Moderate community messages with typed probabilities

01 · Role in the system

What Jev does here

Each message becomes a batch of yes/no Noul questions - one per category plus any rule the owner wrote in plain English - answered through the official typesafe_sdk with retry policies and request chunking that respects the API's input limits, about $0.04 per 1,000 messages with all categories on. Code owns the policy: thresholds decide take-down, flag or ignore, and every decision is logged with its probabilities so a moderator can audit why. Adapters connect the same judge to Discord, Twitch, YouTube and Reddit bots, an MCP server exposes the judge to agents, and the benchmark directory documents what the API can and cannot do - including measured token ceilings per language found by probing the live API on 2026-09-27.

02 · Control boundary

Where Jev sits

Jev as the labeling layer of a moderation pipeline: one typed Noul per category or custom rule, code turns probabilities into policy actions against operator-set thresholds, and every verdict is logged with its numbers for audit.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• Built-in category list is the author's; custom rules are plain-English Nouls whose recall is not benchmarked.
• Thresholds and actions are operator policy choices; no measured false-positive rates are published.
• The reasoning benchmark is self-run; platform adapters need each platform's own credentials.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• ohernandezdev ↗Community · github · public · checked 2026-09-28