@zkpoc/broker

Shard model, tier-aware sizing, commit-then-challenge verification, redundancy consensus, stake-derived audit, and a credit ledger — the multi-client verification layer for consented browser compute. Splits into two structurally separate pipelines, deliberately not sharing a verification or reward path:

• Barter — crowdsourced compute, a worker paid for confirmed work.
• Challenge — anti-bot proof-of-work, useful work substituted into the slot hashcash-style widgets (Cloudflare Turnstile, Anubis, ALTCHA...) occupy today. See @zkpoc/challenge for the client-side solver.

Part of ZK-PoC, a consent-governed, verifiable browser-compute project. Zero runtime dependencies.

import {
  LAPTOP_IGPU, chooseShardSize,
  issueChallenge, resolveChallenge, ChallengeOutcome,
} from '@zkpoc/broker';

// Server side: issue a challenge sized for ~2s of GPU compute on a
// measured device tier.
const { shard } = issueChallenge(LAPTOP_IGPU, { targetWallSeconds: 2 });
// ...send shard.toJSON() to the client, which solves it (@zkpoc/challenge)...

// Then verify the response:
const { outcome } = await resolveChallenge(shard, submittedResult, {});
if (outcome === ChallengeOutcome.ADMIT) proceed();

Why commit-then-challenge

An earlier design derived the verification challenge from public shard data alone, which let a worker compute just the checked points and skip the real O(n³) computation entirely. The fix: a worker Merkle-commits every output row first, and the challenge is derived from that root via Fiat-Shamir — so a valid root costs the real computation, full stop. See ADR-0011.

What's in here

| Module | What it does | | --- | --- | | tiers | Device-tier shard sizing from measured constants; refuses to guess for an unmeasured tier | | shard, merkle | Deterministic-but-fresh shard inputs, row-level Merkle commitment, Fiat-Shamir challenge derivation | | queue | Lease-based multi-worker assignment, replica independence | | consensus | Per-submission gate plus majority/dispute tally | | audit, ledger | Stake-derived audit rate (an economic inspection game, not fixed sampling), stake/reward/slash | | challenge | The anti-bot single-submission protocol (issue/resolve), deliberately not built on the barter pipeline above |

Status

Experimental — API surface may still change. 170 tests.

License

MIT OR Apache-2.0.