A spot to park the data toys I’m building for the band I’ve been into for decades. A sortable songbook, some setlist math, and predictions I probably shouldn’t put in writing.
Beta build — Phish. Not linked from the live site and not indexed. Built
2026-08-12 14:34 UTC from the working corpus. The Prediction page is the new work; the others
are here to show the pipeline still runs end to end.
What this build is really showing. The page shell is famoe.ly's, and famoe.ly is a moe. site — the wordmark, the fan-bar link, the footer and the Songbook's own copy all name moe. in hand-written text, so every one of them is wrong here. That is the generalization gap, not a rendering bug. This build reaches parity with the moe. one: all five pages are present and every number on each is measured from this band's own store. The Tape Measure is new — it exists because phish.in now feeds the recordings mirror. archive.org carries no Phish, so there were no durations and no way to get any; phish.in has a per-track duration on 94.6% of its shows, and those go through the same recognition every tape goes through, because the durations arrive from phish.in while the slots belong to phish.net. It reads 4,178 performances over 229 nights, and it says plainly that there is one recording a night and nothing to cross-check against — which is the honest difference from moe., where 10,190 performances rest on two or more independent tapes. The Scorecard is also new and is built here rather than copied: the live one is rendered by the old pipeline out of a different database, so a Phish build used to ship a page of moe. numbers.