How every number here was counted.
Every public number has a method beside it. If the editor returns something other than what was requested, that is the result we keep.
Counts are derived, not typed.
The version line in the repository is generated, and the release gates fail if a count drifts from the tool that computes it. This site reads those numbers from one file, as of 2026-10-15.
MIF_BIND=717 OPS=364 MCP_TOOLS=1145 SUITES=801
6 engines, one full sweep each.
Each of UE 5.3 to 5.8 ran the whole suite corpus against a stock editor: 5.3 → 5.8 · 322–333 suites each, 0 failed, 1 timed out on each, 2026-09-28 to 2026-09-29. No suite failed on any of them. A suite that passed on either of its two runs counts as passed, one that only skipped is not a failure, and one the sweep stopped at its time limit gave no verdict either way, so it is counted as timed out rather than passed or failed.
UE 5.3 : 295 suites, 0 failed, 1 timed out UE 5.4 : 322 suites, 0 failed, 1 timed out UE 5.5 : 322 suites, 0 failed, 1 timed out UE 5.6 : 322 suites, 0 failed, 1 timed out UE 5.7 : 333 suites, 0 failed, 1 timed out UE 5.8 : 312 suites, 0 failed, 1 timed out
Blender on 5 builds.
3.6.23, 4.2.17 LTS, 4.4.0, 5.0.1, 5.2.1 LTS: all 298 Blender suites on each, 0 failed, recorded 2026-10-13 under the version Blender itself reported.
BLENDER_MATRIX: 5 builds × 298/298
What is not verified.
A compile record is not a live sweep, and a passing suite proves only what it checks. Cooked-content behaviour is verified on UE 5.3.2 only, the one cooked project available for testing. Engine-specific caveats stay on their version pages.
status=builds != status=swept
No green badge outranks the editor state that produced it.