Why
Sentinel's abandoned-review notice on quadseven/infra#1998 said:
this PR was merged with its blocking check still failing while Elder's last
review on this commit was still blocking (1 finding(s), severity
high/critical).
This means the finding(s) below may have SHIPPED - please verify.
Then, in the SAME comment:
Elder's last verdict summary
Elder markings - 0 blocking, 1 total
And the finding itself was tagged 👁 low | maintainability | dead-code.
So one comment simultaneously claims "0 blocking" and "still blocking ...
severity high/critical" about the same review.
Why it matters
Sentinel is doing genuinely valuable work here - it correctly caught that I
merged before reading the bot-comment result, which is a real process failure
worth flagging. That signal should stay loud.
But the severity wording is wrong in a way that costs trust:
- A
low maintainability nit is reported as a possible high/critical
shipped incident.
- The claim "blocking check still failing" contradicts
0 blocking from Elder,
quoted three lines below it.
- If low findings routinely read as high/critical, the honest response becomes
skimming Sentinel - which defeats a guard whose entire value is being taken
seriously.
This one happened to be a false positive too (Elder flagged a line the PR itself
deleted), so the notice read as "a high/critical finding may have shipped" for
something that was neither high, nor critical, nor real.
What
Acceptance criteria
Context
Filed from the same session as #771 (shell control-flow FP), #772
(move-awareness), #773 (Elder cancel/re-enqueue deadlock) and PR #774 (base-sha
churn cancelling in-flight reviews).
Size: S
Refs #707
Why
Sentinel's abandoned-review notice on quadseven/infra#1998 said:
Then, in the SAME comment:
And the finding itself was tagged
👁 low | maintainability | dead-code.So one comment simultaneously claims "0 blocking" and "still blocking ...
severity high/critical" about the same review.
Why it matters
Sentinel is doing genuinely valuable work here - it correctly caught that I
merged before reading the bot-comment result, which is a real process failure
worth flagging. That signal should stay loud.
But the severity wording is wrong in a way that costs trust:
lowmaintainability nit is reported as a possible high/criticalshipped incident.
0 blockingfrom Elder,quoted three lines below it.
skimming Sentinel - which defeats a guard whose entire value is being taken
seriously.
This one happened to be a false positive too (Elder flagged a line the PR itself
deleted), so the notice read as "a high/critical finding may have shipped" for
something that was neither high, nor critical, nor real.
What
not a fixed string.
0 blocking, Sentinel should say the merge skipped a NON-blocking review,which is a different and much milder statement.
lowis worth anote; a merged-with-unanswered-
criticalis worth alarm. Same event, verydifferent urgency.
Acceptance criteria
lowfindings outstanding produces a notice thatsays
low.same comment.
weakened. It was correct here and should keep firing.
Context
Filed from the same session as #771 (shell control-flow FP), #772
(move-awareness), #773 (Elder cancel/re-enqueue deadlock) and PR #774 (base-sha
churn cancelling in-flight reviews).
Size: S
Refs #707