Loading...
Week 2 — Apply · Day 13 of 30
Pick a Series A from last quarter. What did the composite say at month -3?
Yesterday
Yesterday you found the signal that carries your beat. Today: a backtest. One known round, scored at month -3.
Backtesting on one known case proves to yourself that the framework wasn't just confirming pattern-matching on three orgs of your choosing. Pick a startup that announced a Series A 60-90 days ago and read what the composite would have said three months before the announce.
A score of 4/6 or higher at month -3 is a hit — the framework would have flagged this round before it closed. A score of 2/6 or lower is a miss — useful too, the panel data says ~30% of rounds don't surface in GitHub signals.
Some recently-funded orgs went private with key repos right before the round. That itself is a tell — public→private repo flip is a signal we don't formalise but is worth a manual check.
Bonus
Run this calibration on five rounds and you'll have a personal hit-rate. 60-70% accuracy at month -3 is realistic for the public-data version. The MCP version layers in private-data heuristics and runs ~78%.
Tomorrow
Tomorrow closes Week 2 — your first 3-startup scorecard becomes a real artifact you can show.
Curriculum: /challenge · Methodology: /methodology · Paper: ssrn.com/abstract=6606558