You are one forecaster in a pre-registered forecasting study. Work alone and quickly.

Read exactly these two files and nothing else:
  1. /Users/eronfalbo/pistomechanics-forecast/forecasts/{STUDY}/protocol.md
  2. /Users/eronfalbo/pistomechanics-forecast/forecasts/primer_baseline.md

Rules (breaking any of them spoils the study):
- Do not open, list or search any other file or directory, in this repository or anywhere else.
- No web access of any kind: do not use WebSearch, WebFetch, curl or any network tool. The study's results may be online and must not be looked up.
- Do not launch subagents. Do not write helper scripts.
- Use only those two files and your own general knowledge.

Task: forecast the study's primary hypothesis exactly as stated under "What you are forecasting" in protocol.md, using its resolution rule, alpha and effect metric.

Write ONE JSON file with the Write tool to:
  {OUT}
with exactly these keys:
{
  "study_id": "{STUDY}",
  "agent": "baseline",
  "run": {RUN},
  "p_primary_hypothesis_supported": <number 0-1: probability the report will show the primary hypothesis supported under the resolution rule>,
  "effect_direction": "favours intervention | favours control | null | other: <text>",
  "effect_size_90pct_interval": {"low": <number>, "high": <number>, "metric": "<the metric named in protocol.md>"},
  "moderator_prediction": "<which subgroup responds more, and why, or \"none\">",
  "claim_ids_used": [],
  "rationale": "<= 120 words"
}
Sign convention for the interval: positive = in the hypothesised direction, as protocol.md defines it.
Then reply with the JSON and nothing else.