# Should accessibility alerts auto-resolve or require a manual close? | Accessignal

Source: https://accessignal.com/blog/2026-10-06-accessibility-alert-auto-resolve-vs-manual-close.html

[Home](../index.md) / Alert operations


Published October 6, 2026

Auto-resolving alerts keeps the queue clean but can hide flaky scans. Manual close adds accountability but creates alert debt. The right answer depends on the alert type, not on team preference.

## What auto-resolve is actually for

Auto-resolve exists for one situation: the scan can prove the issue is gone. A missing alt attribute that is now present, a contrast ratio that now passes, a form label that now exists in the DOM. When the detection is deterministic and the fix is verifiable by the same scan, requiring a human to click 'close' adds ceremony without adding safety.

The mistake teams make is auto-resolving everything the scanner can re-check, including the flaky stuff. A scan that intermittently fails to load a third-party script will auto-resolve the alert on the runs where the script happens to load, and re-fire it on the runs where it does not. The queue looks clean and the issue is still there. Auto-resolve should be restricted to checks with a stable, deterministic pass signal across at least two consecutive runs.

## Where manual close earns its keep

Manual close is for alerts where the scan cannot verify the fix: screen reader behavior, keyboard traps in dynamic widgets, anything the team fixed with a judgment call rather than a code change. It is also for alerts that were closed as accepted risk, where the record needs a human name attached to the decision.

The value of manual close is not the click. It is the forced moment of review. When a developer closes an alert, they are stating that they looked at the evidence and stand behind the resolution. That statement is worth something in an audit trail, and it is worth a lot the first time a client asks who decided a known issue was acceptable.

The cost is alert debt. Every team that requires manual close has a graveyard of alerts that were fixed weeks ago and never closed, because closing is boring. A queue full of stale alerts trains everyone to ignore the queue, which is worse than no queue at all.

## The hybrid most teams land on

Auto-resolve the deterministic checks after two consecutive clean runs. Require manual close for everything the scanner cannot verify, plus anything closed as accepted risk. And add a third path most teams skip: auto-escalate alerts that have been open longer than the team's SLA, regardless of type.

The escalation path is what keeps both systems honest. An auto-resolve rule that is too aggressive will show up as alerts that fire, vanish, and fire again. A manual-close requirement that is too heavy will show up as a growing backlog of stale alerts. Both failure modes are visible if the team reviews the alert lifecycle monthly instead of just the current queue.

## The metric that tells you the policy is working

Track time-to-verified-resolution by alert type, not just time-to-close. An alert that auto-resolved in a day but re-fired three times was never resolved. An alert that sat open for three weeks waiting for a manual close that took thirty seconds was a process failure, not a technical one.

The healthy numbers look like this: deterministic alerts resolving within two scan cycles, judgment-call alerts closing within the sprint they were fixed in, and a re-fire rate near zero. If re-fires are climbing, tighten the auto-resolve criteria. If stale alerts are climbing, the manual close requirement is heavier than the team's discipline. Adjust the policy to the team you have, not the team you wish you had.

## Frequently asked questions

### What does auto resolve mean for accessibility alerts?

The alert closes itself when a later scan no longer finds the issue. That is efficient for issues fixed by a template change, but risky when a scan simply failed to reach the page or the issue moved. Auto closed is not the same as verified fixed.

### Which alerts are safe to auto resolve?

Low severity, rule based findings on stable templates, where rescans are reliable and the fix is usually a single code change. Even there, a sample of auto resolved alerts should be spot checked monthly to confirm the closure matched reality.

### Which alerts should require a manual close?

Critical and purchase path findings, anything that was escalated, and any issue a customer or tester reported. Those deserve a human verifying the fix with the same method that found the problem, and recording who closed it and when.

### What is the best hybrid setup?

Auto resolve the minor tier, manual close for major and critical, and a weekly report of everything auto closed so a human still sees the pattern. The report is cheap, and it catches the day a broken scan starts quietly closing alerts it never checked.