Norn Labs
An independent project

Tools that know
what they don't know.

Norn Labs builds software that grounds its decisions in measured evidence — and labels the difference between what it measured, what it assumed, and what it guessed.

NornPulse

An autonomous short-form video engine. Six agents turn a long video into vertical shorts, and every creative choice — the hook, the cut, the captions, the tags — is scored against real outcomes rather than style-guide advice.

The problem it addresses

Nearly all short-form advice was measured on channels that already have an audience. Applied to a channel that doesn't, some of it isn't merely weaker — it reverses.

Captions · 0–100 subs −4% reach Measured within the size band
Captions · 100k–1M subs +34% reach Same decision, opposite answer
Grounded against 4.56B Real videos, ClickHouse public dataset

The same scrutiny gets applied inward. Checked against real small channels, the population benchmark overstated their actual reach by about 10× — because a public crawl only contains videos discoverable enough to have been crawled. Forecasts are corrected against a channel's own history instead of shipped raw.

Open the demo View on GitHub

How we build

Three habits, applied to everything built here.

  • Provenance over confidence. A measured figure and a typed-in assumption are different claims, and presenting them identically overstates the weaker one. Every output says which it is.
  • Falsifiable before persuasive. Predictions are recorded before the outcome is known, then graded against what happened.
  • Refuse rather than guess. Below a usable sample size, the honest answer is that there isn't one.

Contact

Questions about NornPulse, or anything else here.

contact@nornlabs.ai