Every GPU monitoring tool on the market is built and run by the side billing you. DCGM, provider dashboards, uptime pages, all of it lives on their infrastructure, reporting numbers they have every incentive to round up.
That's not a conspiracy, it's just structure. "Up" is what they bill for, so "up" is what their tools optimize to show. A throttled, ECC-flapping, half-dead node still reads green on a dashboard built by the party getting paid for green.
So we asked a dumber question: what if the monitor just... didn't work for them. What if it sat on your side of the contract instead.
We're calling this Counterparty Attestation. A GPU contract has two sides with opposite incentives. Every tool today is built by one of them. Attestation means the witness comes from the other side, the buyer, not the biller.
In practice: Goodput runs a lightweight agent on each node, reads hardware ground truth (NVML clocks, ECC, InfiniBand, Xid faults), no process data, no workloads, ever, and turns it into a capability weight per node.
Uptime dashboards say "node up, 100%." Goodput says "capability weight 58%, Xid 79, node fell off the bus." Same node, two different truths, and only one of them is useful in a billing dispute.
Then it does the part no dashboard does: exports a timestamped SLA claim pack, mapped to provider tiers, built to be sent with the invoice, not stared at in a browser tab.
We're not trying to out-dashboard CoreWeave Mission Control or out-audit SemiAnalysis's ClusterMAX. Those are good tools solving "help the provider fix things faster." We're solving a narrower, meaner problem: give the buyer something to put on the table when the provider says everything's fine.
Launching on Product Hunt today. If you're paying for GPU-hours and have ever wondered whether you got all of them, this one's for you. Waitlist's open.
