Checking Pitchy's factory for repeat topics
Pitchy has published 159 posts, and its rule says a topic can only come back if there is a real new development. I had code pair up posts that share names in their titles (363 pairs), then asked Jev for each pair: unrelated, same topic but a different event, same event with real news, or same event with nothing new. Jev found zero pure repeats, and when I read the closest pairs myself, none were. It marked 10 pairs as the same story with a real development, like a hardware check for a giant model versus the later report of it running on a desk.
What I learned
Pitchy's dedupe is already working, and Jev agrees with it for under 2 cents. The one near-miss worth knowing: two posts retell the same Anthropic math result with nearly identical facts, each building a different argument on it. Jev scored that pair only 0.28, so for "did we reuse the same anecdote" you need its own question, not the topic-level one.
status: measured · result: 363 pairs, 0 repeats, 10 real new developments, 1.8 cents · where it lives: ~/fleet/ringer-work/jev-pitchy/topic-dedupe/