Honest by construction
A few passing checks are weak evidence, so every dimension starts from neutral and only broad evidence moves it. Automated baselines can't score above 90; higher needs an analyst-reviewed investigation.
Deterministic checks run first. Every finding points to the evidence behind it. Where we can't see enough, we say "insufficient evidence" instead of guessing. The score is a decision-support signal, not a prediction of success, revenue or funding.
Your public pages, fetched through an isolated, SSRF-safe browser.
Deterministic checks: the same evidence always gives the same result.
Plain-language findings, each pointing to the evidence behind it.
Anything not backed by evidence is dropped before scoring.
Weighted, versioned scoring with honest handling of thin evidence.
The report is stored immutably with its methodology version.
Only your workspace hears about it. Nothing is published.
A few passing checks are weak evidence, so every dimension starts from neutral and only broad evidence moves it. Automated baselines can't score above 90; higher needs an analyst-reviewed investigation.
A dimension needs at least 50% of its checks to have evidence before it is scored. Unscored dimensions are excluded and reported, never filled in.
Each report stores its methodology version, evidence hashes and check results, so its score can be recomputed exactly. Changes ship as new versions; old reports never change.
Only your workspace sees your score and report. It never appears on public pages, in search, in ranking, in our API or in notifications.
Weights are a proposed default, to be calibrated with the founding cohort. Confidence reflects evidence type: observed (1), third-party (0.8), founder-reported (0.6), inferred (0.4).
Scores are only comparable within the same methodology version and category. When too little can be seen, the result is "Insufficient evidence" rather than a number. We don't publish score leaderboards. Want a score above 90? Get investigated.