Conflict Modeling¶
CRDT design works best when concurrent writes represent independent facts. HyperToken's containers use stable identities and rank sidecars so a draw, insertion, placement, or reorder can be merged without replacing a whole list.
Model state so that:
- independent entities have independent IDs;
- a placement or membership entry has identity separate from its token value;
- derived views can be recomputed from replicated facts;
- game actions validate domain invariants before committing;
- conflicts that need authority are resolved by a rule, policy, or trusted server rather than assumed away.
Automerge's merge result is deterministic for a given causal history, but it is not a domain-specific winner selection policy. Last-writer expectations, turn exclusivity, resource conservation, hidden information, and fair dealing require explicit modeling. Convergence does not equal correctness or fair play.