Case studies¶
Every framework in this book was extracted from real work on four properties between June and August 2026. These pages are the dated records of that work: what shipped, what broke, what the numbers actually said, and what remains unmeasured. If the rest of the book tells you what to do, this part shows the frameworks operating under real constraints — platform bugs, owner decisions, Google's actual behavior, and the limits of agent automation.
How to read these¶
Each case study follows the same shape:
- "What this proves" — the claims this case is evidence for, up front.
- Timeline — dated, chronological entries. Incidents are marked with warning boxes; they are the most valuable parts.
- The numbers — only real, recorded figures: baselines captured before changes, counts, validation results. Anything not measured is explicitly marked unmeasured — this book does not round up.
- What worked / What failed / What we'd do differently — the honest ledger.
- Chapters this case feeds — where each lesson became a reusable chapter.
Three doctrines govern these pages:
- Honesty. The book preaches authenticity as a ranking strategy (authenticity audits), so it applies the same standard to itself: failures are documented with root causes, and outcomes that postdate the record are not claimed.
- Append, never rewrite. Each case is a point-in-time record. Later developments get appended with dates; the original narrative stays as written.
- Pattern-level detail. No credentials, no customer data, no internal infrastructure specifics. Implementation detail appears only where it is already public (largely via ever-just/agentskills) or where the pattern itself is the lesson.
The four cases¶
| Case | Type | Window | Core lesson |
|---|---|---|---|
| customdomain.ai | SaaS product (human buyers + AI agents) | 2026-07-01 → 07-21 | Intent and agent discovery must be engineered on purpose — and verified at the layer that actually serves users, because the differentiated path can be quietly broken while everything looks shipped |
| Heads Up Outdoor Services | Local service business | 2026-07-09 → 07-30 | The two catastrophic problems were infrastructure defaults invisible from inside the CMS; the durable local strategy is verifiable honesty |
| everjust.app tenants | Multi-tenant CMS platform | 2026-07-02 → 08-03 | A Host-rewriting reverse proxy makes a CMS misread its own identity — robots, noindex, sitemaps, and canonicals then fail as a class, hidden behind layered caches |
| brogav.com | Third-party traffic estimation | 2026-06-12 → 06-13 | Free, open sources triangulate a defensible traffic estimate — if every number carries a confidence tier and micro-traffic findings change the strategy, not just the report |
What each one proves¶
customdomain.ai is the book's origin story for the branded-vs-intent reframe and its extension to a second audience: AI agents. It proves that a product can have every discovery surface formally shipped — schema, registry listing, OAuth discovery — while the agent usability path is broken in ways only an adversarial audit finds. It also supplies the SERP tokenization crisis (a brand name Google parses as two generic words), the maximal schema @graph worked example, and the GitHub-as-discovery funnel.
Heads Up Outdoor Services is the local stack end to end: LocalBusiness schema, policy-safe reviews, service-area pages that survive scrutiny, and authenticity remediation. Its two headline incidents — a WAF challenge that erased the domain from search, and a sitewide noindex that shipped for 11 days — are why this book insists you test the public surface the way crawlers see it.
everjust.app tenants is the platform view of the same period: the reverse-proxy trap catalog discovered one incident at a time, the domain-cutover method that preserves email zones, and IndexNow as a platform feature. Read it if you run any site behind a proxy that rewrites the Host header.
brogav.com is the smallest case and the purest method: estimating a micro-traffic site's reality from free sources with explicit confidence tiers — and what an honest ~1,200-visits-a-month answer means for strategy.
Related¶
- How discovery works in 2026 — the model these cases validated
- Measurement and baselines — why every case records numbers before changes
- Playbooks — the same sequences, generalized into runnable form
- Skill index — the reusable skills distilled from this work