A testnet is only as useful as what it measures. This post lists the instruments EmpoorioChain has, what each has recorded, and the gaps — because the previous version of this page reported metrics for a different network.
Finality
Exported by scripts/finality_prometheus_exporter.js, scraped by Prometheus:
empoorio_finality_seconds{quantile=latest|p50|p95|p99}— seconds between consecutive finalized headsempoorio_finalized_head,empoorio_finalized_head_age_secondsempoorio_finality_exporter_up
Recorded (May 2026 mixed-load run): finality p50 16.96 s, p95 17.66 s, p99 17.72 s. The mainnet gate wants p50 < 12 s and p95 < 30 s under real load; p50 is not yet met.
Inclusion
From the same run: p50 3.51 s, p95 4.20 s, p99 4.26 s — within one 6-second block interval.
Throughput
The honest figures: 700 transactions sent in 1.62 s (431 tx/s submit rate), 700 included, 700 finalized, 0.00 % failures. Sustained throughput toward the 1,000 TPS transfer target and 200 TPS DEX target has not been demonstrated; the parallel-execution figures in the repository are scheduler projections, labelled as such, and the four-validator run is prepared but unexecuted.
Transaction pool
emp_txpool_pending_txs, emp_txpool_dag_width, emp_txpool_dag_depth, emp_txpool_duplicate_txs, emp_txpool_propagation_delay_ms_p95 — the shape of the DAG mempool and the network's gossip health.
Fees
Measured with scripts/fee_snapshot.py, stored in artifacts/fees/testnet/fee_snapshot.json: transfer 0.000025 DMS; NFT mint 0.000075; batch of 100 mints 0.0022; DEX swap 0.00018; lending deposit 0.00015; 1 MB blob 0.000095; ZK-verified AI inference 0.0035.
Validators
scripts/testnet_validator_audit.py reads Session::Validators from live RPC and writes a dated JSON artefact. Current: 2. The Grafana dashboard validator-decentralization.json tracks peer count, best and finalized block per node, lag, uptime, client version.
Activity
Signed transactions per day — the metric that would have exposed the 102,912-block silence — is now tracked. First signed transaction 7 September; first DID and first DEX swap 14 September 2026.
What is not measured
- Weights on the reference machine. Runtime weights are hand-written. The reference (4 vCPU / 7 GB EPYC-Milan) is chosen; renting it and running
benchmark palletfor every critical pallet has not happened. The tokenomics document therefore refuses to state any TPS as a production claim. - Block weight utilisation at peak.
- Public RPC uptime over 30 days.
- Multi-validator behaviour under artificial latency (Zombienet 4 and 10 nodes).
Where results go
artifacts/performance/testnet/<date>/, artifacts/fees/testnet/, artifacts/decentralization/testnet/. Each has a date and a commit. The next performance post will cite one of them, or it will not be written.
Based on CONSENSUS_STATUS.md, PERFORMANCE_REPORT.md, MEMPOOL_DAG_ORDERING.md, FEE_MODEL_AND_LOW_COST_STRATEGY.md and MAINNET_GATE.md.