You are one forecaster in a pre-registered forecasting study. Work alone and quickly. Read these two files first: 1. /Users/eronfalbo/pistomechanics-forecast/forecasts/{STUDY}/protocol.md 2. /Users/eronfalbo/pistomechanics-forecast/forecasts/primer_framework.md You may also consult, and only these: - /Users/eronfalbo/pistomechanics-forecast/data/claims.csv (the claim registry; read it or grep it) - /Users/eronfalbo/pistomechanics-forecast/corpus/ (the framework's source texts) 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; plain grep or reading is fine. - Reason from the framework (pistomechanics) as the primer and registry describe it, plus 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": "framework", "run": {RUN}, "p_primary_hypothesis_supported": , "effect_direction": "favours intervention | favours control | null | other: ", "effect_size_90pct_interval": {"low": , "high": , "metric": ""}, "moderator_prediction": "", "claim_ids_used": ["PM-xxxx", ...], "rationale": "<= 120 words" } claim_ids_used must list every claim_id from data/claims.csv that your rationale relies on (at least one), each an id that exists in that file. Sign convention for the interval: positive = in the hypothesised direction, as protocol.md defines it. Then reply with the JSON and nothing else.