Things that have to work

A community-kept concert archive: every show, every segue, every tape. 5,441 shows across seven bands, from 1965 to last weekend — and it says out loud when the record is missing.

Setlist sites flatten a segue into a comma. This one treats a run of joined songs as a single musical object and renders it that way, notation intact, because the segue is the point. It also refuses to let absence look like emptiness: a show with no surviving tape, a show nobody documented, and a show whose sources contradict each other are three different facts, and all three are recorded rather than rendered as a blank page. Corrections are proposed by anyone and voted through by committee, which is how the record gets better instead of just bigger.

• Next.js
• TypeScript
• Postgres
• 2026

A fleet-monitoring platform watching 327 production sites — statistical anomaly detection against each site’s own history, an archive of what the host discards, and one rule everywhere: an unchecked site must never look healthy.

Most monitoring answers "is it up?" This answers "is it behaving like itself?" — each site is measured against its own history rather than a shared threshold, so a site that has always been slow does not cry wolf and a site that just got slow cannot hide. The hard rule underneath is that silence is never treated as health: a check that did not run reports as unknown, loudly, because a dashboard that goes quiet when the collector dies is worse than no dashboard.

• Drupal 11
• Next.js
• Pantheon
• 2026

A monolithic Drupal 7 application with millions of nodes, rebuilt as Drupal 10 while the site kept publishing — custom migration paths, counts reconciled every run, zero records lost.

The site could not stop publishing, so the migration ran incrementally against a moving target for months. Every run reconciled its own counts and refused to advance on a mismatch, which is the only reason "zero records lost" is a statement rather than a hope. Cutover was measured in minutes.

• Drupal 7→10
• Migrate API
• Behat
• 2023

A replacement for Drupal core’s taxonomy UI, built for vocabularies from a dozen terms to 100,000 — drag-and-drop, undo history, bulk merge and clone. Running on 90+ production sites by drupal.org’s own count.

Core’s taxonomy screen was written for a short list and falls over on a long one. This handles vocabularies three orders of magnitude larger with drag-and-drop reordering, an undo history, and bulk merge and clone — and the install count is drupal.org’s own, not ours.

• Drupal 10/11
• Vanilla JS
• SortableJS
• 2025–