Distributed systems · 2026 · Solo — research, systems design, build, verification

Causal Canvas

Partition two replicas, write independently, then inspect why opposite delivery orders still converge.

The problem

Offline-capable collaboration fails in ways that are hard to explain from sync logs alone: arrival order can differ from causal order, concurrent edits have no natural wall-clock winner, and convergence claims need proof. Causal Canvas turns that distributed-systems reasoning into one bounded, interactive experiment without pretending to be a production CRDT.

Architecture

Key decisions

01

Separate causality from arrival order

Each operation captures the creator's version-vector frontier. Reconnect deliberately presents opposite delivery orders while materialization follows causal dependencies and a documented deterministic tie-break.

02

Keep the algorithm pure and the interface inspectable

The merge engine owns validation, immutable operations, vector comparison, and topological ordering. React only dispatches actions and renders the same evidence as replica leaves, operation nodes, and a selected proof.

03

Bound the model instead of overstating it

The release supports two replicas and append-only runs. It explicitly excludes deletes, arbitrary cursor edits, persistence, real networking, and production CRDT or Automerge parity.

04

Make publication evidence transactional

The private product repository and this portfolio site remain separate. Production was linked only after tests, build, browser behavior, English, visual differentiation, and hiring-signal gates passed; Vercel then matched the pushed Git SHA.

Metrics

7
automated tests passing
2
replica delivery orders proved
2
responsive viewports asserted
0
document network requests