We graded our own website. It got a C-.

Entry 1 · August 28, 2026Published this page. The ugly number is the proof that the instrument tells the truth, including about us.

Entry 2 · August 28, 2026Fixed the critical flag: the scan now shows half your score free, unlocks the full report with your email, and texts your 72-hour fix plan when you add your phone. One field per screen. The same progressive gate the data says converts 25 to 40 percent better than long forms.

Entry 3 · August 28, 2026One front door. The homepage now asks one thing: point me at your website. Everything else stepped back. And the meta description now answers the question owners actually type: why is my website not making sales?

Entry 4 · August 28, 2026The scanner humbled us mid-climb. After our first fixes it re-graded us DOWN to D 69, because the hero still held three doors (the scan, an app link, a slider). Our own instrument caught our own half-measure in public. We cut the hero to one door, literally one field and one button.

Entry 5 · August 28, 2026Re-scan after the cut: C+, 77 of 100. From 72 to 69 to 77 in one evening, every number published as it happened. Next up, per the scanner: show a sample grade inline in the hero, surface pricing sooner, and the one thing that cannot be built: our first customer's words.

Entry 6 · August 28, 2026A truth about our own instrument, learned by pointing it at ourselves four times in one night: it wobbles. Same site, four runs, 72 then 69 then 77 then 73, with a different top leak each time. An AI audit is a sharp estimate, not a caliper, and we would rather tell you that than pretend otherwise. So from here we optimize the flags that appear in EVERY run, and we ignore the noise. The flags that never go away: no customer proof yet (only a paying customer fixes that), and one page trying to sell two products. That is the real work, and the score will follow it.

Entry 7 · August 29, 2026Highest score yet: B-, 82 of 100. What moved it was not polish, it was money plumbing. We hired four merciless outside testers (a revenue auditor, a conversion expert, a due-diligence skeptic, an AI-citability judge) to attack the live site, and they found the ugly stuff: our own scan report was sending ready buyers to a quote form instead of the checkout, people who unlocked by email never saw a price at all, and a stranger asked to prepay 490 could not find who they were paying or what happens after the charge. All of it fixed the same night: the report now hands you to the posted price with your grade attached, every unlock path shows the fix plan and the pay button, and the fix page walks you through exactly what happens after you pay, who Class Company US is, and the guarantee with third-party numbers you can verify. The arc so far: 72, 69, 77, 73, 82. The distance from 82 to the 93 we hold ourselves to is mostly one thing no code can build: the first customer's before and after.

The night shiftFrom tonight, Tarvis re-grades this site automatically at 3am Central, every night, no human involved. The freshest overnight score prints right here the moment it lands. This is the part of the machine that works while we sleep.