scripts/bench/02_launches.py

Derived analysis output (scratchpad scripts/bench/02_launches.py).

This is a derived table/summary; the finding's excerpt is usually a paraphrase of its counts, so it may not be highlighted verbatim.

"""Launch structure: first wall-time appearance of each (family, cohort-tag) instance key; hourly/daily counts.
Cohort tag = Mon+DD found in signature or near cohort/run/task words (approximate instance key; one agent may post
under a new tag, and 2 instances can share a tag -> counts are approximate)."""
import json, re, collections, csv
S = '/tmp/claude-0/-root-swarm-hackathon/a8e4c17a-c034-48ae-9f11-70fc0a8841c9/scratchpad'
MON = r'(Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|Dec)[a-z]*[ -]?(\d{1,2})(?!\d)'
NEAR = re.compile(MON + r'(?:[ ,]+20(2[5-9]))?[^.;:]{0,12}?\b(cohort|run|task|twin|live|scaffold|slow|fast|tier|OAI|update|reply|R\d|here|watcher|scout)', re.I)
rows = [json.loads(l) for l in open(f'{S}/derived/bench/lines.jsonl')]
first = {}
years = collections.Counter()
for r in sorted(rows, key=lambda r: r['t']):
    f = r['fam']
    if not f: continue
    keys = set()
    if r['sig']:
        for a, b in re.findall(MON, r['sig']):
            keys.add(a[:3] + b.zfill(2))
    m = NEAR.search(r['line'])
    if m and not keys:
        keys.add(m.group(1)[:3] + m.group(2).zfill(2))
    for m2 in re.finditer(MON + r'[ ,]+(202[5-9])\b', r['line']):
        years[m2.group(3)] += 1
    for k in keys:
        if int(k[3:]) > 31: continue
        first.setdefault((f, k), r['t'])
# launches table: per family per day and hour
by_fh = collections.Counter(); by_fd = collections.Counter()
for (f, k), t in first.items():
    by_fh[(f, t[:13])] += 1; by_fd[(f, t[:10])] += 1
with open(f'{S}/derived/bench/launches.csv', 'w', newline='') as fo:
    w = csv.writer(fo); w.writerow(['family', 'hour_utc', 'new_cohort_tags_first_seen'])
    for (f, h), n in sorted(by_fh.items(), key=lambda x: (x[0][1], x[0][0])):
        w.writerow([f, h, n])
with open(f'{S}/derived/bench/cohort_first_seen.csv', 'w', newline='') as fo:
    w = csv.writer(fo); w.writerow(['family', 'cohort_tag', 'first_seen_utc'])
    for (f, k), t in sorted(first.items(), key=lambda x: x[1]):
        w.writerow([f, k, t])
fams = sorted({f for f, _ in by_fd})
days = sorted({d for _, d in by_fd})
print('family'.ljust(22), ' '.join(d[5:] for d in days), 'total')
for f in fams:
    print(f.ljust(22), ' '.join(str(by_fd.get((f, d), 0)).rjust(5) for d in days), sum(v for (ff, _), v in by_fd.items() if ff == f))
tot_h = collections.Counter()
for (f, h), n in by_fh.items(): tot_h[h] += n
print('\nhourly new cohort tags (all families):')
for h in sorted(tot_h): print(h, tot_h[h], '#' * min(tot_h[h], 80))
print('simulated years mentioned', years)