LLM authoring capture
=====================

Prompt:
  examples/demos/llm_roundtrip_timing/prompt.txt

Model/client:
  Claude Opus 4.8 via Claude Code CLI

Observed round trip:
  76.94s

Generated Heat program:
  examples/demos/llm_roundtrip_timing/claude_response.heat

Compile:
  /usr/bin/time -p timeout 10s /tmp/heatc examples/demos/llm_roundtrip_timing/claude_response.heat -o /tmp/heat_llm_roundtrip_demo

Observed compile time:
  real 0.05

Run:
  /usr/bin/time -p timeout 10s /tmp/heat_llm_roundtrip_demo

Observed run time:
  real <0.01

Manual edits to generated Heat program:
  0

Caveat:
  LLM timing includes Claude Code CLI, client, network, provider, and model
  latency. It is a live observation, not a guaranteed SLA.