Jev analysis

Jev analysis: score a list, a page set or a pipeline with calibrated probabilities

Language models are good at prose. Judging two hundred items against the same criteria is a different job, and Jev exists for it.

By Mark Fulton, founder of AGI LayerUpdated September 29, 2026

The short answer

Jev by TypeSafe is a model that returns calibrated probabilities instead of prose. In AutoAGI you tag @Jev, attach a data source such as a CSV, a document or a set of pages, and say what should be judged. AutoAGI designs the evaluation, Jev scores every item, and you get a written report with a visual scorecard. It is included on every membership.

How a Jev analysis runs

Tag @Jev in the composer and attach the source: a CSV export, a document, a folder of proposals or a list of page URLs.
State what should be judged, for example "how likely is each of these leads to fit a founder-led agency of under 20 people".
AutoAGI designs the evaluation and writes the criteria it will apply, so you can see and adjust them.
Jev scores every item and returns a probability for each, not a paragraph of opinion.
You receive a written report with a visual scorecard, exportable to PDF, Word, HTML or Markdown, and you can ask for targeted revisions.

Jobs it suits

Job	What you attach	What you get
Lead qualification	A CSV of leads or a CRM export	Every lead scored against your ideal customer, ranked, with the reasoning for the top and bottom of the list
Page audits against search intent	A list of URLs and the query each should answer	A score per page for how well it answers that intent, and where it falls short
Analytics review	An export from your analytics tool	Which segments or campaigns are most likely to be worth acting on
Proposal scoring	A set of proposals and your selection criteria	A ranked scorecard you can defend to the team
Ticket triage	A batch of support tickets	Each ticket scored for urgency and category so the queue can be ordered

Jev also routes the work

Jev estimates how complex a task is before AutoAGI chooses a model. Simple parts stay on economical models, and frontier models take the demanding ones. That routing is part of why weekly usage in dollars stretches further than picking one large model for everything.

What a probability is and is not

A calibrated probability is a ranking and a level of confidence, not a guarantee. Treat a scorecard as a way to focus your attention on the items most likely to matter, then check the top and the borderline items yourself. Estimates from any model can be wrong, and your data is processed under the providers' policies described on the data handling page.

Questions

What is Jev?

Jev is a model from TypeSafe that returns calibrated probabilities instead of prose. AutoAGI uses it to score lists of items against criteria you state, and to estimate task complexity when routing work across models.

How do I use Jev in AutoAGI?

Tag @Jev in the composer, attach a source such as a CSV, a document or a set of pages, and say what should be judged. You get a written report with a visual scorecard.

Does Jev cost extra?

No. Jev analysis is included on every membership, drawing from the same weekly usage allowance in dollars.

What can I score with Jev?

Common jobs are lead qualification, page audits against search intent, analytics reviews, proposal scoring and support ticket triage.

How is Jev different from asking a chat assistant to rate a list?

A chat assistant answers in prose and tends to apply criteria unevenly across a long list. Jev returns a probability for every item against the same criteria, so the whole list can be ranked and compared.

Keep reading

Related comparisons and guides

Pricing