A fixture changed after 43 predictions were locked
I built PunditBench to make it difficult to tidy up a benchmark after the result.
43 current models predict football matches across five leagues. Each round is collected roughly 36 hours before kickoff, turned into canonical JSON, SHA-256 hashed and tagged.
On 2026-08-21, ESPN reversed one Ligue 1 fixture after all 43 picks had locked: PSG–Rennes became Rennes–PSG.
The easy fix would be to swap the teams in the saved picks. I did not do that.
The 43 original picks are still public against the fixture they received. The real fixture was corrected separately. The match is excluded from benchmark scoring and the €5 brief.
That part works.
Getting people to see it is harder.
From 2026-08-17 through 2026-08-23, GA4 measured 27 users, 68 sessions and 728 pageviews.* Nine non-home league routes made up 396 of the 482 views in the top ten. The brief did not appear there.
Zero eligible payments so far.
I am trying a few distribution routes instead of polishing the benchmark again.
If you were starting from here, what would you put in front of people first: the live match scoring, the method, or the open data?
Full sample and the €5 experiment: https://punditbench.com/briefs/opening-round-2026/?utm_source=indiehackers&utm_medium=community&utm_campaign=2026-08-24_integrity_case&utm_content=fixture_changed_after_lock
*GA4 runs behind analytics consent, so these are lower bounds.
18+. Informational only, not betting advice.
The strongest part is refusing to rewrite history after the fixture changed. Keeping the original prediction immutable and separating the corrected fixture makes the benchmark much easier to trust.