The bright data row is wrong, with the 5k free tier 6000 answers cost about $1.50 instead of $9.00, and the best fix is one column for effective usd per 1000 successful answers. Parse's 82 day study is the strongest case for reruns, just keep 52.6% to 28.9% survival separate from median 0.18 overlap and keep the small panel caveat

The 30 to 50 runs rule does not hold up since reruns correlate, close claims need more prompts not more runs. Tag empty arrays with not_requested, not_triggered, unsupported or parse_failed so nothing served never reads as nothing captured