#!/usr/bin/env python3
"""Export visible assistant replies for the fixed EP1 cohort (no reasoning/headers/tools)."""
import hashlib, json, pathlib
ROOT=pathlib.Path(__file__).resolve().parents[1]
SNAPSHOTS=['exp-20260915-ep1-f005-docker.json','exp-20260918-ep1-f005-glm-probe1200.json']
def visible_reply(path):
    response=json.loads(path.read_text())
    body=response.get('body')
    chunks=[]
    if isinstance(body,dict):chunks=[body]
    elif isinstance(body,str):
        for line in body.splitlines():
            if not line.startswith('data: '):continue
            try:chunks.append(json.loads(line[6:]))
            except json.JSONDecodeError:continue
        if not chunks:
            try:chunks=[json.loads(body)]
            except json.JSONDecodeError:pass
    text=[];has_tools=False
    for chunk in chunks:
        for choice in chunk.get('choices',[]):
            delta=choice.get('delta',choice.get('message',{}))
            if isinstance(delta.get('content'),str):text.append(delta['content'])
            has_tools=has_tools or bool(delta.get('tool_calls'))
    return ''.join(text),has_tools

def main():
    runs=[]; excluded_titles=[]
    for name in SNAPSHOTS:
        for row in json.loads((ROOT/'data/experiments'/name).read_text())['runs']:
            if row['task']!='ep1-f005':continue
            messages=[]
            for p in sorted((ROOT/'data/runs'/row['run_id']/'turns').glob('*.response.json')):
                text,tools=visible_reply(p)
                try: title=json.loads(text)
                except (json.JSONDecodeError,TypeError): title=None
                if isinstance(title,dict) and set(title)=={'title'} and isinstance(title['title'],str):
                    excluded_titles.append({'source':str(p.relative_to(ROOT)),'sourceSha256':hashlib.sha256(p.read_bytes()).hexdigest()})
                    continue
                if text.strip():messages.append({'source':str(p.relative_to(ROOT)),'sourceSha256':hashlib.sha256(p.read_bytes()).hexdigest(),'text':text,'hasToolCalls':tools})
            assert messages, row['run_id']
            runs.append({'id':row['run_id'],'model':row['model'],'attempt':row['attempt'],'budget':row['params']['timeout_s'],'messages':messages})
    assert len(runs)==20
    out=ROOT/'site/evidence/ep1-reference/answer-corpus.json'
    out.write_text(json.dumps({'scope':'Visible assistant content only. No reasoning_content, request headers, tool arguments or tool outputs. Fixed 20-run cohort. Excludes auxiliary JSON-only title replies.','excludedAuxiliaryTitles':excluded_titles,'runs':runs},ensure_ascii=False,indent=2)+'\n')
    print(json.dumps({'runs':len(runs),'excludedAuxiliaryTitles':len(excluded_titles),'visibleMessages':sum(len(r['messages']) for r in runs)}))
if __name__=='__main__':main()