automerge-lens

Explains why an Automerge CRDT merge resolved the way it did, and empirically fuzzes application order to catch convergence violations — the debugging tooling local-first frameworks are missing.

The gap this fills

Local-first / CRDT tooling matured fast through 2026 — mature libraries (Automerge, Yjs, Loro), production sync engines (PowerSync, ElectricSQL, Zero). But the debugging side lagged behind on purpose: CRDTs guarantee your data converges, not that you can see why it converged to what it did. When two replicas edit the same field concurrently, Automerge deterministically resolves the conflict — but today the only way to know a conflict even happened is to manually call getConflicts() at the exact path where you suspect one, and there's no tooling that walks a whole document for you, and nothing that verifies the underlying convergence guarantee actually held for a given set of changes.

What it does

Three independent, composable pieces of Automerge tooling:

getCausalHistory(doc)

Every change in the document's full history, topologically sorted by its deps so causality is always visible — not just whatever order the underlying store happens to return.

explainConflicts(doc)

Recursively walks the entire document (not just one path you already suspect), and for every key with a live conflict reports every candidate value, which actor wrote it, and which one automerge currently presents — so a conflict isn't just "detected," it's explained.

checkConvergence(changes, options)

The one thing a CRDT promises is that applying the same changes in any order produces the same final state. This applies a flat set of changes (gathered from any number of diverged replicas) in many random orders and asserts they all converge to identical heads. If they don't, that's a real bug — almost always non-deterministic or side-effecting code inside a change() callback, which Automerge itself has no way to catch for you. Runs are seeded, so a failure is reproducible.

historyToMermaid(history) / conflictsToMermaid(reports)

A JSON array of changes is not how a person actually wants to look at a merge history. These render getCausalHistory and explainConflicts output as Mermaid diagrams — paste the output into a ```mermaid fenced block anywhere that renders Markdown (GitHub does this natively, no library needed) and get an actual picture instead of a wall of hashes.

For the classic concurrent-edit scenario from Automerge's own docs (two actors both rename the same pet), the conflict graph looks like this:

graph TD
  classDef won fill:#1a7f37,stroke:#116329,color:#fff
  classDef lost fill:#cf222e,stroke:#82071e,color:#fff
  path0["pets.0.name"]
  path0_e0[""Babe"\n(aaaaaaaa)"]
  path0 --> path0_e0
  class path0_e0 lost
  path0_e1[""Beethoven"\n(bbbbbbbb)"]
  path0 --> path0_e1
  class path0_e1 won

Loading

Install

npm install automerge-lens @automerge/automerge

Usage

import * as Automerge from "@automerge/automerge";
import { getCausalHistory, explainConflicts, checkConvergence } from "automerge-lens";

const merged = Automerge.merge(doc1, doc2);

for (const report of explainConflicts(merged)) {
  console.log(`Conflict at ${report.path.join(".")}`);
  for (const entry of report.entries) {
    console.log(`  ${entry.isCurrentValue ? "WON " : "lost"} "${entry.value}" (actor ${entry.actor})`);
  }
}

const result = checkConvergence([...Automerge.getAllChanges(doc1), ...Automerge.getAllChanges(doc2)]);
if (!result.converged) throw new Error("convergence violated -- see result.divergence");

Run the annotated demo (the exact conflict scenario from Automerge's own docs, explained):

npx tsx examples/demo.ts

Non-goals (v1)

• Not a replacement for Automerge's own APIs — it's a thin, well-tested layer that makes existing APIs (getConflicts, getAllChanges, decodeChange) actually usable for whole-document debugging, not a reimplementation of them.
• Automerge only in v1. Yjs and Loro have different internal models (no getConflicts-equivalent API in the same shape); a v2 direction, not silently half-supported here.
• Static diagrams, not a live interactive UI. historyToMermaid/conflictsToMermaid produce Mermaid source you render (GitHub, most Markdown/docs tools do this for free); a live, clickable, in-browser visualizer is a natural next step, not v1 scope.

Development

npm install
npm run typecheck
npm test
npm run build

License

MIT

Part of a series

This is one of 8 independent open-source infrastructure tools, each built against a real, researched gap:

• embedguard -- catches silent embedding-model swaps under a vector index
• agora -- propose/vote/veto/quorum for multi-agent systems
• mcp-versionbridge -- bridges old and new MCP protocol handshakes
• yjs-lens -- verifies Yjs convergence and summarizes causal history
• gatewayproof -- conformance-tests OpenAI-compatible LLM gateways
• wit-breaking -- detects breaking changes in WIT interfaces
• pglite-diff -- diffs migration query results on isolated Postgres

---

Built by ArkNet Digital.