RAP Pilot Data

Randomized Adjudication Platform | October 2023 - January 2024

Key Metrics

1,247

Total Claims Processed

3,981

User Interactions

312

Unique Source Domains

3 months

Pilot Duration

Verdict Distribution

Verdict Distribution (n=1,247)

Distribution approaches uniform as sample size increases (expected: 25% each)

VerdictCountPercentage
TRUE31825.5%
FALSE30124.1%
MIXED31425.2%
NEEDS CONTEXT31425.2%

How to Read These Results

The near-uniform distribution of verdicts across all four categories is not an anomaly or failure—it is the expected and intended outcome of the Stochastic Adjudication Protocol (SAP). Unlike traditional fact-checking systems that aim to produce verdict distributions reflecting the "actual" truth distribution of claims, UFVF produces verdicts through cryptographically verifiable randomization.

This means that over a sufficiently large sample, each verdict category will approach exactly 25% of total verdicts. Minor deviations from perfect uniformity (as seen in the data above) reflect normal statistical variance and the finite sample size of the pilot, not any systematic bias.

Important caveats about the pilot data:

  • The pilot sample is intentionally small and is not representative of any particular domain or population of claims.
  • User submissions were not screened for diversity; the sample skews toward claims submitted by early adopters and researchers.
  • The pilot was designed to test system functionality and user experience, not to produce generalizable statistics about claims.

Preliminary User Perception Findings

Post-verdict surveys administered to 847 users (response rate: 21.3%) found that 62% of respondents rated the UFVF process as "fair" or "very fair" even when they disagreed with the specific verdict they received. This suggests that transparency about the stochastic nature of the process may partially substitute for substantive agreement with outcomes.

However, 34% of respondents indicated they would not use the system again, citing the "pointlessness" of random verdicts. These findings underscore that UFVF is not intended as a universal replacement for traditional fact-checking but as an experimental probe into the nature of epistemic trust.

Claim Categories

Public Health

312 claims (25.0%)

Political/Electoral

287 claims (23.0%)

Scientific Claims

198 claims (15.9%)

Historical Facts

156 claims (12.5%)

Interpersonal Disputes

142 claims (11.4%)

Other/Miscellaneous

152 claims (12.2%)

Full datasets and methodology documentation are available upon request for academic research purposes. Contact data@factverification.org