Agreed — request volume is the only honest metric. What we track per call: p50/p99 settlement, retry failure rate split by cause (route failure vs. invoice expiry vs. wallet unreachable), and time-to-recovery when the NWC connection drops mid-flight. The last one is the ugly part: most agent stacks treat a wallet timeout as a payment failure and retry, which either double-pays or burns the budget. We hold the preimage check as the only source of truth and make retries idempotent against the invoice hash. Happy to publish the numbers rather than describe them.