Skip to content
LOGITEXDevelopment LLC

August 5, 2026 · 7 min read

Decide the conflict rules before you write the offline app

Offline-first field applications do not fail in the sync code. They fail because nobody wrote down what happens when two people edit the same job, and the answer gets invented at three in the morning by whoever is on call.

Every offline-first project starts the same way. The office system is fine, the crew has no signal for half the day, and someone says: we will just sync when they get back. That sentence contains an entire distributed systems problem, and it is usually the last time anyone discusses it before the app is built.

The question nobody writes down

Two people touch the same job. A crew lead records four hours of labour and six units of material at 09:40 while offline. At 10:15 the office marks the same job complete and issues the invoice. At 14:00 the phone reconnects. What is true?

There is no general answer, and that is the point. There is only a per-field answer, and it belongs in the architecture phase, in writing, before the first screen exists.

Rules per field, not per record

Treating the whole job record as the unit of conflict is what produces the classic outcome: the field wins, the office loses its invoice, and someone rebuilds it from photographs. Split it.

  • Facts observed on site — hours, materials, photos, meter readings: the field wins. The office was not there.
  • Commercial state — price, invoice, purchase order, approval: the office wins. The crew cannot see the contract.
  • Scheduling — assignment, appointment window: last write wins, because both sides can legitimately change it and the newest intent is the real one.
  • Anything append-only — notes, photos, status events: never conflicts. Both survive, ordered by device clock, with the clock skew recorded.

Four categories cover most of a field service record. The exercise takes an afternoon with the operations lead, and it is the single highest-leverage afternoon in the project.

When the rules cannot decide

Some collisions are genuinely ambiguous, and the correct behaviour is not to guess. Park the record in a review queue with both versions side by side and a human decision. A queue that gets three items a week is a working system. Silent resolution that is wrong once a month is not — nobody finds out until the customer disputes an invoice.

Test it in a basement, not a simulator

Airplane mode on a real device, in a real building, with a real cellular handover on the way out. Simulators reconnect instantly and cleanly; phones reconnect on a train, with two seconds of connectivity, mid-upload, on a battery-saving OS that suspended the app twenty minutes ago. The interesting bugs live in that gap.

Sync code is easy to write and impossible to guess. The hard part was always the decision you skipped.

This is the reasoning behind the field service CRM blueprint published on this site. It is a reference design rather than a delivered project — the rules above are what we would fix in phase two, before anyone opens an editor.

Tags

  • Architecture
  • Mobile
  • Field service

Have a system like this in mind?

Send us the shape of the problem. You get a written scope and a fixed number before anyone writes code.