"""Dump candidate clusters to blind label sheets (SELECTION.md, protocol steps 1-3).

  python3 make_sheets.py

Writes, for development (seed 11) and test (seed 12):
  candidates-<set>.json   cluster id -> member chunk indices (what signals are computed on)
  sheet-<set>.txt         per cluster, member summaries only, shuffled; no titles, no scores
  labels-<set>.json       an empty template to fill in: {id: {"label": null, "name": ""}}

Cluster ids are assigned in a shuffled order so position carries no information.
"""
import json, pathlib, random, sys

K = 60
SETS = {'dev': 11, 'test': 12}

sys.argv = ['measure.py', 'summary_embeddings.json']
exec(pathlib.Path('measure.py').read_text(encoding='utf-8').split("print(f'{path.name}")[0])

summaries = json.loads(pathlib.Path('summaries_clean.json').read_text(encoding='utf-8'))['summaries']
text = {x['chunk']: x['summary'] for x in summaries}

for name, seed in SETS.items():
    cands = []
    for c in kmeans(K, seed):
        size, nst, dom, _ = describe(c)
        if size >= MIN_SIZE and nst >= MIN_STORIES and dom <= MAX_DOMINANT:
            cands.append(sorted(c))
    rnd = random.Random(seed * 1000 + 7)
    rnd.shuffle(cands)
    prefix = name[0].upper()
    ids = {f'{prefix}{i + 1:02d}': c for i, c in enumerate(cands)}

    pathlib.Path(f'candidates-{name}.json').write_text(json.dumps(ids, indent=1))
    lines = []
    for cid, members in ids.items():
        shown = members[:]
        rnd.shuffle(shown)
        lines.append(f'=== {cid}')
        lines.extend(f'  - {text[i]}' for i in shown)
        lines.append('')
    pathlib.Path(f'sheet-{name}.txt').write_text('\n'.join(lines), encoding='utf-8')
    labels = pathlib.Path(f'labels-{name}.json')
    if not labels.exists():
        labels.write_text(json.dumps({cid: {'label': None, 'name': ''} for cid in ids}, indent=1))
    print(f'{name}: seed {seed}, {len(ids)} candidate clusters, {sum(map(len, ids.values()))} scenes')