Case studies

The case study is this site. Every iteration is annotated-tagged in the git history with the gate scores at that iteration. This page summarizes them; the authoritative source is git log --oneline --decorate --graph.

iter-0-baseline — Cloudflare defaults floor

Tag: iter-0-baseline · What changed: placeholder index deployed to agent-ready-poc.pages.dev

The first measurement: a single-page placeholder, no JSON-LD, no _headers, no robots.txt. Establishes which points come free from Cloudflare's defaults versus what the build adds.

• Performance: 100 · Accessibility: 94 · Best Practices: 100 · SEO: 92
• axe-core: 0 violations · W3C HTML: 0 errors
• LCP: 800 ms · CLS: 0 · INP: 0

Free wins identified: HTTPS, HTTP/3, referrer-policy, x-content-type-options, access-control-allow-origin: *.

Gaps identified: missing _headers, robots.txt, llms.txt, sitemap.xml, content-signals, JSON-LD entity graph, WCAG 2.5.8 touch-target sizing.

iter-1-defaults — discoverability and bot access

Tag: iter-1-defaults · What changed: added the Cloudflare-can't-infer-this files

Added _headers (CSP, HSTS, Permissions-Policy, content-signals), robots.txt (non-blanket AI bot rules), llms.txt (page list), sitemap.xml (generated from eleventy collections), JSON-LD entity graph stub in base.njk, and a CSS touch-target fix for WCAG 2.5.8.

• Performance: 100 · Accessibility: 100 (+6) · Best Practices: 100 · SEO: 100 (+8)
• axe-core: 0 violations · W3C HTML: 0 errors
• Gate 2: GREEN at ceiling

iter-2-structure — full page tree

Tag: iter-2-structure · What changed: scaffolded 11 crawlable pages + 1 404

Added /about, /services (index + 3 detail pages), /case-studies, /how-we-engage, /blog, /faq (with FAQPage schema), /contact, /404. Created a nav partial with aria-current handling, added a Nunjucks startswith filter for prefix matching. Lighthouse harness updated to run both desktop and mobile per Gate 2 spec.

• Per-page Lighthouse: 100/100/100/100 on both mobile and desktop, across all 11 pages
• axe-core: 0 violations across all pages
• W3C HTML: 0 errors across all pages
• Gate 2: GREEN on the full page tree

iter-3-content — real content with citations

Tag: iter-3-content · What changed: brand selection (Option B: meta) and real content shipped

Adopted "Agent Ready POC" as the brand — meta, openly the build process. Wrote real content for home, about, three service pages, case studies (this page), how-we-engage, FAQ, contact. Each service detail page now carries 5–6 verifiable citations to industry sources, two-to-three statistics per major section, and an original named framework. No fabricated expert quotes — better to leave that Gate 3 dimension partial than to fabricate.

Custom domain agentreadypoc.com attached.

(Gate scores logged at the iter-3 tag annotation in git log.)

What the iteration log proves

Three things:

1. The agent can ship the structural and quality bar without human authoring. Iterations 0–2 are entirely structural; the operator approved each tag but did not edit code.
2. Brand and content decisions remain a checkpoint. The operator picked the brand direction at Checkpoint 4 (Option B). No autonomous brand fabrication.
3. The verification harness scales. Adding pages from 1 → 12 did not require rewriting the gate scripts; lighthouse, axe, validate, and the sitemap all picked up new pages automatically.