Case Ledger

Results move a lot from one repository and language to the next. Each table lists every pull request in the test set with a one-line summary of the bug, its severity, and whether each reviewer flagged it. Every tool name opens that tool’s run, and every mark opens the pull request itself.

Which reviewer is right for you depends on what you weigh. Some wrote fuller summaries, some returned in under a minute, some simply stayed quieter. Read the cases that look like your stack and your tolerance for noise.

Flagged = a line-level comment sitting on the faulty code that explains the impact. A passing mention in the summary does not count.

Pull request / bug	Severity	Mezzotint	Plateworks	Halftone	Baren	Deckle
Speed up pagination for high-volume audit logs Imports a cursor paginator that no longer exists	High
Evict from the span buffer during insert Negative offset lets the cursor skip page boundaries	Critical
Support upsampled error counts A sample rate of 0.0 is falsy and silently skipped	Low
Harden the single sign-on callback Null reference when the session state is missing	Critical
Self-serve bulk delete for session replays Breaking change to the error response shape	Critical
Health monitoring for the multiprocess buffer Metric tagged ‘shard’ on one path and ‘shards’ on another	Medium
Cross-system issue synchronisation Shared mutable default on the timestamp field	Medium
Reorganise incident and occurrence creation Reads a stale config value instead of the refreshed one	High
Queue-backed parallelism for background workers Shutdown exception is swallowed and never re-raised	High
Emit occurrences from the stateful detector Handler body is a bare pass statement	High
Total flagged		8/10	4/10	3/10	4/10	1/10