Numbers you can re-run yourself.
Businesses call me after something went wrong. What went wrong is nobody else's business. So you will not find client names on this page, ever, and that is part of what you are hiring, not a gap in the record. What you will find: measurements, the free public tool that produced each one, the date it was taken, and instructions for checking it yourself.
The Greg Drake Digital evidence page documents anonymized client work with before-and-after measurements from public tools on stated dates: a Sonoma County website rebuild measured June 5 and July 6, 2026, a same-evening site rescue, a multi-site review, and this site's own readings, published so anyone can verify them.
One Sonoma County business, before and after.
A Sonoma County business needed a permanent home after an emergency; the story of that emergency is just below. I rebuilt the site from the ground up: no database, no plugins, no admin login to steal, every change recorded so it can be undone, and labels in the page code that tell Google and AI tools exactly who the business is (structured data). Then I put the old site's numbers next to the new site's numbers, using tools anyone can run for free.
Google's public speed test (PageSpeed Insights). Mobile score out of 100. The range covers repeated runs, because lab scores vary a few points each time.
Time until the main content appears (a measurement Google calls Largest Contentful Paint). Past about three seconds, people start leaving: Google's own research measured 53% of mobile visits abandoned there (published 2016).
The security instructions a site sends every browser, graded A to F at securityheaders.com. Most small sites send none; that is my judgment from the audits I run.
Page weight. Five times lighter, which is most of why the new site feels instant on a phone.
Measured June 5 and July 6, 2026
The speed figures come from Google's public speed test (PageSpeed Insights, which runs a tool called Lighthouse); the security figures come from live header checks. The 89 to 94 range reflects normal run-to-run variance in mobile lab scores. Every figure can be re-run, and I will re-run any of them in front of you.
One more number from the same two dates: the old site loaded 46 separate scripts, pieces of code fetched from around the web before the page could settle. The new site loads 1. The count is one more thing I will re-run in front of you.
The citation flip. On July 3, 2026, ChatGPT, Perplexity, and Google's AI all recommended this business's competitors when asked; the business was absent from the answers. Within 24 hours of the July 4 relaunch, ChatGPT was citing the new site in its answers. I do not promise that result to anyone, and nobody honestly can. I recorded it because it happened, and because it shows what can follow when the machines can finally read a site clearly.
The site that came back the same evening.
Situation. That same Sonoma County business's hosting company deleted its server and confirmed it could not be restored. Its entire web presence, gone in an afternoon.
Stakes. No website, no inquiries, and nothing to rebuild from, except that a recent audit had recorded the site's structure and content outside the server.
Action. Rebuilt from audit records and public archives and relaunched the same evening. Then the permanent rebuild measured above: no server to delete, no database, no plugins, no admin login, machine-readable facts on every page, every change recorded so it can be undone.
Several brand websites, one clear remediation path.
Situation. A business group runs several brand websites on a single shared platform. I reviewed every site from the outside, the way any visitor's browser sees them: the security instructions each site sends, the labels in the code that identify the business, whether search engines can find every page, an automated accessibility check, and whether anyone was measuring visits at all. No passwords needed, nothing touched.
Findings. A genuinely well-built content layer sitting on software past its public end-of-life dates, meaning the makers stopped fixing security holes in it. Several externally verifiable gaps in the security instructions the sites send every browser. No measurement of visits anywhere. A handful of silent live defects nobody had noticed.
What they got. A severity-ranked brief with a verification date on every finding, fair credit for what was working, and a remediation sequence in which one fix covers the whole portfolio, because everything shares one platform. Fact separated from inference throughout, stated plainly where something could not be verified from outside.
This site, on the same instruments.
Anyone can publish a table of their own results. So this site is the standing exhibit: the same measurements, taken on gregdrake.com, with the free public tool for each one. Checking them takes about five minutes, costs nothing, and requires no accounts and no technical skill. Open the tool, type in gregdrake.com, compare what you see to what I published.
| What is measured | Reading | Check it yourself |
|---|---|---|
| Google's public speed test (PageSpeed Insights), mobile | 96–98 | Go to pagespeed.web.dev and enter gregdrake.com |
| Accessibility and search-readiness scores from the same test | 100 / 100 | Same run, same tool; the scores appear side by side in the report |
| The security instructions this site sends every browser (security headers), graded A to F | 6 of 6 | Go to securityheaders.com and enter gregdrake.com |
| An automated accessibility check (axe) | 0 violations, all 18 pages | Install the free axe DevTools browser extension and run it on any page here |
| The labels in the page code that tell Google and AI tools who the business is (structured data, JSON-LD) | Present on every page | Paste any page address from this site into validator.schema.org |
If any number here has drifted when you check it, email greg@gregdrake.com. You will have found something my own monitoring should have caught first, and I want to know.
Why there are no names on this page.
Businesses call me after something went wrong: a deleted server, a hacked page, a vendor holding things hostage. What went wrong is nobody else's business. Client names stay off this page permanently, and no case here carries enough detail to identify anyone. If we work together and something goes wrong on your watch, you get the same silence.
What does get published: the numbers, the public tool behind each one, and the date it was taken. I keep the raw records, I separate what I found from what I infer, and I will re-run any measurement on this page in front of you. Reports are decision support, not legal advice.
Get the same numbers for your own site.
The free baseline reading uses the same instruments as this page. You send one web address. I send back five dated measurements: Google's speed score, the time until your main content appears, the security instructions your site sends every browser, whether your code carries the labels that tell Google and AI who you are, and what one AI assistant says when asked about you, written down word for word. Each explained in one sentence. Free, and it does not obligate you to anything.