In systems that maintain historical state (databases, event logs, blockchains), I ran into a recurring issue:
It’s easy to make data append-only, but much harder to guarantee that:
past state is never silently rewritten
contradictions don’t coexist undetected
interpretation of old data doesn’t drift over time
For example:
append-only logs still allow reinterpretation bugs
schema changes can alter meaning of past data
distributed systems can expose conflicting views without detection
I tried modeling a system with stricter invariants:
all updates are forward-referenced (no retroactive mutation)
state ordering is deterministic
contradictions must be explicitly represented or rejected
interpretation is versioned alongside state
The goal wasn’t performance — just structural integrity of history.
Question for people working with real systems:
Where do these guarantees usually break in practice?
Are append-only/event-sourced systems enough, or do they still leave gaps around interpretation and consistency?