All capabilities

Classification · Catalog selection

Navigate a decision tree

Use jev-tree to choose a known category with a bounded call budget.

What this is worth testing

Your agent keeps the larger task and any action that follows. System One asks Jev for a typed answer on this check.

Research value
MediumEvidence confidence is low. This rating describes the research record, not a live Jev probability or a measured saving.
Sample size
About 83 input tokensThe recipe sample length divided by four. This is a planning estimate, not a measured tokenizer count.

$0.03 at Jev's $0.042 per million list rate

$2.49 at a $3 per million example rate

$2.46 input-cost difference for 10,000 uses of this sample

Excludes output charges, host tool calls, retries, hosting, and paid plans. Equal input volume does not establish equal answer quality or savings on a fixed-price subscription. $3 per million is an example rate, not a quoted model price.

Open the shared cost comparison

When an agent uses this

System One asks Jev for a typed answer. Your agent keeps the larger task, the original evidence, and any action that follows.

Example evidence

This is the recipe sample. Replace it with the evidence from your task. It is not a measured result.

A test fails because the refresh token remains valid after revocation.

What your agent keeps

Escalate uncertain branches. A selected category does not execute an action.

Finding

The current 320-category experiment is synthetic, and a deterministic parser solved every case.

Research value: medium. Evidence confidence: low. That rating describes the research record, not a live Jev probability.

Choose a category in a large catalog

Try it

Connect System One, then ask your agent for recipe taxonomy. Inspect the input on the recipe page before you send your own evidence.

More in Classification