◌ causal atlas / 001
simulated network
a distributed systems observatory

a world
learning to agree.

Five places. Five versions of the truth.
Pull them apart. See what survives.

experiment 01 / concurrency

Tokyo went dark. London and Tokyo wrote different futures. Neither has seen the other.

T + 00:00.00seed 86
worlds apart5 independent replicas
drag to orbit
select a region below
causality travels at the speed of a message
replica agreement1/ 5
messages delivered00 lost to the network
wire traffic0 kbactual serialized state payloads
concurrent values1preserved, never silently overwritten
every effect leaves a trace

the event horizon.

0 events
trust the laws. test the implementation.

try to prove
it wrong.

Random writes, deletes, and merges. Hundreds of hostile histories. Every merge must obey three laws.

01 order does not matter02 duplicates do not matter03 grouping does not matter
a model you can inspect, not an animation you have to trust.
field notes / how this world works

the beautiful part
is the constraint.

This is a real replicated data structure inside a simulated network. Five in-memory replicas merge serialized state; no remote data centers are contacted.

what converges?

An observed-remove map containing a multi-value register. Concurrent writes survive together. A deletion removes only the causal dots it has observed. A later write that has seen both versions replaces both.

what makes repair possible?

Periodic anti-entropy sends immutable state snapshots. After links heal and delivery becomes reliable, replicas converge. Message order, duplicates, and intermediate loss do not change the final merge.

what this does not claim

No linearizability, consensus, authenticated peers, production durability, or bounded metadata. The explicit causal context grows with writes; real deployments need a safe membership and compaction design. Latencies are illustrative geographic distances plus seeded jitter, not measured internet performance. Rewind inspects history; it does not fork time.

why a state-based model?

It makes the full merge semantics inspectable. Wire traffic is real JSON payload size, not a delta optimization. The randomized checker tests convergence laws for generated reachable states; it is evidence, not a formal proof.

delta state replicated data types ↗