Published accuracy.
A historical evaluation report, with its recorded values and measurement limits.
20 cases in the report · recorded run date 2026-07-27
The published record, right now
What Trust Me Bro has published, counted from the stored receipts and refreshed hourly. These are counts of what exists. None of them says a receipt is correct.
22,050
published receipts
80,966
distinct sources cited
11%
cite both sides
31%
rest on a primary source
0%
revised after publication
3
sources per receipt (median)
Verdicts published
SUPPORTED
13,552
PARTIAL LEAN
3,062
REFUTED
2,142
COMMON KNOWLEDGE
1,197
CONTESTED PARTIAL
1,197
CONTESTED
793
OVERSTATED
104
INSUFFICIENT EVIDENCE
2
NOT QUITE
1
Cited sources by tier
peer reviewed
26,752
unrated
24,860
reference
21,115
peer_reviewed
12,479
official
1,053
uncountable
674
primary
472
fact checker
170
official record
147
other
94
primary data
61
Receipts published per month
Readers have raised 1 open source disputes. Withheld receipts are not counted. The same numbers are at /accuracy.json.
The current release is not qualified by this report.
Historical results are preserved below. Per-metric denominators and source revision are not recorded in this report, so the case count is not a denominator for every percentage. The report also lacks per-case outcomes and error counts. Its labels, source support and quote fidelity are not established as independently verified. The descriptions below explain the legacy runner in this codebase; the report does not bind itself to an exact version of that runner. No improvement or release pass is inferred.
Recorded false-attribution rate
0%
The legacy runner counts returned sources rated STRONG or DECENT on cases whose stored label says no supporting source is expected. A low score does not independently establish source identity or support.
Recorded source-return rate (recall)
86%
The legacy runner counts whether a source object was returned on cases labelled as expecting a source. It does not independently verify that the returned source supports the claim.
Recorded verdict-label agreement
100%
The legacy runner compares citation verdicts with each case's stored set of acceptable labels. This report does not establish who verified those labels and is not an independent measure of current claim-verdict accuracy.
Recorded quote flag rate
100%
Stored under quote_fidelity. The legacy runner counts the quote_verified flag on returned quotes and treats a missing quote_verified flag as true. This is not an independent comparison with source text.
Caveats: percentages are rounded from the stored rates. Missing or invalid values
appear as n/a. These metrics have different populations and do not measure accuracy
across all claims, sources or users. A citation identifies a source; whether it
supports a statement requires checking the evidence and its context.