DarkPool. Prediction Markets on DeepBook V3

The @darkpool/resolver package runs two jobs in one Express process:

OO settlement endpoint. HTTP service for atomic Pyth + redeem PTBs (used by the frontend when a market is ready to settle).
Auto-settle keeper. Polymarket-style daemon that scans predict::PositionMinted events, groups by (oracle, manager, strike, isUp), and submits per-position predict::redeem_permissionless PTBs once an oracle expires.

Port: :8082. Configurable via RESOLVER_PORT.

Startup log

[resolver] RESOLVER_KEY loaded. Signing as 0x…
DarkPool resolver on :8082

If you see RESOLVER_KEY is not set, the keeper runs in unsigned mode and never settles. Fix by ensuring .env lives at repo root and the value is a single-line suiprivkey1… string (no quotes).

Atomic settlement PTB

Oracle to auto-settle to permissionless redeem flow. Step 1: Pull Hermes VAA via the Pyth Sui SDK. Step 2: pyth::update_price_feeds consumes the hot-potato. Step 3: OracleSVI auto-settles inside Predict. Step 4: market_key::new or range_key::new per redeemer. Step 5: predict::redeem_permissionless per redeemer, which credits the manager. Step 6: predict_facade::attribute_redeem emits an event. — One PTB pulls a Hermes VAA, pushes Pyth, settles OracleSVI, then redeems each opted-in position and emits attribution. Per-position scoping isolates aborts.

If ENOKI_PRIVATE_KEY is set, the PTB is sponsored, user-perceived gas is zero. Falls back to the resolver’s own keypair paying gas if Enoki throws.

Routes

GET  /health                          -> { network, sponsorMode, keeperSigner }
POST /markets/:id/resolve
  body: {
    oracleSviObjectId: string,
    optedInRedeemers: [{ predictManagerId, marketKind }],
    pythFeedIdHex?: string,
    pythPriceInfoObjectId?: string,
  }
  -> { digest } | { ptbBytes }    # depending on sponsor / RESOLVER_KEY config

Auto-settle keeper

File: packages/resolver/src/autoSettleTick.ts. Runs every 60s. Algorithm:

Page recent predict::PositionMinted events (up to 1000).
Aggregate by (oracle, manager, strike, isUp). Unique positions.
For each oracle: check expiry_ms, active, and settlement_price.
If expired and not yet settled: submit one predict::redeem_permissionless per PTB (one position per tx so a single abort can’t sink the batch).
Memoise tried tuples. Successes never re-fire; failures back off 30 min.
Always self-pays. Enoki allow-list rejects Predict’s Move targets.

Log excerpt:

[resolver] tick: 12 expired tuples, 0 already-tried, 12 to try
[resolver] redeem succeeded oracle=0x… strike=$65000 isUp=true digest=0x…

Files

resolver/
├── package.json                           # Express + Pyth Sui SDK + Enoki
├── tsconfig.json
└── src/
    ├── index.ts                           # Express server on :8082
    ├── buildAtomicSettlementPTB.ts        # Pyth pull + OracleSVI settle + redeem
    ├── enokiSponsor.ts                    # Sponsored-tx wrapper with self-pay fallback
    └── autoSettleTick.ts                  # Keeper loop

Gotcha: don’t run two resolvers

A tsx watch reload can leave the old setInterval alive while a new one starts. You’ll see duplicate-digest log lines and validator sequence-number conflicts. Fix: ⌃C every resolver shell, pkill -f "resolver/src/index", restart one.

Getting Started

Core Concepts

Move Modules

Backend

Frontend

AI Agents

Guides

Scripts Reference

Deployment

Reference

Resolver

Startup log

Atomic settlement PTB

Routes

Auto-settle keeper

Files

Gotcha: don’t run two resolvers

​Startup log

​Atomic settlement PTB

​Routes

​Auto-settle keeper

​Files

​Gotcha: don’t run two resolvers

Startup log

Atomic settlement PTB

Routes

Auto-settle keeper

Files

Gotcha: don’t run two resolvers