WCAG 2.0 Evidence Audit

Runs axe-core against the pages you give it, maps every finding onto the 38 WCAG 2.0 Level A/AA success criteria, and produces a client-ready evidence report (HTML + PDF + JSON) with screenshots and the exact CSS selector of every failing element — not just a raw rule dump.

What it does

1. For each start URL: validates it isn't a private/loopback address (SSRF guard), loads the page, and checks for bot-challenge / interstitial pages (Cloudflare, CAPTCHA, etc.) so you never get a false "pass" from a page that never actually loaded.
2. Runs axe-core (wcag2a, wcag2aa, wcag2a-obsolete rule tags) and aggregates the results per WCAG 2.0 success criterion into one of three statuses: 不適合 / non-compliant, 要手動確認 / needs manual review, or 適合 / compliant (automated).
3. Optionally captures a full-page mobile-width screenshot per page.
4. Writes one row per page to the Dataset, and a combined report.json / report.html / report.pdf to the Key-Value Store (linked from OUTPUT).

Why this instead of a generic axe-core Actor

Most accessibility-testing Actors on the Store dump raw axe-core rule violations as dataset rows. This Actor is built for handing evidence to a client or auditor, not just triaging rules yourself:

• WCAG 2.0 success-criterion mapping (all 38 A/AA criteria, not just the axe rule IDs) — the format auditors and clients actually expect (JIS X 8341-3:2016 / WCAG 2.0 style reporting).
• Evidence report (HTML + PDF) with the disclaimer, environment/version block (Node/Playwright/axe-core versions + a report_json_sha256 for reproducibility), and per-violation CSS selectors/help text/help URL — ready to attach to an email or deliverable, not just a JSON blob.
• Interstitial/bot-challenge detection: if the page you asked for came back as a Cloudflare challenge or similar, that page is reported as a failure with interstitial_detected rather than silently "passing" an empty shell page.

Input

Field	Type	Default	Notes
startUrls	array of { "url": "..." }	— (required)	Pages to audit
maxPages	integer	10	Hard cap 50; extra URLs are skipped (reported in OUTPUT.pages_skipped_by_max_pages)
includeScreenshots	boolean	true	Full-page mobile-width (390px) screenshot per audited page, uploaded to the key-value store
outputPdf	boolean	true	Also render report.pdf (one extra headless render at the end of the run)
reportTitle	string	"WCAG 2.0 Evidence Audit Report"	Title shown on the HTML/PDF report
locale	en | ja	en	Controls axe-core's own rule/help text language. The report's structural labels (headers, disclaimer) are Japanese in both modes — see Limitations

Output

Dataset — one row per requested URL:

{
  "url": "https://example.com/",
  "final_url": "https://example.com/",
  "status": "不適合",
  "criteria_summary": { "不適合": 3, "要手動確認": 12, "適合": 23 },
  "violations": [{ "rule_id": "image-alt", "help": "...", "selectors": ["img.logo"], "help_url": "..." }],
  "incomplete_nodes": 4,
  "screenshot_url": "https://api.apify.com/v2/key-value-stores/.../records/screenshot-0"
}

Rejected/failed URLs get { "url": ..., "status": "error", "error": "..." } instead, and are not charged.

Key-Value Store (linked from the default OUTPUT record):

• report.json — full machine-readable report (all 38 criteria × all pages)
• report.html — client-ready HTML report
• report.pdf — same report as PDF (if outputPdf is true)
• screenshot-{i} — per-page PNG screenshots (if includeScreenshots is true)

Pricing (pay-per-event)

This Actor uses pay-per-event pricing with a single event, page-audited, charged once per successfully audited page (navigation + axe-core run completed). URLs rejected by the SSRF guard, that fail to load, or that are detected as bot-challenge/interstitial pages are not charged.

Limitations

• Automated checks only. This is not a legal conformance statement or a substitute for a manual WCAG audit — many success criteria (keyboard operability, meaningful reading order, correct use of ARIA semantics, etc.) require human review and are reported as "要手動確認 / needs manual review" rather than a pass.
• Report chrome is Japanese-labeled (table headers, disclaimer text) regardless of the locale input; only axe-core's own rule/help text follows locale. Full bilingual templating is a possible future improvement.
• Pages that require login, JavaScript-gated content beyond initial page load, or that block headless browsers (some bot-detection products) will show up as error / interstitial_detected rather than a real result.