Tolerance

Right Answers from Wrong Parts.

Six live instruments that get the right answer out of noisy, broken, drifting, or lying parts, and show how each one fails:

#	Exhibit	What it shows
01	The Vote	Triple modular redundancy, von Neumann's threshold, and common-mode failure (Ariane 5)
02	Scratch the Code	A real, scannable QR code (version 3-H) you can deface, with the Reed–Solomon decoder repairing it live
03	Where Am I	GPS pseudoranging: solving for position and clock, dilution of precision, and why altitude is worse
04	Two Wrong Answers	A GPS + accelerometer Kalman filter that also learns the accelerometer's bias
05	Light Through a Finger	Pulse oximetry's ratio of ratios, motion artifact, perfusion, and camera heart rate with your own finger
06	Drop It on Purpose	TCP CUBIC through a bloated router buffer, compared with CoDel and fq_codel

jkastl.github.io/tolerance

The Lab

Four new concepts, each on its own page and built a piece at a time. Every page carries its own design and build plan, and CLAUDE.md explains how to advance one.

#	Page	Idea	Status
07	liar/	Twenty questions against an adversary allowed to lie (Ulam's liar game, Berlekamp's volume bound)	blueprint
08	traitors/	Pass-the-phone party game running the Byzantine Generals protocol	building (playable)
09	fingerprint/	Your phone's sensor calibration errors as an ID (SensorID, 2019), on-device only	blueprint
10	tearproof/	Split a secret across printable tags; any k of n rebuild it (Reed–Solomon / Shamir)	blueprint

No build step, no dependencies, and no backend: everything runs in the browser, and no data leaves the device. Each page loads an anonymous GoatCounter visit counter (no cookies) and works without it. Every exhibit and Lab page lives in its own folder. Exhibit pages share site.css and core.js, and Lab pages share lab.css. Everything is served by GitHub Pages from main. The QR encoder/decoder (GF(256), Berlekamp–Massey, Forney), Gauss–Newton GPS solver, Kalman filter, and TCP/CoDel simulation are all written from scratch in the page. The QR output matches the Python qrcode library bit for bit.

The camera demo uses getUserMedia, so it needs HTTPS (GitHub Pages is fine). Video frames are averaged on the device and never sent anywhere.

A sibling of A Field Guide to Emergence.

License

MIT