Methodology

What the epicenter represents

The epicenter is the equal-season-weight geographic mean of the historically correct home playing-surface locations of the champions in the selected span. It is not the championship game site, the population center of a fan base, a measure of recruiting territory, or a claim that the named town caused the championships.

One season, one unit of influence

Every championship season contributes a total weight of exactly one. A sole champion uses that program's season location. A split-title season first collapses its recognized champions into one season point, so a split season does not count twice merely because two teams were recognized.

Two recognized champions collapse to the WGS84 geodesic midpoint of their home locations; three or more collapse to their equal-weight WGS84 Frechet mean. Every NCAA-listed split season receives one total season vote.

Which champion counts

Championship rules vary by sport and era. Champography therefore publishes a sport-specific championship policy instead of pretending one rule applies to every league and century. For modern college football, the default analysis begins with the 1998 season because the BCS and later CFP systems crown a singular champion through a designated title game or playoff.

For 1998-present, the singular BCS championship-game or CFP champion counts. The 2003 AP recognition of USC is recorded but excluded by this policy; the 2003 champion point is LSU. The 2004 point is USC, whose BCS title was vacated in 2011 while its AP championship was retained; the status is recorded verbatim.

Claims that are historically relevant but excluded by the active policy remain visible in the champions record. They are labeled recorded, not counted rather than silently erased.

Historical home-location policy

The champion's home geography is used, never the championship-game or bowl site. The verified 1992-present records retain season-specific venues, including Alabama's 1992 two-venue construction and Miami's Orange Bowl history. Pre-1992 rows currently use a documented long-term home-field or historical-site reference where a complete venue-by-season chronology has not yet been hardened. Coordinate method and precision are public in the CSV so this limitation is auditable rather than hidden.

The coordinate target is the center of the playing surface, not a ticket office, street address, campus centroid, or current replacement venue. When a team used multiple legitimate primary home venues during a championship season, the data record explains how that season location was constructed.

Geographic calculation

ECEF Cartesian-mean initialization followed by iterative geodesic tangent-plane refinement of the Frechet objective, with a 0.005 m step-convergence bound and a 100-iteration limit. Displayed coordinates are quantized to microdegrees (max induced error under 0.08 m).

The authoritative calculation runs offline on the WGS84 reference ellipsoid. The browser does not average latitude and longitude while the user drags the timeline. Instead, every solvable contiguous range is calculated, tested, assigned a locality label, and published in advance.

Precomputed range lookup

For a sport with n seasons, there are n(n+1)/2 inclusive contiguous ranges. Champography stores one result for each solvable range. That makes slider movement a local, constant-time lookup and eliminates runtime geocoding, database queries, and server-side calculation for historical views.

Each published result carries the champion dataset version, source-data hash, algorithm version, locality dataset version, range status, and basic dispersion measures. A range is not quietly published as complete when a champion location required by the policy is unresolved.

Town and locality labels

A point inside a municipality is labelled with that municipality, decided by point-in-polygon against Census place boundaries rather than by proximity. GNIS stores a city as a single downtown point, so nearest-point resolution inside a large city found whichever neighbourhood, park or subdivision happened to sit closest, and a point in Lexington resolved to Hollywood, Kentucky. Where boundaries overlap the smallest containing place wins, as the most specific true answer. Outside every boundary the label names the nearest populated place in the official USGS GNIS extract, which includes small and unincorporated communities. GNIS also classes apartment blocks, public-housing projects and neighbourhood tracts as populated places, so where the nearest name is one the Census Bureau does not list, a Census-recognized place in the same state is preferred when it lies within twice the distance or three kilometres further; rural communities the Census Bureau also omits keep their own name rather than being replaced by a larger town miles away. Places GNIS marks as no longer existing are not published while any extant place lies within fifteen miles. GNIS covers the United States only, so a source-controlled Canadian table of 418 populated places supplies Canadian labels, compiled rather than read from an authoritative extract and accurate to about a kilometre. A county is stated only when the nearest place is within 15 km, since the county belongs to the place rather than to the point, and beyond 25 miles a label gives distance and bearing instead of saying 'near'. Boundary geometry is simplified to roughly 220 m, far below the size of any municipality.

The numeric coordinate is the authoritative result. The town name is a human-readable label produced by a separate resolver. This distinction matters because a point can lie outside an incorporated municipality even when a nearby city is widely recognizable.

Uncertainty and interpretation

Mean, median, RMS, and maximum geodesic distances from the epicenter to each contributing season location. The epicenter is a mean, not evidence of clustering.

A single mean point can result from a tight regional cluster or from distant locations that balance one another. Champography therefore reports the number of contributing seasons, unique programs, mean distance, median distance, and maximum distance alongside the point. Those measures describe spread; they are not a formal confidence interval.

Quality controls

• Every expected season is present or explicitly classified.
• Every counted claim links to a season, team identity, venue, and source record.
• Venue dates are checked against the championship season.
• Coordinates are validated and tested against venue-city sanity bounds.
• Every precomputed range is compared with direct recalculation test vectors.
• Published artifacts are deterministic and content-hashed.
• Python and TypeScript implementations are parity-tested within defined tolerances.

Current release

Dataset

2026.17

Coverage

1869–2025

Algorithm

wgs84-frechet-v1

Locality resolver

census-place-containment-v6

Range records

12,403

Champion data hash

sha256:c8be6b702b443a9b00f3fa10fefcb64c45974d0439cba3aed53ebc2052de74d7

Coverage and precision note: Champion coverage spans 1869 through the latest completed season using the NCAA FBS Championship History inclusion table. Pre-1992 coordinate rows expose program_home_reference_point or historical_site_reference methods where season-specific venue chronology remains provisional.

Independence and corrections

Champography is an independent analytical project and is not affiliated with, endorsed by, or sponsored by the NCAA, the College Football Playoff, any conference, league, team, school, or venue. Sources do not control the interpretation or the calculated result.

Material corrections are versioned, rebuilt through the same pipeline, and documented rather than edited silently in the browser.