Most WCAG criteria resist computation, but SC 2.4.11 Focus Not Obscured (Minimum) almost invites it. The W3C Understanding document for SC 2.4.11 frames the goal as keeping “the focused item visible,” with benefits for sighted keyboard users, low-vision users using a pointer for orientation, and people with attention or short-term-memory limits.
Thesis: SC 2.4.11 as a computable contract
The thesis is that because SC 2.4.11 turns on one testable clause — a focused component must not be “entirely hidden due to author-created content” — an AI design agent can compute useful pass/fail evidence for it in the DOM before shipping UI, provided the output is labeled as evidence from a proposed, unvalidated sampling method rather than as conformance. The clause is quoted from the W3C Understanding document for SC 2.4.11.
Computable rules: the contract an agent can encode
The contract has ten rules: rules 1–4 define the pass/fail buckets, and rules 5–10 define scope, remediation, and edge cases, each anchored to exact W3C wording in the Understanding document for SC 2.4.11.
- Enumerate and focus (ours). The agent enumerates every focusable component, focuses each programmatically, then applies the hit test in the next section. The anchor is the SC text: “When a user interface component receives keyboard focus, the component is not entirely hidden due to author-created content.” (SC 2.4.11)
- Entirely hidden is the blocker (threshold ours). If all sampled points are blocked, classify the component as “entirely hidden” and treat it as failing SC 2.4.11. The anchor wording is the same: “When a user interface component receives keyboard focus, the component is not entirely hidden due to author-created content.”
- Partially obscured passes at AA. If some points are blocked and some are clear, the component passes SC 2.4.11 at AA. The document says “this AA criterion allows for the component receiving focus to be partially obscured by other author-created content.” It also says: “For best visibility, none of the component receiving focus should be obscured. This preferred outcome is covered by the AAA criterion Focus Not Obscured (Enhanced).” That preferred outcome is SC 2.4.12, which is AAA.
- Any blocked point fails SC 2.4.12 (AAA). At the AAA level, the SC text is “When a user interface component receives keyboard focus, no part of the component is hidden by author-created content.” The goal line is “Don’t cover any part of the item with focus.” The document adds that semi-transparent overlays, blur, or dimming fail: “it does fail the normative requirements of this success criterion.” (SC 2.4.12)
- Author-positioned and auto-appearing overlays are always in scope. Content “pre-positioned by the author (such as a sticky footer), or content that appears without direct user initiation, such as system warnings, must not prevent the item receiving focus from being immediately visible in the viewport.” Typical offenders are “sticky footers, sticky headers, and non-modal dialogs.” A cookie banner “will fail this success criterion if it entirely obscures a component receiving focus.” (SC 2.4.11)
- User-opened content has an escape hatch. “Content opened by the user may obscure the component receiving focus. If the user can bring the item with focus into view using a method without having to navigate back to the user-opened content to dismiss it, this criterion would be passed.” The document names Escape, scroll keys, and F6. Our proposal: the agent can apply this only if the overlay’s trigger is declared metadata; otherwise flag for human review. (SC 2.4.11)
- User-movable panels: initial position only. “only the initial positions of user-movable content are considered for testing and conformance.” (SC 2.4.11)
- Modality gate. “A properly constructed modal dialog will always pass this SC.” A non-modal overlay that does not take focus and does not dismiss on focus loss is “at risk of failing this SC, where it is positioned such that it can obscure other focusable items.” (SC 2.4.11)
- Measure the component, not the indicator. The criterion evaluates the focused component, not the focus indicator; “However, if the focus indicator is fully obscured, it would likely fail 2.4.7 Focus Visible.” Semi-transparent overlays may pass 2.4.11 but “may separately fail 1.4.11 Non-text Contrast.” The indicator itself is discussed under 2.4.7 Focus Visible. (SC 2.4.11)
- Remediation contract (what the agent emits). The preferred fix is scroll padding. The SC 2.4.11 Examples section gives a sticky-footer case that passes: the focused item is “not completely visually obscured by the footer because content in the viewport scrolls up to always display the item with keyboard focus using scroll padding” (SC 2.4.11). The SC 2.4.12 (AAA) Understanding document gives the matching failure: “a focused item is partially obscured by the footer because content in the viewport scrolls without sufficient scroll padding” (SC 2.4.12). Technique C43, “Using CSS scroll-padding to un-obscure content” (C43), is listed as sufficient under SC 2.4.12. MDN defines
scroll-paddingas “The scroll-padding CSS shorthand property sets scroll padding on all sides of an element at once. It specifies offsets that define the optimal viewing region of a scrollport within a scroll container.” and says Baseline widely available since April 2021;scroll-marginis “The scroll-margin CSS shorthand property sets all of the scroll margins of an element at once, assigning values much like the margin property does for margins of an element.” with Baseline widely available since July 2021. (scroll-padding, scroll-margin)
How to verify — our proposal
Our proposal is a DOM sampling method: focus each focusable element, sample points of its bounding rect, and ask document.elementFromPoint whether each hit is the element or a descendant; it is our proposal, not a W3C technique, and we did not run it. The SC 2.4.11 Understanding document defines the criterion, but not this method.
// pseudocode — ours, not executed
for (const el of focusableElements) {
el.focus();
const rect = el.getBoundingClientRect();
const points = samplePoints(rect); // center + inset corners
const blocked = points.filter(p => {
const hit = document.elementFromPoint(p.x, p.y);
return !(hit === el || el.contains(hit));
}).length;
if (blocked === points.length) bucket = "entirely hidden";
else if (blocked > 0) bucket = "partially obscured";
else bucket = "clear";
}
Output contract: three buckets per element — clear, partially obscured, entirely hidden. MDN also has an Element.scrollIntoView reference (MDN), but this post does not draw specifics from it.
Limitations: what this check cannot settle
The proposed check has seven stated limits, all our analysis, and none of them are settled by the SC 2.4.11 Understanding document:
elementFromPointreturns the topmost hit-testable element only;pointer-events: noneoverlays and visual-only opacity are not seen.- Point sampling does not prove full coverage between samples.
- The check measures the component, not the focus indicator (per the spec note in rule 9).
- A static DOM read cannot distinguish user-opened from author-created content.
- Sticky elements that appear only after a scroll or on a timer need a scripted interaction pass.
- User-movable panels are tested at initial position only (which does mirror the SC’s own rule).
- The method is proposed, not validated; we ran nothing.
The Bottom Line
The bottom line is that 2.4.11 is close to an ideal computable contract—one clause, “not entirely hidden,” that reduces to a hit test an agent can run before emitting UI—while conformance remains a judgment, as the SC 2.4.11 Understanding document frames the criterion. Compute the three buckets, treat “entirely hidden” as a blocker and “partially obscured” as AAA debt against SC 2.4.12 (AAA), and hand the residue—transparency, user intent, indicators—to a human. Agents can compute evidence here; conformance remains a judgment.
FAQ
What does WCAG 2.4.11 require? SC 2.4.11 Focus Not Obscured (Minimum), a Level AA criterion, says: when a user interface component receives keyboard focus, the component is not entirely hidden due to author-created content. Partial obscuring is allowed at AA. Sticky headers, sticky footers, and non-modal dialogs are the typical causes of failures.
Does a cookie banner fail it? Only if it entirely obscures the focused component. W3C’s Understanding document says a cookie banner “will fail this success criterion if it entirely obscures a component receiving focus.” Passing fixes: make the banner modal, or use scroll padding so it does not overlap content. Notifications that do not require user action can also pass by closing on loss of focus. Partial overlap passes AA but fails SC 2.4.12 (AAA).
What does the elementFromPoint check actually do?
Our proposed check focuses each focusable element, then samples the center and inset corners of its bounding box, asking document.elementFromPoint whether each point resolves to the element or a descendant. All samples blocked: entirely hidden. Some blocked: partially obscured. This is our proposal, not a W3C technique, and we have not run it hands-on.
What can’t this check catch?
The hit test cannot see overlays with pointer-events: none, the visual effect of semi-transparent overlays, or coverage between sample points. It measures the component, not the focus indicator (a 2.4.7 concern). Static DOM sampling cannot distinguish user-opened from author-created content, and overlays appearing after scroll or on a timer need scripted interaction.
How should an agent fix a failure?
Use scroll padding. MDN defines scroll-padding as offsets that define the optimal viewing region of a scrollport, Baseline widely available since April 2021; scroll-margin (Baseline July 2021) works on the element side. W3C lists technique C43, “Using CSS scroll-padding to un-obscure content,” as sufficient under SC 2.4.12 (AAA).
How This Guide Was Built
This guide is a desk analysis based on the W3C Understanding documents for SC 2.4.11 and SC 2.4.12 (AAA), and MDN’s scroll-padding and scroll-margin pages, fetched 2026-11-21. The SC 2.4.11 Understanding document, SC 2.4.12 Understanding document, scroll-padding, and scroll-margin are the sources. MDN’s Element.scrollIntoView page was saved but not drawn on. The elementFromPoint sampling method is the authors’ proposal, not a W3C technique and not a validated tool. We did not run the check hands-on and report no test results.
