Last week this happened:
For those curious, the methodology is detailed further in this post. The short procedure:
- Define rare: With some wrinkles, based on my preference to see a word with a clean edge, as in decaf99rather thandecafea(more on this below)., repeated characters, repeated sequences of characters, palindromes, and ascending and descending sequences.
- Pattern match: combine possible paths to the pattern(s), divide over 16⁷ possible hashes. That yields our drop rate, assignable to a I drew the tier thresholds mostly by vibe, but I wanted to see an order of magnitude between tiers..
- Seek fortune.
Below you'll find a detailed (and configurable!) breakdown of how the scorer matches patterns. I can't vouch for every permutation of the configuration, but I can vouch for the one I like!
Everything below is computed in the browser:
- Scoring a hash: each run of 3 to 7 characters is checked against every pattern, and patterns that don't overlap stack. The probability is the rarest stack's matching hashes ÷ 16⁷.
- Saving new rules: a Web Worker recounts all 16⁷ hashes, marking each stack's matches in a 2²⁸-bit table. That gives the tier sizes and the map's data.
- Moving a tier threshold: no recount; the counted stacks are regrouped into the new tiers.
- The map: saved rules also compile to a GLSL shader, which calculates the scores of the hashes in view on the GPU.