Error Correction Cadence
also Evaluation frequency
How often you judge a project's success or failure: most learning chunks within a day, never more than a week, with difficulty roughly halved after each failure.
Error correction cadence is CF’s rule for how frequently you should reach a verdict on whether you are learning. CF treats slow feedback as the core danger in learning projects: if you only find out you failed after months, you have wasted months. So a learning project should be split into sub-projects small enough that each can be judged within a week, and most should be finished and evaluated in under a day. Going near the weekly limit should be uncommon. The cadence measures time-to-judgment, not time-to-success — failing within a week still counts as on time, because catching failure early is itself the point.
What gets judged is a binary success-or-failure call against an explicit goal, made mostly in isolation so the chunk is evaluable on its own. This operationalizes CF’s wider commitment to fast, frequent error correction: many small checkpoints surface problems while they are cheap to fix.
The cadence pairs with a difficulty rule. After a failure, halve the next project’s difficulty (or cut more) — exponential backoff — and do at least one other, easier project before retrying, ideally a success on the same topic. Repeated failures are exponentially bad: two is far worse than one, and five should essentially never happen, signalling badly misjudged difficulty. This keeps your success rate high and drives failure streak recovery. The maximum difficulty to attempt is roughly 20% above your best success on that topic, judged per-topic, not double it — though with zero failures yet you may probe upward faster to find your existing skill level.