U+22A8 documentation

Score content against trait-scoring models you define from examples — deterministic, with no language model at score time.

U+22A8 scores content against models you define from labelled examples. A model encodes one or more traits — axes of judgment such as clarity or intent — and scoring returns a number per trait, a composite, and the context to interpret it. Scoring is deterministic: the same text and the same model produce the same score every time, and no language model runs at score time.

Quickstart

Get a key and score your first text in a few minutes.

Score card

What a score returns — scores, tiers, confidence, headroom.

REST API

The resource-shaped /v1 API: models, samples, training, scoring.

Console

Issue and manage API keys.

How it works

1. Define a standard

   A model is a set of traits, each taught by labelled example samples — not rules, prompts, or fine-tuned weights.

2. Train

   Training fits each trait’s geometry in embedding space and calibrates the thresholds that separate score tiers.

3. Score

   Send content to the model. It returns a score per trait with confidence and headroom — deterministically, without invoking a language model.

The public catalog under u22a8.* ships ready-to-use models. You can also author your own through the API.

Next

Quickstart

Score your first text.

Score card

Interpret what you get back.