Studies: citable road-trip data tables

halfway.travel studies are computed once, published at a permanent URL, and left citable forever. Each one ranks a real travel question — where America's Thanksgiving corridors get long, where they thin out — and answers it with a table whose every number traces back to a documented pipeline. The register is deliberate: findings in the present tense, methods in the past tense, and a strict wall between numbers that were measured and numbers that rest on a labeled assumption.

The v1 site runs on an estimate engine, not routed miles: drive times come from straight-line distance times a road factor, town facts from a GeoNames-derived seed, and lodging counts from population-derived estimates. Studies built on that engine say so in the first paragraph of their methodology — v0.9 of the Thanksgiving study is exactly that, an honest first cut that the routing upgrade will re-run under a new version number.

Every study follows the same six-step discipline. First, a corpus is fixed by numbered inclusion rules — for the Thanksgiving study, rule E1 froze a 50-metro universe and rule E2 took the top 50 unordered pairs. Second, a ranking metric is defined by formula before any number is computed; the Thanksgiving ranking used gravity (population × population ÷ estimated road distance squared), stated openly as a modeled proxy for holiday volume, with ties broken by distance then alphabetical pair. Third, every computed column gets a one-sentence metric definition with its formula printed verbatim. Fourth, each non-measured input is logged in an assumptions register as ASSUMPTION, and everything measured is logged as MEASURED — the two never mix in a sentence. Fifth, the methodology block lists at least three known limitations, unflattering ones included. Sixth, the table of record ships as a versioned CSV and the build script that produced it is named in the repo.

The result is a page a reporter can quote without calling anyone. Each stat is either traceable to the printed formula or flagged as an assumption, and a corrections policy with a five-business-day commitment sits at the bottom of every study.

Newsrooms may quote any stat, table, or chart with attribution. Do not screenshot numbers without the version: the engine is upgraded over time, so a v0.9 figure and a later figure can differ, and the version is what keeps a citation true. The cite-as line for the current study, ready to paste:

CSVs carry a CC BY 4.0 license; the narrative and computed scores on each page are CC BY 4.0 as well. Charts are build-time static SVG files that may be republished unmodified with the same attribution line printed on the asset itself.