{
  "revision": "meridian-iterative-showcase-v1",
  "goal": "Complete usable local-model artifacts through browser feedback and iterative OpenCode repairs.",
  "initial_protocol": "protocol.json",
  "evaluation": "Binary audience preference between current local-model artifacts; Astra is a separate cloud reference. Functional checks are not a numerical quality score.",
  "development": "Codex inspects browser behavior, reports concrete defects, reviews source changes and verifies results. Each local model edits its own artifact through OpenCode with access limited to that artifact. Sessions may be restarted from the saved file for focused follow-up tasks.",
  "provenance": "Original outputs, intermediate snapshots, task prompts, token receipts and diffs are retained. Current builds may have different repair counts and settings; this is not a controlled one-shot capability ranking.",
  "settings_note": "Later repair sessions request chat_template_kwargs.enable_thinking=false. Response limits and any agent step caps are recorded per round. Step caps were removed after earlier rounds stopped before completing edits; response limits were increased after a large edit was truncated. Full request-body capture was not enabled, so the thinking flag is requested configuration, not independently verified runtime telemetry.",
  "public_voting": "Local browser preview only. No public aggregate votes or remote collection enabled. Ballots record artifact SHA-256 hashes.",
  "limitations": "The watches are interactive illustrations, not physical simulations or certified mechanical designs. Browser checks cover documented interactions and viewport sizes, not every device or browser."
}