Score video statements by viewpoint

01 · Role in the system

What Jev does here

An install wizard takes a video file or link, a preset, and an optional transcript of who speaks. Each preset is an index with its own named tells: the debate preset flags evasive answers, dodged questions, emotional appeals and self-contradiction; earnings calls flag vague guidance and blame-shifting; podcasts flag unsupported claims and outrage; pitches flag buzzwords and urgency pressure. Sentences are scored per tell and the meter is rendered onto the video, cut to 16:9 for posting. The author reports a full debate costs about five cents and badges preset accuracy at 99 percent on a held-out split under eval/RESULTS.md. The positioning is a viewpoint meter, not a fact-checker: it scores how a statement is made against the preset's rubric, not whether it is true.

02 · Control boundary

Where Jev sits

Transcript to per-sentence typed scoring under a preset rubric; deterministic rendering cuts the video and draws the meter.

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

• A viewpoint meter, not a fact-checker: it grades delivery against a preset rubric and says nothing about truth.
• The 99 percent held-out accuracy badge and five-cent figure are author-reported from the author's own eval split.
• Requires local video tooling through the install script; no hosted demo.

04 · Attribution

Public sources

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

• ChetasLua ↗Community · github · public · checked 2026-09-23