Overview
jevmeter (github.com/ChetasLua/jevmeter) is chetaslua's Python CLI. Whisper timestamps sentences, then parallel Jev Noul probes (per preset questions on evasion, spin, hot takes, or hype) score each line. Pillow draws 1920x1080 frames into ffmpeg for a 16:9 jevmeter.mp4 beside your source. Presets cover debates, earnings calls, podcasts, and launch hype. README badges quote about five cents for a full debate render on the battle demo thread and held-out preset accuracy claims in eval/RESULTS.md.
Problem: Long talking-head videos hide evasive lines in volume; viewers need sentence-level BS signals, not a vibe check after forty minutes.
Built for: Creators, researchers, and terminal-friendly editors who want open-source overlays that score every sentence with Jev and export a postable clip.
First indexed on Jev Directory: 2026-09-17
Creator and team
- Name
- chetaslua
- Handle
- @chetaslua
“Every sentence scored. Every dodge flagged. Rendered as a 16:9 edit you can post.”
How Jev is used
- Role in the product flow
- Per-sentence BS and spin scoring with parallel Noul probes after Whisper segmentation
- Primitives
- NoulScore
- State in
- Sentence text, speaker context, preset rubric questions, and prior lines per README pipeline (Whisper word alignment optional with transcript file).
- Decision out
- Calibrated per-sentence scores driving on-screen meters and highlight edits in the rendered video.
- Transcribe with Whisper timestamps (optional transcript alignment)
- Fire parallel POST /v1/systemone Noul questions per sentence and preset
- Aggregate scores for highlight or full-length edit modes
- Render 1920x1080 frames via Pillow into ffmpeg output
jevmeter is a batch media pipeline, not a chat wrapper. Whisper handles audio; Jev handles judgment. Each sentence triggers structured Noul questions drawn from the preset (evasive, dodged the question, hype, and similar flags in the wizard copy). README documents parallel Jev threads and render workers so long videos stay practical. The open-source BS meter clip on X shows the overlay in motion; this directory embeds that remote video in the showcase (referrerPolicy no-referrer so twimg playback works in Chromium). API keys live in user settings per install.sh, not in the repo. Mention the optional battle post only as cost context; the featured embed follows the open-source launch post.
Sourced performance claims
- README badge cites about five cents for a full debate render (battle demo post).Source: github.com/ChetasLua/jevmeter README and x.com/chetaslua/status/2100473581251748216
- Public GitHub repo ChetasLua/jevmeter had about eighty one stars when this listing was drafted.Source: GitHub star count September 2026
Features and stack
Features
- Terminal wizard for presets, speakers, and highlight vs full render
- Whisper transcription plus parallel Jev scoring
- 16:9 ffmpeg output ready to post
- Featured showcase clip with remote X video
Stack
- Python
- Whisper
- ffmpeg
- Pillow
- TypeSafe System One
Pricing: Open source; TypeSafe spend scales with sentence count and --threads. README cites micro-dollar debate totals for the sample battle.
Demo
Showcase clip with the same lightbox player as the homepage. Click to play.
jevmeter live BS meter on any video
Whisper slices the rant into sentences; parallel Noul probes light up every dodge. Sixteen by nine ffmpeg guilt you can post before the talking head finishes pivoting.
Links
FAQ
- Which primitive scores each sentence?
- README pipeline uses Noul questions per preset flag via systemone, with parallel requests per sentence.
- Do I need to commit videos to GitHub?
- No. Run the CLI locally; this directory only hotlinks the X/Twitter mp4 for the showcase player.
- What does the five cent badge mean?
- It comes from the jevmeter README battle demo badge referencing the full debate render. Your cost scales with length, preset, and thread settings.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- System One model
The model family behind Jev: parallel typed questions, one forward pass, probabilities you can threshold.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.