CalibratedDecisions.

Repo · Coding & code review

jevseek

DeepSeek proposes next-token candidates; TypeSafe Jev (via OpenRouter System One) chooses which token to append—turning a decision model into a…

Open github.com ↗

From the repository

DeepSeek proposes the next token, TypeSafe's Jev chooses it: a decision model used as a sampler

deepseekjevllmsamplingtypesafe

How builders describe it

DeepSeek proposes the next token, TypeSafe's Jev chooses it: a decision model used as a sampler.
DeepSeek proposes next-token candidates; TypeSafe Jev (via OpenRouter System One) chooses which token to append—turning a decision model into a sampler.
Experiment turning Jev into a sampler: DeepSeek proposes top-k next tokens with logprobs and Jev picks each token, also judging when to stop or cut rambling.

The decision Jev makes

Label a change, score its risk, pick the relevant file, or decide whether a coding agent should continue.

Where it fits

Code review and coding agents are full of small questions: is this change risky, what kind of fix is this, which file matters. These projects answer them with Jev and route the rest to people or bigger models. All 450 coding & code review projects →

Related projects