"Internet-scale capital markets" is a phrase from another chain's roadmap that this site once published as its own. Here instead is what EmpoorioChain's throughput and latency actually are, with the document each number comes from.
Latency (measured, testnet, May 2026)
| p50 | p95 | p99 | |
|---|---|---|---|
| Inclusion | 3.51 s | 4.20 s | 4.26 s |
| Finality (GRANDPA) | 16.96 s | 17.66 s | 17.72 s |
Block time 6 s. Finality about three blocks behind the head, tightly clustered because GRANDPA votes on chain prefixes. Mainnet gate wants finality p50 under 12 s; not yet met.
Throughput (measured)
700 transactions in a mixed workload — transfers, swaps, lending, staking, bridge simulation, read pressure — submitted in 1.62 s (431 tx/s submit rate), 700 included, 700 finalized, 0 % failed. This is a burst test of correctness and latency. It does not demonstrate sustained throughput at the 1,000 TPS transfer target or 200 TPS DEX target; the report says so.
Throughput (projected)
The parallel-execution scheduler projects ≥ 5,000 TPS at 0 % conflicts, ≥ 3,200 at 20 %, ~1,800 at 50 %, ~500 at 100 % (sequential-equivalent). Projections from a scheduler benchmark on a laptop, labelled as such. The four-validator run that would measure them is prepared and unexecuted.
Bandwidth (the network layer)
Substrate's libp2p stack for blocks; a DAG-aware gossip for transactions and pre-partitioned batches. Health is exported: emp_txpool_propagation_delay_ms_p95, DAG width and depth, duplicate count. No public figures yet for propagation delay under load; the metric exists so there can be.
Fees (measured)
Transfer 0.000025 DMS; swap 0.00018; 1 MB blob 0.000095; ZK-verified inference 0.0035. Weights are hand-written, so these are approximate until benchmarked on the reference machine.
Where the ceiling is
Two validators in one rack. Every latency number above is what two co-located machines produce; a geographically spread set of ten will be slower per round and the design must absorb that. The honest statement of EmpoorioChain's performance is: fast enough for payments today on a network too small to prove it at scale.
What increases it
Parallel execution (native, then EVM with revm), the DAG mempool feeding pre-partitioned batches, tuned thread pools per validator — all in the runtime or the roadmap, all waiting for the multi-node benchmark to become numbers.
Based on PERFORMANCE_REPORT.md (2026-05-30), PARALLEL_EXECUTION_REPORT.md, MEMPOOL_DAG_ORDERING.md, FEE_MODEL_AND_LOW_COST_STRATEGY.md and MAINNET_GATE.md.