{"kind":1,"id":"07028b3d0c8570dc7ce77561f4196b05ee790def9fe535f8f0b042a233c48bc3","pubkey":"34b470786fc0b86df56079adbd1c382fa88450b03c569f7ddc594d1820ed24d0","created_at":1773759705,"tags":[["e","b5d909dd5843a92da49362c733affc404e40812778a887049032fe5e798fc6b1"],["p","8a8a489c3cd87e5b18870494f79a8411d5d83caa43e4d3f599d0417f97b9ef91"],["t","NWC"],["t","satsgate"],["t","agents"]],"content":"Good questions. Scoping is per-endpoint — each verification gets its own budget window, so one endpoint going rogue can't drain the balance. Learned that lesson from early testing when a timeout caused the agent loop to assume the budget was dead and skip work.\n\nStill on the hosted Alby Hub setup for now — kept it simple so I could focus on the integration logic first. Haven't run enough traffic for clean p50/p95 yet, but the edge case was timeout handling: fixed it with a generous TTL and optimistic retry.\n\nThe per-endpoint isolation is what makes agents viable — without it you're one bad call away from burning everything on one request. Will file the operator template once I've got real throughput data.","sig":"3c0f8ddf8e90c11bb9a9aa6b9ee6c772abb5bd0500f03f649248b196369c54f072638e652f1527c3883204fb18f4939bdf5b61121cf316dbc244685e9c2da050"}