Running the Prover

The prover turns Pulsar's block history into settle() transactions on Mina. It runs as five processes (a coordinator plus four workers) over Redis and MongoDB:

Prerequisites

• Node ≥ 20, pnpm
• MongoDB and Redis
• A Pulsar gRPC endpoint and a funded Mina fee-payer account
• ~8 GB memory per proving process, and ~2–3 GB disk for the compile cache (prover/cache/, safe to delete but slow to rebuild)

Setup, in this order

pnpm install                 # repo root
cd prover
cp .env.example .env         # fill in, see below
pnpm run seed                # ONCE, with the chain reachable
pnpm run deploy              # only for a fresh contract deployment
pm2 start ecosystem.config.cjs   # or: pnpm run start

• seed writes the genesis and anchor blocks the pipeline builds from. Run it once, before first start. It is idempotent.
• deploy compiles everything (several minutes) and anchors a new contract. It refuses to run unless its address matches the chain's configured contract, and the deploy keypair is single-use because the verification key is sealed at deployment. If you are joining an existing deployment, skip it and just set CONTRACT_ADDRESS.

Environment

Variable	Meaning
MONGO_URI, REDIS_HOST/REDIS_PORT	State and queues
PULSAR_GRPC_ENDPOINT	The chain (:443 gets TLS automatically)
CONTRACT_ADDRESS	The settlement contract
MINA_PRIVATE_KEY	Fee payer for settle transactions
MINA_NETWORK	devnet, mainnet, or lightnet
SETTLER_WINDOW	In-flight settle txs (default 5; keep well under Mina's ~10 per-account mempool cap)
EPOCH_START_HEIGHT	Frozen at seed time; changing it means pnpm run reset plus a redeploy

Health

pnpm run doctor    # one-shot verdict: what is blocking, or "throughput, not a wedge"
pnpm run gap       # contract height vs chain tip; -- --watch 30 to follow

Doctor checks heights, fee-payer balance, the pipeline's state distribution, the exact epoch the contract needs next, in-flight settle transactions, and the known failure classes, with copy-pasteable recovery commands where one exists.

Scaling

• Scale freely: pulsar-block-prover, pulsar-aggregator, pulsar-settlement-prover (pm2 scale …). Stale-claim sweeping is age-gated precisely so sibling instances don't steal each other's live work.
• Never scale pulsar-settler. Mina transactions are nonce-based and settle transactions chain by state precondition, so they must be signed and broadcast in strict height order by a single sequential worker. Two settlers would reuse nonces and evict each other's pending transactions. Settler throughput comes from pipelining (SETTLER_WINDOW), not parallelism.

Crash recovery

Automatic. Proofs are durable in MongoDB; a stalled settle transaction is detected by timeout and re-sent (not re-proven); interrupted work is swept back to the queue after its age gate. pm2 restarts complete the picture.