How we check

Fairness tests

Two fairness tests run before a method version may check public episodes. We publish every result here, including the ones that did not pass.

Simulated runs only. No run with the real reading software yet.

Same claim, other side (mirror test)

Pairs of claims where only the political side changes, for example the same claim about a Democratic and a Republican politician. The results must match in at least 98 percent of pairs, with no lean to either side. The full set of 300 pairs runs before public launch. Each result below says whether the reading software was really used, or simulated to test the rules only.

• FactUnion Method 1.0: Simulated run: passed: 30 pairs, 100 percent gave the same result on both sides, no lean to either side (sign test 1). Offline run with simulated readings, which tests the rules but not the reading software.

Same result each time (stability test)

The whole check runs 3 times on each test episode. To pass, at least 85 percent of claims must be found in every run, at least 90 percent of those must get the same result, and the scores must differ by at most 5 points.

• FactUnion Method 1.1: Simulated run: passed: 1 test episode, 3 full runs each; claims matched across runs 100 percent (needs 85 percent), same result 100 percent (needs 90 percent), scores differ by 0 points (at most 5). Offline run with simulated readings, which tests the rules but not the reading software. 17 October 2026.

Results are copied from our method's own output files when this site is built. See also the transparency report.