98 days, 374 commits, 16 releases. Scrutinizer was a sprint through the frontier of biologically-inspired rendering, where theoretical correctness meets visual stability at 60fps.
This page documents the development arc. For the vision science and shader implementation, see Inside the Math.
November 2025 – February 2026
Scrutinizer was built in a 98-day sprint, from November 2025 through February 2026. Here's how it compares to its sibling project, Psychodeli+ (a year-long evolution of a 1996 screensaver into an AI-directed audio-reactive visualizer).
| Metric | Scrutinizer | Psychodeli+ |
|---|---|---|
| Commits | 374 | 1,703 |
| Calendar span | 98 days | 340 days |
| Active days | 37 (38%) | 175 (52%) |
| Lines written / deleted | +46,908 / -13,504 | +492,839 / -271,393 |
| Net codebase | 11,057 lines / 42 files | ~130k lines / 275 files |
| Shader lines | 2,052 (3 files) | ~4,500 (2 files) |
| Documentation | 46 markdown files | 128 markdown files |
| Golden captures | 95 PNGs / 8 versions | 78 images (essay) |
| Releases | 16 tags (v1.0–v1.6.0) | Continuous deploy |
| Reverts | 7 (3 in one day) | ~15 |
| Peak commits/day | 63 (Nov 23) | 45 |
| Longest streak | 14 days | 29 days |
| AI co-authored* | 3 (1%) | 88 (41% in Feb) |
Scrutinizer is a sprint where Psychodeli+ is a marathon. It shipped one release every 6 days and made 63 commits on launch day. Every perceptual experiment ran through the shader, which was touched in 9% of all commits.
*AI co-author counts reflect git workflow. The co-author tag records who runs
git commit, which does not show who wrote the code. In Scrutinizer, the developer
handled all git commits directly. In Psychodeli+, the AI agent started making commits.
2025-11 | +24,680 -7,555 (net +17k) ← THE BIG BANG
2025-12 | +17,690 -4,916 (net +13k) ← Perceptual accuracy push
2026-01 | +2,218 -374 (net +2k) ← Mobile + calibration
2026-02 | +2,320 -659 (net +2k) ← De-monolith refactorThe reverts are the most telling part of the git history. On December 13, 2025, three rollbacks happened within 30 minutes. Each one pulled the shader back to a less ambitious but more stable tier of perceptual fidelity.
Most of the engineering effort went into the gap between "theoretically correct" and "visually stable at 60fps."
The three reverts on December 13th were the project's inflection point. After them, the goal was perceptual plausibility, and biological accuracy became secondary. The human visual system is noisy, approximate, and good enough, and the shader can be too.