bitrep (JavaScript / WebAssembly)

Exact, order-invariant, bit-identical floating-point reductions and CRDTs — a WebAssembly binding over the Rust bitrep engine, so every result is the same across any machine, any merge order, any sharding. Runs in the browser and in Node.

npm install bitrep

import init, { SumF64, MomentsF64, SumMap } from "bitrep";
await init(); // load the wasm

// Exact sum — naive float gives 0; this gives 1, and it's order-invariant.
const s = new SumF64();
[1e16, 1.0, -1e16].forEach((x) => s.add(x));
console.log(s.value());            // 1

// Exactly-rounded statistics
const m = new MomentsF64();
[2, 4, 4, 4, 5, 5, 7, 9].forEach((x) => m.add(x));
console.log(m.mean(), m.variance(), m.count()); // 5 4 8n

// Receipt: same multiset -> same 32-byte hash regardless of order.
console.log(Buffer.from(s.state_hash()).toString("hex"));

// CRDT map that converges without a coordinator
const a = new SumMap(); a.add("k", 1.5); a.add("k", 2.5);
const b = new SumMap(); b.add("k", 2.5); b.add("k", 1.5);
a.merge(b); // idempotent, order-invariant

What's included

Exact accumulators (SumF64, SumF32, FastSumF64), exact dot product (DotF64, dot), convergent statistics (MomentsF64, Moments4F64 with skewness/kurtosis, CovF64 regression, WeightedMomentsF64, PnMomentsF64 retractable, CovMatrixF64 multiple regression — with the v0.5 exact tier: regression_exact() for correctly rounded coefficients and sub() for exact downdating/unlearning, plus SumF64.try_unmerge), HistogramF64 (exact counts + honest quantile bounds), ExtremaF64, signed receipts (state_hash), and the CRDT layer (SumMap, MomentsMap, ReplicatedSum, DeltasSum).

The convergence laws of the core are machine-checked in Lean 4. See the main repository for proofs, the Rust crate, and the paper.

License: MIT OR Apache-2.0.