I wrote a 403-check quality audit for a static site in ~130 lines of Node — zero dependencies

Static sites rot quietly. You add a page, forget a meta description, break the nav on one file, your JSON-LD stops parsing — and nothing tells you. Lighthouse is great but heavy, needs Chrome, and won't check your site's specific contracts (like "every tool page must link the privacy policy").

So I wrote a single-file audit script: 403 deterministic checks, ~130 lines, zero dependencies, runs in 3 seconds offline. Here's the approach.

The idea: checks as regex + tiny helpers

No headless browser, no HTML parser. For a static site you control, tolerant regex over the raw HTML covers 95% of what matters:

function check(file, name, ok) {

  total++;

  if (ok) passed++;

  else fails.push(`${file}: ${name}`);

}



const title = get(/<title>([^<]*)<\/title>/i, raw);

check(f, 'title 15-65 chars', !!title && title.trim().length >= 15 && title.trim().length <= 65);

check(f, 'canonical', /<link\s+rel="canonical"/i.test(raw));



// a11y: every input needs a label

const inputIds = [...raw.matchAll(/<(?:input|select)[^>]*\bid="([^"]+)"/gi)].map(m => m[1]);

const labelFors = new Set([...raw.matchAll(/<label[^>]*\bfor="([^"]+)"/gi)].map(m => m[1]));

check(f, 'every input has a label', inputIds.every(id => labelFors.has(id)));

Checks that caught real bugs

JSON-LD must parse. I once pasted a schema block after </head>. Valid-looking page, dead structured data:

const ld = [...raw.matchAll(/<script type="application\/ld\+json">([\s\S]*?)<\/script>/gi)];

check(f, 'JSON-LD parses', ld.length > 0 && ld.every(m => { try { JSON.parse(m[1]); return true; } catch { return false; } }));

WCAG contrast from your own palette. Parse :root hex variables, compute relative luminance, assert 4.5:1.

Link-grid consistency. Every page must share the identical nav set, and the homepage must link every tool page. I was maintaining this by hand and missed one — now it's a check.

Thin-content guard. Word-count per page; under 300 words a tool page fails. Ad networks reject thin pages, and so do readers.

The rule that made it work

Every real bug becomes a permanent check, same day. Misplaced schema → structure check. Hand-checked navigation → link-grid check. Slow bloat → 15 KB/page byte budget. The audit grew from 114 to 403 checks while the site stayed at 100%.

Result

• 9 pages (bilingual portfolio + 6 calculator tools), ~60 KB total, no frameworks


• one command → METRIC quality=100, 3 s, fully offline


• Live: https://trixer666.github.io (calculators in Polish — the audit pattern is universal)


• The automation/radar code that feeds my freelance pipeline: https://github.com/trixer666/income-radar
  

Questions about specific checks welcome — happy to paste more in the comments.