE and HEADFORCE — designs that combine high rig power with sensitivity traits (heel response, drag penalties, condition dependence) and the widest spread between best and worst corrected-time finishes in the fleet.
Signal strengthens with data.
Archetype assignments are not static labels. As new race data is processed, confidence accumulates. A boat with three race observations has a provisional assignment — the structural position is identified, but the confidence bounds are wide.
A boat with thirty or more observations across varied conditions has a high-confidence profile — the archetype assignment is stable and the dimensional coordinates are precisely resolved.
The methodology itself is validated against the fleet. The 87.14% variance explained by four latent factors is not a target — it is a measurement.
The eleven archetypes are not imposed — they emerge from the data through consensus clustering. As the fleet grows and new events are processed, these numbers update. The methodology is transparent and reproducible.
Three floors. Every page obeys the same rule.
Structural readouts — archetype distribution, designer or class signatures, dimensional profiles — publish down to any N ≥ 2. Describing what a cohort IS requires only that the cohort exists.
Comparative or result-ranking narratives — who finished where, which archetype performed better on this course, who closed the gap — require N ≥ 5. Comparison below five boats is anecdote, not analysis.
Crew-attribution or performance-residual narratives — decomposing a result into what the boat gave and what the crew added — require N ≥ 10 within a single event. Below that floor, noise dominates the residual.
Cohorts below the applicable floor carry a Low-sample badge and limit themselves to the structural tier; comparative narrative is suppressed. Programmatic enforcement follows in a later release; the policy itself is the public invariant every page is measured against today.
Five public evidence badges.
Every analytical card on FleetEdge carries a badge telling you what kind of claim it is. The lexicon is closed: five values, stable, auditable.
- Production
- Live on-corpus measurement. Results are stable enough for comparative use.
- Validation
- Methodology is under on-site validation. Results are directional while the validation run completes.
- Structural-only
- Dimensional or archetype analysis without race-result attribution. Describes what the fleet IS, not what it did.
- Low-sample
- Cohort is below the narrative floor for comparative analysis. The card ships a structural readout only; no comparative takeaways.
- Pre-race scenario
- Forward-looking analysis using forecast or historical-prior inputs. Not a measurement; a projection under stated conditions.
See Methodology in the Sydney Hobart fleet.
Sydney Hobart racing brings methodology into sharp relief — see how the physics meets the fleet's regime and rivalries.
Explore Sydney HobartRigorous methodology. Transparent pipeline. Every boat measured the same way.
Thirty-one features, four factors, eleven archetypes — applied to every boat in the ORC fleet.