Hi Aderyn team,
I maintain a free dataset that might be useful as a triage layer on top of Aderyn's detector output — not a competing tool, just a data point.
What it is: acceptance rates per vulnerability class (reentrancy, access-control, oracle-manipulation, etc.), computed from Sherlock audit-competition judging data, restricted to the subset where results reconcile exactly against Sherlock's own published outcomes (10 of 105 crawled contests, 1,032 findings — smaller than the full corpus, but every number is verifiable, not estimated). Full methodology + reproduction steps: https://github.com/holistis/bug-bounty-intelligence-mcp/blob/main/METHODOLOGY.md
Standalone CC0 JSON, no server, no account: https://github.com/holistis/bug-bounty-intelligence-mcp/blob/main/vulnerability-acceptance-rates.json
The idea: Aderyn's raw findings are unfiltered by design (same false-positive tradeoff every static analyzer makes — my own project's benchmark against Slither documents the same problem from a different angle). A user staring at a detector list might benefit from knowing, e.g., "reentrancy findings in this class are confirmed valid 78% of the time when specifically hunted for by paid auditors, but oracle-manipulation findings are confirmed only 36% of the time" as a prioritization signal — which findings to look at first, not a verdict on any specific finding.
Not asking for anything specific — flagging it in case it's useful as a linked resource, or not a fit at all, your call. Happy to answer questions.
Hi Aderyn team,
I maintain a free dataset that might be useful as a triage layer on top of Aderyn's detector output — not a competing tool, just a data point.
What it is: acceptance rates per vulnerability class (reentrancy, access-control, oracle-manipulation, etc.), computed from Sherlock audit-competition judging data, restricted to the subset where results reconcile exactly against Sherlock's own published outcomes (10 of 105 crawled contests, 1,032 findings — smaller than the full corpus, but every number is verifiable, not estimated). Full methodology + reproduction steps: https://github.com/holistis/bug-bounty-intelligence-mcp/blob/main/METHODOLOGY.md
Standalone CC0 JSON, no server, no account: https://github.com/holistis/bug-bounty-intelligence-mcp/blob/main/vulnerability-acceptance-rates.json
The idea: Aderyn's raw findings are unfiltered by design (same false-positive tradeoff every static analyzer makes — my own project's benchmark against Slither documents the same problem from a different angle). A user staring at a detector list might benefit from knowing, e.g., "reentrancy findings in this class are confirmed valid 78% of the time when specifically hunted for by paid auditors, but oracle-manipulation findings are confirmed only 36% of the time" as a prioritization signal — which findings to look at first, not a verdict on any specific finding.
Not asking for anything specific — flagging it in case it's useful as a linked resource, or not a fit at all, your call. Happy to answer questions.