Label gaps

228 interactive elements had no derivable label at capture time.

Read the caveat below before drawing a conclusion from that number.

All surfacesbuild.maindialog.add-connected-agentdialog.add-knowledgedialog.add-knowledge.advanceddialog.add-mcp-serverdialog.add-skill.blankdialog.add-skill.uploaddialog.add-tooldialog.add.microsoft-iqdialog.settings.agent-detailsdialog.settings.ai-behaviordialog.settings.greeting-promptsdialog.settings.safety-accessevaluate.mainmonitor.mainpreview.mainAll tagsdiviinputAll roleslinknonerowswitch228 of 228 label gaps

Surface ↑	Testid	Tag	Type	Role
build.main	—	i	—	none
build.main	—	i	—	none
build.main	—	i	—	none
build.main	—	i	—	none
build.main	—	i	—	none
build.main	—	i	—	none
build.main	—	i	—	none
build.main	—	i	—	none
build.main	—	i	—	none
build.main	—	i	—	none
build.main	preview-chat-end-user-toggle	input	checkbox	switch
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	preview-chat-end-user-toggle	input	checkbox	switch
dialog.add-connected-agent	—	i	—	none
dialog.add-connected-agent	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	preview-chat-end-user-toggle	input	checkbox	switch
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge	—	i	—	none
dialog.add-knowledge.advanced	—	i	—	none
dialog.add-knowledge.advanced	—	i	—	none
dialog.add-knowledge.advanced	—	i	—	none
dialog.add-knowledge.advanced	—	i	—	none
dialog.add-knowledge.advanced	—	i	—	none

Page 1 of 5

This is not an accessibility audit

derivedLabel is a heuristic, not a strict accessible-name computation per the ARIA specification. It checks visible text, aria-label, aria-labelledby, title, placeholder, and associated label elements — but it does not implement the full accessible-name algorithm, including its traversal rules, its handling of hidden subtrees, or its CSS-generated-content cases.

Two consequences:

False positives. An element listed here may have a perfectly good accessible name that the heuristic did not compute. Reporting these 228 as accessibility defects would overstate the evidence.

False negatives. An element not listed here may still fail a real accessible-name check. A label derived from a stale title attribute satisfies the heuristic and fails a user.

So this is a lower bound on investigation, not a defect count. It is the list of places to point a real accessibility tool, not the output of one.

If you want an actual accessibility result, run axe or an equivalent against the live surfaces. This table tells you which surfaces to prioritize.

What it is good for

Automation targeting. An interactive element with no label and no test id cannot be targeted by role and name, and cannot be targeted by test id. It must be reached positionally or through a structural locator, both of which are brittle. These are the elements most likely to make an automation suite flaky.

Cross-reference with the element explorer:

const untargetable = elements.filter(
  (e) => e.isInteractive && !e.testId && !e.label,
);

Surface prioritization. Facet by surface. A surface with a disproportionate share of the 228 is where both accessibility review and automation hardening pay off most.

Component identification. Facet by tag and role together. A cluster of unlabelled elements sharing a tag and role usually traces to one reused component rather than 30 independent oversights, which makes the fix one change rather than 30.

Casing note

label-gaps.json keys test ids in lowercase, while testids.json preserves source casing. Any join across the two must normalize:

import { normalizeTestId } from '@raybot/object-model';

This inconsistency is in the source data, not introduced by the export. It is documented rather than silently corrected, because a consumer reading the raw files directly will hit it.

Facets available

• surface — which of the sixteen surfaces
• tag — HTML tag name
• role — ARIA role, with none for elements carrying no role

Combining role and surface is usually the most informative view. An unlabelled button on a frequently used surface is a different priority from an unlabelled div with tabindex on a surface reached once.

Why capture this at all

The audit could have reported only what it could name. Recording what it could not name is more useful, for the same reason the dumpFileExists column exists on the surfaces page: an audit that silently drops what it failed to resolve is indistinguishable from an audit that found nothing to drop.