The scan lifecycle
Eight steps between a pasted URL and a re-scan that proves the fixes landed. No black box — every step below names its actual mechanism.
8 steps · ~90 seconds median · evidence captured at every step
Step 01 · intake
Whatever you paste — bare domain, stray fragment, uppercase host — gets normalized to a canonical https URL before anything touches the network.
If the site won’t answer, the scan says so and falls back to a clearly labeled demo fixture. No fake results, ever.
you paste
Marlow-Dental.example/#services
the scanner fetches
https://marlow-dental.example/
Step 02 · crawl
Homepage first, then up to nine more same-origin pages picked from its links, prioritizing the routes visitors actually use: about, services, pricing, contact. Four fetches at a time under a twenty-second budget, with a robots.txt probe on the way in.
The crawler identifies itself as AgencyProofBot/1.0. Pages that fail get recorded as failures, not fatal errors — the scan continues with whatever answered.
Link Mole, Broken Link Unit. Still walks every route.
Step 03 · checks
Each crawled page becomes a structured snapshot — titles, headings, links, images, schema, forms — and the 43 checks read it in one instant pass. Every finding is filed with its evidence attached: the exact tag, filename, or missing line.
Same input, same findings. This layer has no opinions.
from the check roster
Step 04 · ai review
A model reads the copy for what a measurement can’t catch: unbacked claims, template text, language that may need review. The default provider is mock — a heuristic pass with no API key and no page content leaving the machine.
Every finding from this layer is labeled source: ai in the report, and a provider failure never kills a scan.
provider config
AI_PROVIDER = mock | openai | anthropic
on provider error → falls back to mock, scan continues
The homepage welcome copy is the page builder’s untouched output
marlow-dental.example/ · content · high
Step 05 · scores
Every category starts at 100. Each finding deducts by severity, scores floor at zero, and the Launch Readiness Score is the weighted average of all ten categories. That’s the whole model — transparent on purpose, so a client can check the arithmetic.
The ten category weights are published on the product page.
Every scan ends in one of two ways: a stamp, or a list. Both are useful.
deduction model · per finding
| Severity | Deduction |
|---|---|
| Critical blocker | −25 |
| High priority | −15 |
| Medium priority | −8 |
| Low priority | −4 |
| Enhancement | −2 |
category = max(0, 100 − Σ deductions)
launch = Σ (category × weight)
Step 06 · report
Ten sections, from executive summary to a plain-language client note calm enough to forward. Compliance-sensitive wording is flagged for review, never presented as legal advice.
On Studio plans the report is white-labeled — your logo, name, color, and footer.
contents · ten sections
cover carries run metadata: pages crawled · duration · engine AgencyProofBot/1.0 · 43 checks
Step 07 · export
One click in the report viewer copies the complete fix prompt — every finding sorted by severity, with evidence, fixes, and approved rewrites — formatted as instructions for Claude or any coding agent. The report itself exports as Markdown.
export paths
Step 08 · re-scan
Run the same scan again and the score history shows the move. In the demo workspace, Marlow Dental went from 71 to 94 in nine days — and the before/after is the client communication.
marlow dental · demo workspace
Nine days, three critical blockers cleared. The full report is public, fictional client and all.
The demo scan finished in 88 seconds. The median real scan finishes in under two minutes — and running it twice is the whole point.
paste → crawl → checks → report · re-scan after fixes