// receipts

Signed receipt ledger

Every benchmark run produces an Ed25519-signed JSON receipt with the full per-scenario or per-query record pinned. Click any receipt to see scores, environment, and the signature. Verify any of them in your browser at /verify.

// multi-agent convergence
// run 2026-05-19 04:48Z6 receipts
Adapter / ModelSubsetFixturesCorrectCollapseSycophancySigned
autogen / gpt-4o-miniholdout683.3%0.0%0.0%
baseline-anthropic-sequential / claude-haiku-4-5holdout683.3%66.7%0.0%
baseline-anthropic / claude-haiku-4-5holdout6100.0%66.7%0.0%
baseline-azure-openai-sequential / gpt-4o-miniholdout666.7%83.3%0.0%
baseline-azure-openai / gpt-4o-miniholdout666.7%83.3%0.0%
baseline-azure-openai / gpt-5-miniholdout6100.0%100.0%0.0%
// run 2026-05-19 03:42Z6 receipts
Adapter / ModelSubsetFixturesCorrectCollapseSycophancySigned
autogen / gpt-4o-miniall3093.3%10.0%0.0%
baseline-anthropic-sequential / claude-haiku-4-5all3093.3%53.3%0.0%
baseline-anthropic / claude-haiku-4-5all3096.7%56.7%0.0%
baseline-azure-openai-sequential / gpt-4o-miniall3086.7%86.7%0.0%
baseline-azure-openai / gpt-4o-miniall3076.7%73.3%10.0%
baseline-azure-openai / gpt-5-miniall3096.7%96.7%0.0%

Receipts are also available raw at /receipts/<benchmark>/<filename>.json for direct download. The implementation lives in przm-bench/results/published.