Every experiment, every verdict — including the ones we lost
Every change we make to any site we operate is pre-registered with a falsifiable hypothesis in an append-only log before it ships. Each gets a verdict from real Search Console / GA4 data on a fixed date. Nothing is edited retroactively; failures stay published.
The pillar we had to kill
We shipped two AI-written pillar pages with no human layer. Google’s spam system ate one of them within weeks. That failure is now a rule.
What we did: Two new pillar pages (PPC Florida, AEO) plus a full homepage reframe, shipped as raw AI output — fast, complete, and unreviewed.
What happened: The homepage reframe and the wedge page did their job (both remain live, both validated later). The AI-raw pillar was killed by the spam update — dead traffic, dead ranking, page pulled.
We corrected our own baseline by 30×
Our own plan said the site was getting 11 clicks a day. A 40-day daily-series pull said it was actually 0.32. We published the correction.
What we did: While grading week 1 of the plan, we recomputed the baseline from GSC daily series instead of reusing an earlier summary figure.
What happened: The real baseline was ~30× worse than the number we had been carrying: 0.32 clicks/day, not 11. The earlier ’11 clicks/day’ had quietly measured cumulative clicks over a window, not daily average.
Our contact form was broken since inception
The form every visitor hits was rejecting submissions it should have accepted — from day one. We found it with our own audit funnel, the one we sell.
What we did: Server-side validation audit of the site’s own form. Found a phone-number pattern that silently rejected inputs containing spaces. Fixed the same day.
What happened: Two honest findings. First: the fix was verified end-to-end syntactically. Second — and more useful: 28 days later, GA4 shows 13 form-starts, zero form_submit events, and the chat widget out-performing the form (5 opens, 3 messages sent). The verdict we logged: INCONCLUSIVE by volume. Not a failed fix — a funnel without traffic.
The wedge page hit a wall — we’re publishing the wall
Two months of weekly bucketed data show exactly where the ceiling is: not content, not keywords — external authority. Zero new links in 60 days.
What we did: Shipped a product-led page targeting the category keyword ‘AI website builder for roofing contractors’ (validation: passed 09-30). Later, an internal-link fix took the page from orphan (0 impressions) to linked and served.
What happened: The fix did exactly what it promised: 12 straight weeks of sustained visibility at positions 8–14 (28d: 1,049 impressions at p13.9). Then the plateau — flat impressions for weeks. The bucketed data says why: nothing else was added. Zero new external links in 60 days. This is an authority ceiling, not a content ceiling.
Our #1 query went to zero. We think we know why.
The site’s top query dropped from 25–41 impressions a week to zero for the entire month. The page hadn’t changed. Its crawl date had: 89 days stale.
What we did: Pulled the full history: the query had been stable for 12 weeks (p13–18), then the serving page’s freshness collapsed. Diagnosed as crawl staleness, not content decay. Pre-registered the hypothesis, then executed a Request Indexing through Google’s owner console — the page entered a priority crawl queue the same day.
What happened: Pending by design. The re-check is dated (Nov 3, 2026) and registered. Either the refresh recovers impressions, or it doesn’t — and the recovery (or its absence) becomes measurement.
The same discipline, on client sites
Anonymized, because consent is a feature: how the ledger looks outside our own site.
What we did: Four representative situations from the client portfolio this month, all handled by the same registry-verdict loop.
What happened: A food-industry client doubled clicks to 45/28d. A contractor client’s new page captured 139 impressions at position 17 in its first 28 days — interim-positive-cautious, final verdict in December. An industrial-supply domain migration showed a −14% combined dip: we verified the 301s live before concluding no audit was needed (re-check scheduled). A dental client with a −23% freshness decline: we recommended pausing new content entirely and starting reactivation with conversion tracking — and the pause stands until the owner approves.
We don’t sell reports — we publish ours, with the losses in them. The next Ledger drops with the November verdicts.
Pull your own numbers in 60 seconds: adwolvesmedia.com/map-gap
