Form-native visual loopback — seeing learned the way hearing was

NameDescription

Extend the loopback-learning pattern (3rd-party oracle bootstrap -> Form-native candidate -> A/B via co-learning/champion-challenger/jit-oracle-learning parity -> promote, colearning-retire the oracle once native wins) from audio/speech to VISION: geometry, depth, texture, shape, light/shadow, recognition, generation, driven by a video-frame loopback. This completes the shape the body already DECLARED (loopback-learning lbl-video-shape-sample, channel-loopback video:features codecs + video-gen/classify loops, oracle-catalog eye/vision_classify.swift teacher, image-grammar.form + video-as-recipe.form). First native cell shipped: geometry.fk (band 1111111 four-way Go/Rust/TS/fkwu) - vectors, ray-sphere depth, Lambert light/shadow, render-pixel (shape+depth+shade in one computed pixel) over trig.fk fsqrt; the generator side a 3rd-party renderer is A/B'd against. Staged path by tractability: geometry (shipped) -> texture (2D spectrum.fk) -> recognition (nearest-shape.fk/recognition-router.fk, modality-agnostic) -> generation (host-carrier model weights, heaviest/last). Honest split: geometry/texture/edge/recognition/fusion are four-way recipes; video decode (ffmpeg), object/face detection, diffusion inference are host carriers proven by receipt. Teaching: docs/coherence-substrate/visual-loopback.form.

Potential value (USD)Estimated cost (USD)Confidence (0–1)