This is the archived full report of the run “Devnet native-transfer throughput (superseded)”. Back to the run page. This report is superseded; its numbers must not be cited.
Superseded — do not cite any number on this page. This August 2026 devnet native-transfer report is internally inconsistent: the markup KPIs (944 peak / 426 sustained / 545 busiest block) disagree with the JS data (4,626 / 1,611 / 1,428) that fillKPIs() actually rendered, and both counted foreign and reverted transactions as throughput. Replaced by the mixed-workload report built with bench/build-report.py (see bench/README.md).
Throughput stress test

We flooded the Telcoin devnet
and watched the blocks fill.

Five GKE blasters, each in the same cloud zone as the validator it hammers, fired native transfers as fast as the chain would take them. Throughput here is measured from execution results - transactions actually included in blocks - not from what we managed to submit.

944TPS peak
busiest one-second window, summed across every block the chain stamped that second
Peak throughput
944tps
busiest 1-second window (execution results)
Busiest block
545tx
38% of the 1,428-tx block capacity
Sustained
426tps
avg over seconds the chain was producing
Transactions landed
220,714
of 225,850 accepted; ~10% txpool-rejected at saturation
Blast origin
5zones
one blaster per validator, in-zone

Where devnet lands against other chains

Highest sustained throughput each network has been recorded producing. Telcoin's bar is what we demonstrated in this short flood - injection-limited, not the chain's ceiling.

Demonstrated peak TPS

Telcoin devnet: measured peak this run. Others: max recorded TPS (chainspect.app, Aug 2026).

Telcoin devnet (measured) Other networks (max recorded)

The flood, second by second

Transactions included per second across the run. TN produces several blocks per second under load, so each second sums every block it stamped.

Included transactions per second

Bucketed from the on-chain block series. The dashed line marks single-block capacity (1,428 tx).

Capacity versus what we drew

Each block holds 1,428 native transfers (30M gas / 21k). Under the flood the chain filled entire blocks - and stacked several full blocks into a single second.

Single-block fill

Busiest block vs block capacity.

Blocks per second, under load
up to 5blk/s
consensus stamped five full blocks inside a single second during the drain - that one second is the 4,626 TPS peak.
Gas price paid
~1gwei
base fee sat near the floor throughout; throughput was never fee-limited.

How the number was earned

Throughput claims are easy to inflate by counting submissions. Every figure here comes from blocks the chain actually produced.

01

Pinned wallets

Hundreds of accounts, each pinned to one node so its nonce sequence stays intact in that node's mempool - no gaps to stall block building.

02

Pre-signed, no round trips

EIP-1559 transfers signed locally with managed nonces, fired via raw submission. The client never waits on the chain to build the next tx.

03

Measured from receipts

We scan every block produced during the run and count included transactions - realized TPS, bucketed per timestamp-second.

04

Aggregated, not double-counted

Five blasters, one chain. Submissions sum; landed throughput is taken over the union of blocks so the shared chain is counted once.

Same zone as each node

The validators are spread across five regions on four continents, so "in-zone" meant a small GKE cluster in each - the blaster reaches its node in well under a millisecond.

The comparison, in full

NetworkPeak / max TPSTheoreticalFamily