# Seeing Without Dependencies: A Measurement of a Hand-Written JPEG and DEFLATE Image Path, Its Sampling Bias, and Its Abstention Contract Daozhendong Yang, School of Intelligent Science and Engineering, Yunnan Technology and Business University, Kunming, China — Paper about **styx-vision**, a two-crate Rust workspace that gives a local-first assistant a way to look at images without a model and without a decoder dependency: a hand-written baseline {JPEG} decoder, a hand-written {DEFLATE} inflater for {PNG}, a {BMP} reader, a strided picture-statistics layer, a Chinese sentence renderer with an explicit abstention contract, and three optional remote backends behind one `Composer`. ## arXiv classification - **Primary:** `cs.SE` — Software Engineering - **Cross-lists:** `cs.CV` (Computer Vision and Pattern Recognition), `cs.PL` The same classification is stated in the header comment of `main.tex`. ## Abstract (plain-text rendering) A local-first assistant that cannot see images has two unattractive options: ship a vision model, or ship a dependency. styx-vision rejects both. It hand-writes a baseline JPEG decoder, a DEFLATE inflater for PNG, and a BMP reader in pure safe Rust, computes a fixed set of picture statistics from the decoded pixels, and renders them into a structured Chinese sentence for a text-only language model. The claimed cost of the design is two runtime crates; the claimed benefit is that the layer can never be absent, so the assistant can always be told something — including, when the pixels cannot be decoded, that nothing can be told. The decoder is real. All 10 reference JPEGs land inside the project's published tolerance (worst mean per-channel difference 0.0938 levels, worst single-pixel difference 3, no pixel differing by more than 8), and all 52/52 synthesized PNGs and 5/5 BMPs round-trip bit-exactly, including streams built from each of the three DEFLATE block types and each of the five PNG scanline filters. The abstention contract holds: a progressive JPEG is refused, yet its 64×48 dimensions are still reported, and in all 10 remote deployment states the description is non-empty. The gaps are in the parts nobody unit-tests. The ≤120,000 sampling stride changes mean saturation by 1.7e-4 at 3.84 megapixels while visiting 3.1% of the pixels, but inflates the colourfulness estimate by a factor of 32. All five injected one-line decoder defects are caught — though the upsampling defect slips past the criterion's mean bound and is caught only by its outlier condition, and the DEFLATE defect is invisible to the pixel-reference suite by construction. On the remote path, every image costs two detector requests because one backend is invoked twice, an 8 s per-request budget becomes 33.6 seconds against a slow sidecar, a 16-megabyte model reply is passed into the prompt verbatim as 16,777,385 characters, and the release itself does not pass its own cargo fmt or clippy gates — while 127 tests do pass. ## Abstract for the arXiv form (1,920-character limit) The abstract printed in the paper is longer than the form accepts; paste this version into the arXiv metadata field instead. It keeps every measured number and drops only prose. > A local-first assistant that cannot see images has two unattractive options: > ship a vision model, or ship a dependency. styx-vision rejects both. It > hand-writes a baseline JPEG decoder, a DEFLATE inflater for PNG, and a BMP > reader in safe Rust, computes fixed picture statistics from the decoded > pixels and renders them into a structured Chinese sentence for a text-only > model. The claimed cost is two runtime crates; the benefit is that the layer > can never be absent, so the assistant is always told something — including > that nothing can be told. > > The decoder is real. All 10 reference JPEGs land inside the project's > published tolerance (worst mean per-channel difference 0.0938 levels, worst > pixel difference 3, none above 8), and all 52/52 synthesized PNGs and 5/5 > BMPs round-trip bit-exactly, including streams from all three DEFLATE block > types and all five PNG scanline filters. The abstention contract holds: a > progressive JPEG is refused, yet its 64×48 dimensions are still reported, > and in all 10 remote deployment states the description is non-empty. The > gaps are in the parts nobody unit-tests. The ≤120,000 sampling stride > changes mean saturation by 1.7e-4 at 3.84 megapixels while visiting 3.1% of > the pixels, but inflates the colourfulness estimate by a factor of 32. All > five injected one-line decoder defects are caught — though the upsampling > defect slips past the criterion's mean bound and is caught only by its > outlier condition, and the DEFLATE defect is invisible to the > pixel-reference suite by construction. On the remote path, every image costs > two detector requests because one backend is invoked twice, an 8 s > per-request budget becomes 33.6 seconds against a slow sidecar, a > 16-megabyte model reply enters the prompt verbatim as 16,777,385 characters, > and the release itself does not pass its own cargo fmt or clippy gates — > while 127 tests do pass. ## The probe, and why it is disclosed The crate's public API is enough to measure but not to automate: pixel access and machine-readable output are needed. `scripts/probe/harness_probe.rs` is an `examples/` binary **we wrote** that calls only public functions (`decode`, `sniff`, `dimensions`, `analyze`, `analyze_bitmap`, `Composer`) and prints one JSON object per invocation, in five modes (facts, decode, dump-to-PPM, roundtrip, remote). `scripts/build_probe.py` copies the repository (excluding `target/` and `.git/`) to `scripts/build/