CASE STUDY · PUBLIC
deckmill
Browserless slide factory: a 46-entry registry rendered by an embedded engine, zero Chromium anywhere in the pipeline.
Status ruling active development
Evidence for Essay 5, Ideal tooling infrastructure: choosing and building tools. The spec names what that argument has to hit and the figures it will cite.
Problem
Slide tooling assumes a browser: slow, heavy, flaky in CI. deckmill renders from an embedded engine with no Chromium anywhere in the pipeline.
Build
- A 46-entry slide registry: 38 live slide types and 8 retained as removed, each removed entry naming its replacement, so the registry keeps its own supersession record.
- An embedded Blitz engine renders previews directly, so the previews in the repository are engine output rather than screenshots taken by a browser.
Outcome
- 192 tests passing with zero failures at commit 18036f4, measured on the working tree. The README badge said 185 at measurement time: stale by seven, and not the register.
- Registry caveat tracked rather than hidden: superseded entries marked removed remain in the slide registry. The register previously printed a slide-type figure the repository disproves; R17 withdraws it in full and is kept public on purpose.
Provenance
| Gate, executed | 192 passed, 0 failed | cargo test at 18036f4, 2026-10-20 |
|---|---|---|
| Slide types | 46 entries: 38 live, 8 removed | slide_registry.rs key count at fc38859, 2026-10-26 |
| LOC | 27,383 | raw count over src/ and crates/ |
| Test attributes, static | 203 (upper bound) | static attribute count, revised from 196 in R14 |
Reproduce
Clone the repo and run `cargo test` at the pinned commit. The engine rendered previews are committed inside the repository, so the browserless claim can be checked by looking at its output. The registry file is openable; the key count is the figure.
https://github.com/ishan-parihar/deckmill
Published documents
From the documentation
The design system is written before the renderer: OKLCH color, a type scale and a layout grammar, with the 46-entry slide registry as the vocabulary that grammar composes.
DESIGN-GUIDE.md at fc38859 (2026-10-21)
The styling architecture explains why the engine needs no browser: output is composed from the design system, not screenshotted from one.
styling-architecture-v2.md at fc38859 (2026-10-21)
Version 2 of the styling architecture is published under its own filename, so the supersession is visible rather than silently replaced.
styling-architecture-v2.md at fc38859 (2026-10-21)