Engineering

How we rebuilt sync for crews that work without signal

Hana Kowalski · Oct 28, 2026 · 7 min read

Most of our customers spend their day in basements, wind farms and hospital wards. Signal comes and goes, and the old sync engine treated every drop as an error. Field crews were losing photos, and supervisors were re-entering checklists by hand at the end of a shift.

We started from one rule: the device is the source of truth until the server says otherwise. Every edit is written locally first, stamped with a logical clock and replayed in order once a connection returns.

Conflicts are a product problem

Two inspectors editing the same checklist is not rare, it is Tuesday. Instead of silently picking a winner, the new engine merges field by field and shows a small diff when two people changed the same answer.

The result after six weeks in beta: sync failures down 92 percent, median upload time from 14 seconds to 1.8, and zero lost photos across 410,000 inspections.

What comes next

Background uploads on Android, selective sync for very large sites, and an audit trail that shows exactly when each change reached the server.