Bidirectional sync: conflict resolution strategies that hold up in audit

Why last-write-wins keeps getting chosen anyway

It is the default because it requires no decisions. The cost arrives later: a record that oscillates between two values, a support team that cannot explain the current state, and an audit question nobody can answer. Once volumes rise, the corrections themselves generate conflicts.

Strategy 1 — field-level ownership

Declare authority per attribute rather than per record. This resolves the majority of real conflicts because most disagreements are structural, not concurrent: sales owns pipeline stage, finance owns invoice state, the product owns usage. Store the map as data so a new field is a configuration change.

Strategy 2 — version vectors for genuine concurrency

When both sides legitimately author the same field — collaborative notes, field-operations data captured offline — carry a version vector per record and merge deterministically. The important property is not cleverness but reproducibility: the same inputs must converge to the same output on every device and every replay.

Strategy 3 — reconciliation as a metric

• Re-compare both stores on a schedule and report drift per field.
• Alert on sustained non-zero drift, not on a single mismatch.
• Keep a replay path for one record, one segment, or the whole estate.

Make conflicts visible to humans

Every resolved conflict should be logged with both candidate values, the rule applied, and the outcome. When someone asks why a value changed, the answer should take seconds to produce. That log is also the fastest way to discover that an ownership rule was wrong.