> cat rules

Rules

Each rule is one sentence, most with the failure that taught it.

Your agent can run all eight: the vet skill.

  1. 01

    Model fills against the real trade tape.

    Learned from

    A fill model built on price candles predicted that almost every order would fill. Few did.

  2. 02

    Replay only what was public at the time.

    Learned from

    A backtest edge came from knowing a figure would later be revised.

  3. 03

    A number is not citable until its job has finished.

    Learned from

    A per-trade figure from a partial scan was later withdrawn.

  4. 04

    A strong secondary result is not an edge until it replicates as the pre-registered primary.

    Learned from

    One did not.

  5. 05

    Compare with a tolerance, not equality.

    Learned from

    A match rate that looked like a data problem was a floating-point comparison.

    > /reconcile

  6. 06

    A result that looks too good is a bug until you decompose it.

  7. 07

    A pass on the data that chose it needs a second look before it counts.

  8. 08

    Ask whether the question is decidable before you work on it.

    Learned from

    Some needed years of data.

> cat story.md