Benchmarks are usually published to impress. This one was published to be reproduced. The EmpoorioChain testnet performance report of 30 May 2026 records a mixed-workload run against a live node, includes the raw command output, and ends with a decision: do not promote to mainnet from this report alone. Here is what it contains.
Targets
| Target | Value |
|---|---|
| Native transfer throughput | 1,000 TPS |
| DEX swap throughput | 200 TPS |
| Maximum failed-extrinsic rate | 1 % |
| Finality stalls | 0 seconds |
| Minimum validators under test | 4 |
Workload
The generator mixes six traffic types to look like a real ecosystem rather than a transfer-only sprint:
- Native transfers (50 %) — DMS through
balances. - DEX swaps (20 %) — against DMS/testUSDC and DMS/testWETH pools.
- Lending (10 %) — supply, borrow, repay, liquidation-safe churn.
- Staking (10 %) — bond, nominate, unbond, claim.
- Bridge simulation (5 %) — testnet DMS to a wrapped test asset.
- Indexer read pressure (5 %) — RPC and indexer reads while writes continue.
Results
Total transactions sent 700
Included in blocks 700
Finalized (GRANDPA) 700
Failure rate 0.00 %
Submit rate 431 tx/s (1.62 s to send everything)
Inclusion p50 / p95 / p99 3.51 s / 4.20 s / 4.26 s
Finality p50 / p95 / p99 16.96 s / 17.66 s / 17.72 s
Every transaction that was sent was included and finalized. Inclusion happened within one 6-second block interval; finality took about three blocks, which is how GRANDPA behaves when it votes on chain prefixes.
The raw output also lists emp_* RPC methods the node exposes but that the client library had not yet decorated — emp_getAiModelStatus, emp_getBlobData, emp_getDomainOwner, emp_getTrackInfo, emp_getValidatorScore. That is a to-do item for the SDKs, and it is in the report because leaving it out would have been hiding something.
What the report does not claim
- It does not demonstrate 1,000 TPS sustained. A 700-transaction burst is a latency and correctness test, not a throughput ceiling.
- It was run against a single local RPC endpoint, not a 4-validator network. The multi-node parallel-execution run is prepared (Zombienet configuration and release binary exist) but had not been executed because Zombienet was not installed in that environment.
- It lists the evidence still required before any mainnet conversation: validator list and regions, Prometheus exports for CPU, RAM, disk and network, GRANDPA finality measurements over time, failed-extrinsic rate under sustained load, and indexer/RPC p95 and p99 latencies.
Cost comparison, with its assumptions visible
The report includes measured testnet fees (a DMS transfer at 0.000025 DMS, a swap at 0.000180, a ZK-verified AI inference at 0.0035) and compares them to other L1s at their observed May 2026 fees. Its methodology section states every assumption — gas prices used, token prices used, which contract calls were assumed for a swap or a mint. Note that the report used a $1.00 reference price for DMS; the canonical public-sale price is $0.65, so the dollar figures for EmpoorioChain in that table are conservative by about a third. The comparison is illustrative, and it says so.
Why publish it this way
Because the alternative — a landing page with a big number and no methodology — is what this ecosystem has explicitly decided to stop doing. The next benchmark will be the four-validator run under artificial latency, and its results will replace these under artifacts/performance/testnet/<date>/. Until then, these are the numbers, and this is what they mean.
Source: PERFORMANCE_REPORT.md, generated 2026-05-30, testnet only.