I measured QNet's real TPS. Here's the honest number — and why my old numbers were wrong.

The throughput figures I published earlier were what most chains actually publish: a lab benchmark — one node, localhost, transactions pumped straight into the engine, no network between machines, no finality. That measures how fast a single process can chew transactions. It's not a blockchain number. I've dropped it.

What I measured instead. A live 5-node testnet on ordinary budget VPS (~€10/month class; the weakest are plain 8-vCPU machines with slow disks) — real network gossip between five independent machines, producer rotation every 30 blocks, failover armed, post-quantum signature verification on every node, deterministic re-execution of every block by every validator, a Merkle state root every block. A transfer only counts when it is finalized by a BFT checkpoint — not when it lands in a mempool, not when it's merely included.

Result: 13,000 finalized transfers per second, sustained for 10 straight minutes. 99.8% of everything submitted was included AND finalized, on 1-second blocks (the 12k rung ran 932 blocks in exactly 932 seconds), hard finality ~109s median. Peak in 5-minute windows: 16,000/s. Push harder and the network degrades gracefully — it queues and self-drains, no forks and no network halt. 13k is the honest sustained ceiling of €10 machines, found by stepping the load up until it broke and confirming the last step that held.

Single transactions (one post-quantum signature each): ~450/s on this hardware. Not a design flaw — the price of quantum safety. An ML-DSA-65 signature is 3,309 bytes, and no standardized batch verification exists for ML-DSA — each signature verifies individually. For scale: one such signature doesn't even fit in Solana's 1,232-byte packet.

This is where QNet's architecture earns its keep. Batch transfers put up to 1,000 payments under ONE post-quantum signature — 3.3 bytes of signature per transfer. Every batch is signed and verified with plain, unmodified FIPS 204 ML-DSA — no custom crypto, the signature simply covers more payload. That's what makes PQ crypto practical at scale today, and it's why a network of €10 VPSes moves 13k quantum-safe transfers a second while a PQ-retrofitted classical chain would manage an estimated ~2-4k.

Potential: signature verification and block application scale with CPU. On Solana-validator-grade servers the same code projects to ~5k singles and 40-80k batched transfers per second.

What's next. The rolling node-update system and super-node onboarding are done and hardened. What's left is consensus soak-testing: if it runs clean for several days straight, I'm opening the testnet. The LLC is also nearly ready — which unlocks publishing the mobile apps to the stores.
Every number above is reproducible end-to-end over public RPC: submit → include → finalize.