// nuwright
◔ building · in training; not yet released

sen

A small judgment model that answers typed questions about a message with probabilities, built to run on an ordinary server CPU.

job
typed judgment on CPU
domain
care · judgment model
runs
server CPU
class
judgment model
scale
target: < 1 GB, p95 ≤ 150 ms on CPU
since
2026

One call

  1. 01packThe message and its questions packed into one sequence
  2. 02encodeA small encoder reads them together
  3. 03scoreA probability for each question
  4. 04decideCode applies the thresholds and makes the call

In · out

inputs
  • a message
  • the typed questions to answer about it
outputs
  • a probability for each question

Where the person is

The model can raise an alarm but never lower one. Code makes the final call.

code decides● yes · the final say is deterministic

What is measured

Against: Planned bar: at least as good as Jev, a much larger judgment model, on externally labelled data.

No result yet. Nothing on this page is a result until the status says otherwise.

Stack · links

Python
last updated 2026-10-05 · from source, not marketingbuilding · no result is claimed