{
  "environment": "Real web app, separate dev servers and fresh databases, Linux Chrome, 1280x800 CSS pixels, light theme",
  "seed": "Same server settings RPC: reasoningEffort=high with unadvertised contextWindow=1m; same catalog model on both sides",
  "base": {
    "commit": "9cd7a04797184c92a8f636ae1e918f42c139d850",
    "afterInitialSend": { "savedEffort": "high", "draftContainsStaleContextWindow": true },
    "afterAgentConfigure": { "savedEffort": "medium", "composerEffort": "High" },
    "afterUntouchedFollowup": { "savedEffort": "high", "agentReadback": "Follow-up saved effort: high", "composerEffort": "High" }
  },
  "head": {
    "commit": "77a02ecbce6df963f11bd4af0680b8192a15a185",
    "afterInitialSend": { "savedEffort": "high", "draftReleased": true },
    "afterAgentConfigure": { "savedEffort": "medium", "composerEffort": "Medium" },
    "afterUntouchedFollowup": { "savedEffort": "medium", "agentReadback": "Follow-up saved effort: medium", "composerEffort": "Medium" }
  },
  "evidenceProcessing": "Continuous recordings retain original timing; both cropped identically to conversation and composer. Runtime identifiers masked. Environment and commit caption added. Still frames captured after identical screenshot pairs over 500ms apart. No application state or effort labels edited.",
  "attribution": "AI-assisted; verified with the tests and tools listed above."
}