attest

FlowAnalysis diffs an ordered sequence of audit reports — the output of
tester.auditFlow — so each step's introduced and resolved findings are
explicit. Instead of "here are N reports" it says "opening the dialog
introduced these two issues", matching by the stable fingerprint so a pure
layout shift between steps is not counted as a change.
introducedByInteractions excludes the initial screen's own findings.

This makes interaction-aware auditing actionable — it names the interaction
responsible. Releases attest 1.7.0 (FlowAnalysis, StepDelta) and attest_flutter
1.2.1 (requires it, documents FlowAnalysis in the flow section); attest_cli is
unchanged.

attest/generic-link-text flags links whose whole accessible name is a generic
phrase — "read more", "click here", "here", "learn more" and the like — that
tells a screen-reader user nothing about the destination, which matters most
when they pull up an out-of-context list of links.

WCAG 2.4.4 allows the purpose to be clear from surrounding context, so this is
a heuristic: a suppressible warning, scoped to actual links (not buttons), with
a declared precision bar of 0.9. It moves 2.4.4 from manual to partial in the
coverage matrix. Ships with rule fixtures, nine corpus cases (precision and
recall 1.0), and its heuristic-contract entry.

Fifteen bundled rules; corpus at 137 cases. Releases attest 1.6.0 and
attest_flutter 1.1.2 (dep bump + docs); attest_cli is unaffected.

attest/adjustable-value flags adjustable controls — sliders, steppers — that
respond to the increase or decrease action but expose no current value. A
screen reader lets the user change the setting yet cannot announce what it is,
and a visible label ("Volume") is not the value. This is the Value leg of
Name, Role, Value that the existing rules did not cover.

Deterministic, tree-only, so no Flutter-side change: the builder already
captures increase/decrease actions. Ships with rule fixtures, ten corpus cases
(precision and recall 1.0), and its entry in the coverage matrix under 4.1.2.
Fourteen bundled rules; corpus at 128 cases.

Releases attest 1.5.0 and attest_flutter 1.1.1 (which just requires the new
core and refreshes its docs); attest_cli is unaffected.

Added

• ConformanceReport: a machine-readable, clause-by-clause conformance document for a standard pack — every criterion with its coverage status, guidance and the findings mapped to it. Built with ConformanceReport.build; it is the stable substrate a dashboard or VPAT / accessibility statement consumes, with a schemaVersion bumped only on a breaking shape change.

Added

• CoverageMatrix.renderTable() now prints each criterion's guidance note under it, so the text output — and attest coverage — is an actionable checklist rather than just a list of clause numbers.

Added

• The coverage matrix: WcagRegistry (the single source of truth mapping every WCAG Level A/AA success criterion to automated / partial / manual coverage, with an own-words note per clause), CriterionCoverage, CoverageStatus, and CoverageMatrix.forStandard. It states, per pack, exactly what attest verifies automatically and what needs human review — no new detection, no false-positive risk. A consistency test keeps it aligned with the bundled rules.

Added

• attest/non-text-contrast (WCAG 1.4.11): icons and other non-text glyphs are held to the flat 3:1 minimum instead of the text minimum. ContrastSample gained an isNonText flag; the contrast rule now skips non-text samples.

Fixed

• A small icon rendered below 4.5:1 but at or above 3:1 is no longer reported as a text-contrast failure. Icons are non-text content (governed by 1.4.11), so holding them to the text minimum was a false positive.

First stable release. The public API is frozen (see below) and the rules ship with measured, published precision and recall against a validation corpus.

Added

• SemanticsNodeData.identifier and Finding.identifier, carrying the developer-assigned semantics identifier (SemanticsProperties.identifier). A finding resolves to the offending node's identifier, or its nearest ancestor's.
• ContrastSample.identifier, so a contrast finding is anchored to the sampled text node the same way tree-rule findings are.
• The validation-corpus API (package:attest/corpus.dart): CorpusCase, ExpectedFinding and CorpusCategory, the labelled ground-truth types the precision/recall harness measures each rule against.
• MetricsHarness and CorpusMetrics: run the corpus, compute per-rule precision, recall and false-positive-on-clean rate, and evaluate a CI gate (deterministic precision must be 1.0, zero false positives on clean cases, heuristics must meet a declared threshold, and recall must not regress against a committed baseline).
• declaredHeuristicPrecision: an explicit precision bar (0.9) for each of the four heuristic rules, enforced by the metrics gate. A contract test verifies every heuristic declares a bar, tags its findings heuristic, and is suppressible in one line via RuleConfig.disabledRules.
• Determinism and performance guarantees are now regression-tested: identical input produces byte-identical output regardless of rule registration order, a pure layout translation leaves every fingerprint unchanged while a genuine violation change moves exactly one, and a 2000-node screen must audit within an explicit per-screen time budget.

Changed

• The public API is frozen for 1.0. Everything exported from package:attest/attest.dart is stable within a major, with two documented exceptions annotated @experimental (free to change in minors, always changelogged): the validation-corpus library (package:attest/corpus.dart) and TranscriptGenerator, whose announcement wording awaits cross-validation against real screen readers.
• Stated the version-support policy: Dart SDK ≥ 3.6 (a tooling floor, not a language-feature requirement — documented in the pubspec), and toolkit-wide support for the current and previous three stable Flutter releases.

Changed

• attest/state-exposed now fires on a group of two or more custom controls (down from three), catching small segmented/toggle pairs that expose no selected state.