Where the lawsuit-risk score comes from.

Accessibility score (0–100)

Unified headline score. A weighted ratio of axe-core passes vs open failures — user-status-aware: marking an issue done increases the score without a rescan.

weighted_passes   = nodes_passed_total × 3
weighted_failures = Σ count_open(impact) × weight(impact)
  where weight: critical=10 / serious=5 / moderate=2 / minor=1
accessibility_score = round(100 × weighted_passes / (weighted_passes + weighted_failures))
Returns null if denominator = 0 (degenerate scan).

• Only open issues (status = open) contribute to weighted_failures. Issues marked done or ignored are excluded — triage directly improves the score.
• nodes_passed_total is the cumulative count of DOM nodes that passed an axe rule, captured at scan time.
• The passes coefficient (×3) approximates the average severity weight of a passing test. Calibrated empirically on reference scans.
• Single exposed score. Scans created before the deployment show null until the next recompute (triggered by a manual issue triage).

Legal coverage %

Engine availability ratio — the share of a legal reference for which an automated rule exists. NOT the share of your site that passes.

legal_coverage_pct = count(rules_with_engine_coverage) / count(rules_in_reference) × 100

• A high coverage % means the engine can test many rules of that standard — it does not mean your site is compliant on those rules.
• Coverage is reported per standard (WCAG 2.2 AA, RGAA 4.1, Section 508, EN 301 549) and aggregated weighted by the count of testable rules in each.
• Manual-only criteria (e.g., WCAG 1.2.2 Captions accuracy) are by definition NOT in coverage.

WCAG principle subscores (Perceivable / Operable / Understandable / Robust)

Four per-principle compliance rates derived from the issues found, mapped via WCAG SC → principle.

subscore(principle) = 1 − (failed_issues_in_principle / testable_rules_in_principle)
where failed_issues excludes manual_required and needs_review

• Issues tagged with multiple WCAG principles (e.g. an issue mapped to both 1.3.1 Perceivable and 4.1.2 Robust) increment each principle independently. The four subscores can therefore sum to more than 100% of the total issue count — by design.
• manual_required and needs_review outcomes are excluded from the failed count to avoid penalising criteria the engine cannot decide.
• Used for diagnostic radar charts only — not for the headline conformity score.

Lawsuit risk — Litigation likelihood (0-100, per issue)

Per-issue likelihood score blending impact × WCAG multiplier × jurisprudence boost when historical lawsuit data exists for that rule.

base = impact_base[issue.impact] × max(wcag_multiplier[tag] for tag in issue.wcag_tags)
if rule_id in RULE_JURISPRUDENCE_TABLE:
  settlements_boost = min(SETTLEMENTS_BOOST_CAP, settlements_count / DIVISOR)
  median_boost      = min(MEDIAN_BOOST_CAP, median_usd / DIVISOR)
  score = round(base × (1 + settlements_boost + median_boost))
else:
  score = base
score = clamp(score, 0, 100)

• Shown in the app and in the Defense PDF (per-issue priority column).
• Not a USD estimate — it is a relative likelihood ranking used to triage which issues to fix first.
• Rules absent from the jurisprudence table fall back to legacy impact × WCAG only.

Source: RULE_JURISPRUDENCE_TABLE sourced from UsableNet ADA Web Lawsuit Trends 2025 + Seyfarth ADA Title III FY2024.

Lawsuit risk — USD exposure estimate (Defense PDF only)

Estimated monetary exposure. Surfaced ONLY in the Defense PDF, never as a standalone number in the dashboard, to avoid being read as a prediction.

raw = sum_issues( litigation_weight(rule) × impact_multiplier × volume_factor(nodes) )
exposure_usd = raw × BASE_SETTLEMENT_USD × jurisdiction_multiplier
exposure_usd = clamp(exposure_usd, LAWSUIT_RISK_MIN_USD, LAWSUIT_RISK_MAX_USD)
# BASE_SETTLEMENT_USD = $55,000
# jurisdiction_multiplier defaults to 1.0 (US)
# Caps applied: floor $20k (single-scan), ceiling $150k (single-scan)
# Hard cap $500k applies only to optimistic/pessimistic scenario bands

• BASE_SETTLEMENT_USD ($55k) = midpoint of the $25k-$75k median range reported by Seyfarth ADA Title III Year-End FY2024 for pre-litigation digital ADA settlements.
• Caps clamp the result inside the statistical band observed in 2024-2025 datasets ($20k-$150k floor/ceiling on single-scan estimate; $500k hard cap on scenario bands). Outliers are NOT extrapolated.
• The jurisdiction_multiplier is 1.0 (US) by default. Non-US jurisdictions are not yet calibrated — treat as US-only.
• This number is indicative, not a legal opinion. Re-verify annually when the next Seyfarth report ships.

Source: Seyfarth Shaw LLP — ADA Title III Federal Lawsuits Year-End Report FY2024.

Quick wins — gain_points

Triage list of issues with high impact and low effort. The `gain_points` value is a count of templating-unit fixes, NOT a delta in the score.

gain_points = number_of_unique_template_locations_to_fix

• Do NOT interpret gain_points as 'points added to the score if I fix this'. It is a count of distinct template/component patches.
• A single template fix can resolve dozens of page-level instances (e.g., one image-alt fix in a product card component).

Fix effort (hours per issue)

Indicative time-to-fix estimate per axe-core rule, used to rank quick wins. Internal estimates based on historical fix data.

effort = FIX_EFFORT_HOURS.get(rule_id, FIX_EFFORT_HOURS_DEFAULT)
# Range observed: 0.1h - 4h per rule. Default = 1.0h.

• Estimates are PER UNIQUE TEMPLATE LOCATION, not per page-level occurrence.
• Effort estimates are calibration values from internal project history. They do not account for QA, deployment, or regression testing.

Viewport gap warning (desktop vs mobile)

Flags scans where the desktop and mobile density scores diverge significantly, suggesting responsive-only failures.

viewport_gap_warning = abs(desktop_score − mobile_score) > VIEWPORT_GAP_WARNING_THRESHOLD
# VIEWPORT_GAP_WARNING_THRESHOLD = 15 (empirical project threshold)

• Threshold of 15 points is empirical (project calibration). It is not a WCAG-derived value.
• When triggered, the report adds a viewport-specific recommendation block — but does not penalise the headline score.

Regression detection (vs previous scan)

Compares the current scan against the previous scan with matching legal references, surfacing rules that newly fail.

regression = (rule failed_delta ≥ DEFAULT_REGRESSION_MIN_FAILED_DELTA)
# DEFAULT_REGRESSION_MIN_FAILED_DELTA = 1

• A delta of 1 is intentionally aggressive — even one new failing node triggers the regression flag.
• The comparison is filtered by matching legal references between scans (issue #211 fix). A scan against a different referential will not regression-compare.

Multi-page consistency — amplification

When the same rule fails across multiple pages, the penalty contribution is amplified to reflect the recurring template defect.

penalty += SCORE_WEIGHT_SERIOUS × pages_affected
# No cap currently — tracked for a future iteration.

• An issue affecting 10 pages contributes 10× the SERIOUS weight, regardless of the issue's own impact level. Rationale: a recurring template failure is by nature serious.
• No upper cap on pages_affected yet — very large sites with widespread template defects can see the headline score saturate at 0.