GridCase Lab — 24 Pathfinding Regression Fixtures

Catch small grid-navigation regressions before they reach your game. GridCase Lab is an engine-neutral dataset of 24 deliberately specified JSON cases, with a zero-dependency Python reference runner.

What you get

• 20 route cases and 4 explicit invalid-input cases.
• Cardinal/diagonal movement, destination terrain weights, corner-cut policies, unreachable goals, start=goal, edge channels and equal-cost routes.
• Expected reachability/cost or error, one example optimal path, case notes, a SHA-256 manifest and usage guide.
• Python source under MIT; fixture data licensed for commercial project use, not standalone redistribution.

Try 4 cases free

The demo contains baseline, no-route, weighted-detour and forbidden-corner cases. It includes the same reference runner and license details. The remaining 20 fixture files are only in the paid ZIP.

Run locally

Python 3.10+ and the standard library only. Extract the ZIP, open that folder in a terminal and run python3 runner.py fixtures. Compare your own implementation against the documented contract; the runner does not automatically connect to your engine.

Compare your export — version 1.1

Export result JSON from your own pathfinder, then use compare_results.py to check reachability, minimum cost and route validity against the fixture contract. Missing or unexpected case IDs fail; any valid minimum-cost route is accepted. Optional JUnit XML output and non-zero exit codes support your own test workflow. See CI_GUIDE.md in both downloads.

reference_results.json is a startup/schema example from the included Python solver, not evidence that your engine passes. XML generation and parsing were tested locally; no hosted CI pipeline or engine integration was run. The original 24 fixture JSONs, reference runner, licenses and price are unchanged.

Verified scope, clear limits

Valid costs cross-checked with independently constructed Bellman-Ford edges and Dijkstra search. 34 automated tests plus 240 seeded randomized cross-checks passed locally on Linux. This is not a map editor, performance benchmark, engine plugin or guarantee of bug-free code. Godot, Unity, Unreal, mobile and browser integrations have not been tested. Any optimal route is acceptable: do not assert exact equality with the example path.

AI disclosure

AI-assisted case design, code, documentation and code-drawn SVG preview graphics. Cases target individually specified failure modes; no scraped maps, external artwork, music or competitor code are included.

Price: USD 5. No subscription, account, network connection or paid dependency is needed to use the downloaded pack.