Most partnership announcements describe two logos agreeing to like each other. This one started with a score.
In August, Dan Bondarev, the solo founder of Docsbook, ran his pre-seed deck through EvalLens. Six judges, unanimous verdict: medium, 5.2 out of 10. The report did what reports are for: it named the one gap every judge saw independently, the deck sells one thing while the paying customer buys another. Product and founder pulled the score up. Problem and market pulled it down, because they rested on untested hypotheses.
Dan read the report and did the rare thing. He did not argue with the judges. He asked for the room.
The trade
So here is the partnership, stated as the trade it is.
EvalLens gives Docsbook the room. Weekly founder-to-founder working sessions, run against the report: a scout-mode pitch run with interruptions, a truth file read out loud, one testable hypothesis per week, written protocol the same day. Not mentorship, not encouragement. The rules ban invented numbers and unmeasured claims, and every session gets scored on the same dimensions the panel used. AI prepared the analysis; Dan decides what to do with it, week by week, on the record.
Docsbook gives EvalLens the product. Our documentation now runs on Docsbook: repository in, branded AI-readable docs site out. We use the full plan as a demanding early user, and what we find goes straight back to Dan as the kind of feedback a founder can act on, specific, dated, and tied to a real workflow.
Both sides open their networks as results come in. Nothing in this arrangement involves either side pretending the other is further along than it is. Docsbook today is a live SaaS with working billing and one paying customer; the honest slide about that fact was the thing our judges flagged as a rare strength.
Why we do these
Every evaluation we ship ends with the same handoff: the panel prepares the evidence, a person makes the call. A partnership like this is that handoff stretched over weeks, with a founder who chose to treat a 5.2 as a work order instead of an insult. If the format works, the numbers will show it: the session regulament sets per-dimension targets and a checkpoint at session four. We will publish what the checkpoint shows, either way.
AI prepares the analysis. The founder decides. That line is easy to print on a website. This is what it looks like as a weekly calendar invite.
Update, two weeks in (September 7)
We promised to publish what the sessions produce, either way. This is the first installment.
The gap the judges named is closing in public. The report's central finding: the deck sells one thing while the paying customer buys another. Two weeks of sessions turned that diagnosis into a repositioning. Docsbook now pitches documentation that sells: a team of agents that improves your docs every day, gets your product found and recommended by Google and by AI assistants, and shows you how many customers came in as a result. That one-liner did not exist at session one. Dan built it during a role-play call, after the opening question, "so what is it?", went unanswered four times. By the end of the same day he had an answer that survives interruption.
Customers drove the pivot, not us. Between sessions Dan ran sales calls and logged the raw transcripts (a session rule: raw logs, not summaries). He found the same pattern the panel did: nobody pays for automation up front, buyers want measurable results first. The product followed. Agents now track metrics, propose changes, execute them, then verify whether the predicted impact happened. A unit-economics model replaced the hypothesis.
Our side of the trade is live too. EvalLens documentation is moving onto Docsbook. The onboarding runs as a working chat between Dan and our CTO, bug reports reach him within minutes of us hitting them, and the first cut is up at docsbook.io/evallens.
Next: a case study with us as the client. The plan on the table: run Docsbook's agents against real projects, ours included, and publish what the agents analyzed, what they changed, and whether the changes moved anything. The same standard we hold pitch decks to: predicted impact, then verified impact, on the record.
The session-four checkpoint is ahead. The numbers will say what they say, and we will print them here.