Research idea from DoS skill noise on Relay/Symbolicator: review findings often flag pre-existing issues near changed code, so reviewers ignore real signal.
Proposal
- After findings are produced, run a cheap/fast auxiliary LLM pass
- Disqualify concerns the PR did not actually touch or introduce
- Keep only findings grounded in the PR's changed surface
Related: #313, #408
Requested by David Cramer.
--
View Junior Session [Sentry]
Research idea from DoS skill noise on Relay/Symbolicator: review findings often flag pre-existing issues near changed code, so reviewers ignore real signal.
Proposal
Related: #313, #408
Requested by David Cramer.
--
View Junior Session [Sentry]