Community · E2 · artifact verified
Classify SVGs by weighting typed responses
A research task treating Jev's answer distributions as classifier features: many small typed questions about an SVG, responses weighted and combined until the signal classifies the image - a study of whether decision calls can stand in for maths on pixels.
01 · Role in the system
What Jev does here
The premise: regular image classifiers do maths on an image and output a value; asking Jev a question is a comparable primitive, so enough well-chosen questions about an SVG should combine into a classifier. The pipeline decomposes each SVG into per-feature typed questions, collects the probability distributions, weights the responses, and reads the combined signal as the class - a study of weighting schemes rather than a single prompt. A live demo renders the classification, the Python client is small, and the README links the sibling JevGraph project that grew from the same obsession: decision distributions as a computational substrate.
02 · Control boundary
Where Jev sits
Jev as a feature extractor for image classification: one SVG becomes many typed questions, response distributions are weighted and combined in code, and the aggregate - not any single answer - is the classification.
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
- Research scope: SVGs only, weighting schemes compared informally rather than benchmarked.
- Many calls per image; cost scales with question count.
- 0 stars; single-author study with a live demo at jev-svg.lexic.cloud.
04 · Attribution
Public sources
This is a Community record: the project was published by a third-party community author.
- Fox-Islam ↗Community · github · public · checked 2026-10-03