DETOUR

This is DETOUR's reference route: … to …, 4,353 GPS points, about … of driving. The candidate towns were found by the tool's own detect_corridor() — project the route to an equal-area CRS, densify it, and take the perpendicular distance from every Craigslist market to the line. That geometry ran in Python, because reimplementing an Albers projection in JavaScript would be exactly the kind of lookalike this site keeps promising it isn't building.

What runs here is the part you actually turn: cheapest and spread from corridor/detour.py, ported honestly — including the bit where spread trims 30 minutes off the start and 45 off the end, divides what's left into equal segments, and makes one rescue pass at 50% wider before giving up. Segments that are genuinely empty stay empty; the tool degrades honestly rather than padding the list.

Which exits are worth taking?

This is the actual output of the engine on the reference route — 11 marketplace regions found inside a 25-mile corridor over a 7.2-hour drive. Flip between the two selection strategies and watch the stops redistribute. That difference is the whole product.

What you’re watching: in Cheapest mode the greedy pick clusters Hudson Valley, Hartford and the early stops into the first few hours — you’d shop one region and burn the budget before New Hampshire. Spread cuts the drive into equal time segments and takes the best stop in each, so the finds are distributed — and it honestly leaves segment 1 empty rather than forcing a stop that isn’t there.

Corridor geometry, done right.

Real numbers, from the tested path.

The corridor detection has a golden-file test on this exact route: it asserts Hartford, Worcester, Boston and the NH/Maine regions appear in the right order, that Maine resolves to subareas including Portland, that nothing exceeds the buffer, and that spread and cheapest produce different sets. The geo core is finished and green.