A live, read-only telemetry window into murmur's real-money x402 settlement rail on Arc mainnet. Every figure below is pulled client-side from the same free, keyless, CORS-open endpoints the swarm itself runs on — no synthetic load, no testnet, no LLM anywhere in the loop. A population of autonomous fruit-fly nervous systems trades around the clock and settles real USDC over x402 / EIP-3009, folded into per-pair nets to amortise gas.
reads api.muros.live/economy · /history · /predictions · /proofs — view source, it's all client-side.
Terminal outcomes of real on-chain net settlements. successRate = settleOk ÷ settleAttempts over
genuine broadcast attempts (shadow dry-runs excluded), plus the measured submit→finality latency of each
mined settle (timed at the facilitator facade, so it covers whichever rail is live). All additive lifetime
counters, persisted across deploys — read live from /economy.
Micropayments are not broadcast one-per-trade. Per-pair nets accumulate bilaterally and only the difference
moves on-chain, so gas is amortised over more value per transaction. The fold figures are derived live from
/economy totals — count (gross trades) against settleAttempts (broadcasts) —
and the net-pending gauges below read the live in-flight accumulator (pairs folded, not yet broadcast).
The swarm also bets real USDC each cron on the next tick's Arc temperature, resolved parimutuel by the chain
(zero-sum, no house). A FLAT resolution — inside the flat band — refunds every stake as
net-zero flows, so nothing broadcasts on the settlement rail: a FLAT round is a
resolution outcome, not a refund transaction. Rate below is read live from /predictions.
From the independent per-cron D1 archive (/history) — one row per cron, separate from the live
Durable-Object state, so these curves are cross-checkable against the on-chain receipts.
Each on-chain net settlement sets its EIP-3009 nonce to the sha256 of a receipt bundling the
frozen buyer/seller neural read-outs that caused it, hash-chained via prevChain and committed on Arc
(every proof carries its commitTx). A receipt invented after the fact could never hash to a mined nonce —
so the trades behind every figure above are provably decided by connectomes, not a human or an LLM.
chain head · loading live anchor…
GET /proofs → the chain (last 64 served) · chainHead · GET /proofs/verify?tx=0x… → { selfConsistent, match, registry } · or verify in-browser at muros.live → neural proofs.
All six headline metrics are now instrumented. What follows are the honest edges of each figure — what it does and does not claim — so nothing above is over-read.
Arc finalises deterministically (sub-second blocks, no reorg), so settleMs* measures the RPC broadcast→receipt round-trip, not a reorg-confirmation wait. Values are naturally small; the signal is the tail (max) and any drift over time.
netPending/netPendingTrades are sampled at snapshot time and drain at the next cron flush; they are not lifetime totals. Circle /status reconciliation stays defined-but-unwired (the internal rail settles on-chain directly).
The FLAT/refund rate is computed over the recent resolutions /predictions serves (a lifetime FLAT tally is not exposed). And “txs avoided” is the transaction-count reduction that amortises gas — murmur does not yet publish per-tx gas units.