Learning center

Read a dismissed code review finding before you score it

A dismissed finding is not automatically a reviewer's mistake. Before you count it, find the author's reason, check that reason against the code, and decide whether the dismissal scores against the reviewer, against the author or against neither.

Separate the reason from the outcome#

Authors dismiss findings for different reasons, and each one means something different for a review trial:

Reason givenWhat to checkHow to score it
The finding is wrongThe code or documentation the author citesIncorrect, if the reason holds
The problem is real but handled elsewhereThe other place, at the reviewed commitIncorrect, but note that the reviewer lacked that context
The problem is real but out of scopeThe follow-up issue or pull requestValid, deferred
No reason, or "won't fix"Nothing to check until someone gives a reasonUnresolved

Without the reason, a trial counts every dismissal as reviewer noise. That penalizes a reviewer for findings an author deferred on purpose, and rewards one whose findings nobody bothered to answer.

Follow a reasoned dismissal#

In October 2026, a Home Assistant pull request fixed automatic backups that ran an hour off their configured time after a daylight saving change. The fix converted the stored time of the last backup to local time before using it to seed the scheduler:

-            seed_time = manager.config.data.last_completed_automatic_backup or now
+            # The stored timestamp has a fixed UTC offset, which does not follow DST
+            seed_time = dt_util.as_local(
+                manager.config.data.last_completed_automatic_backup or now
+            )

A maintainer reviewing the change asked: "on second thought - does this need to be fixed in more places?" They worried that the stored value was still off and that other code, such as the frontend, might show the wrong time. Review comment

The author answered with a specific claim: "The stored value is the right instant (01:00:01+01:00 is exactly when it ran), it just carries a fixed offset. That only matters when it's used to work out the next "01:00", and the cron seed is the only place that does that." They added that the frontend parses the value with new Date(...) and the sensors store UTC, and offered an alternative that would convert the value when the configuration loads, at the cost of changing stored text and several tests. Author's reply

The maintainer replied: "if that really is the case I am fine with the current solution". Maintainer's reply The pull request merged on 28 October 2026 without further changes to that code.

Check the reason, not the tone#

"If that really is the case" is a conditional acceptance. The claim is checkable, so check it before you score the finding. In the merged homeassistant/components/backup/config.py, the stored value appears where the configuration is read and written, through isoformat(), and in one calculation, the scheduler seed the pull request changed:

seed_time = dt_util.as_local(
    manager.config.data.last_completed_automatic_backup or now
)
cron_event = self.cron_event = CronSim(cron_pattern, seed_time)

Merged file

Within that file, the author's claim holds: the fixed offset matters only where the next scheduled time is calculated, and that calculation now converts it. A reviewer scoring this finding should record what they checked and what they did not. This check covered one file. A use of the stored value in another module or in the frontend would need its own read.

So this finding scores as a reasonable question, answered with evidence, that did not identify a defect. It is not noise. It cost the author one careful reply and gave the maintainer a reason to approve, which is a review doing its job.

Compare a dismissal that corrects the reviewer#

A vLLM pull request in May 2026 shows the other common case. An automated reviewer said "several type annotations for KVCacheBlock need to be converted to string literals to avoid runtime NameError exceptions". Review The author declined: "Variable annotations in function bodies and on self.attr assignments are not evaluated at runtime", the file already used the bare name in untouched lines, and "the module loads and all 13 scheduler tests pass." Author's reply

Python's language reference agrees: when the target of an annotated assignment is an attribute, "the annotation is never evaluated", and annotations "are never evaluated and stored in function scopes." Annotated assignment statements That finding scores as incorrect. The author's reply also cost more time than the finding's one line suggests, because disproving a runtime error meant checking Python's evaluation rules and running the tests.

Record the dismissal as carefully as an acceptance#

For each dismissed finding, record the author's reason, the code or documentation that supports it, what you checked and the resulting score. Use the same row you keep for an accepted finding, with the reason in place of the fixing commit. An author who writes that reason into the thread, as the dismissal reply guide describes, saves you the reconstruction. A trial that keeps these rows can tell a reviewer that asks good questions from one that is often wrong, even when both have their findings dismissed at the same rate. If the reviewer repeats a dismissed finding on the next push, the follow-through guide shows how to count that cost.

One month of free code review.

New organizations get one calendar month of hosted code review for private and public GitHub repositories. No card or cloud connection required. Eligibility and fair-use terms.

Start free with GitHub

Related reading