Strategy Runbooks and On-Call Documentation

A runbook is the document that lets someone other than the strategy's author keep it running safely overnight or on a weekend. It typically states what the strategy trades, what its normal position size and P&L ranges look like on an ordinary day, which upstream data feeds and services it depends on, and — most importantly — a numbered list of "if X happens, do Y" instructions for the failure modes the author has already thought about: a feed going stale, a position limit breach, an unexplained P&L swing, or an exchange connectivity drop.

For example, a runbook entry might read: "If the strategy's overnight position exceeds $2m notional and no corresponding signal fired in the last log entry, kill the strategy via the emergency stop script and page the strategy owner — do not attempt to manually flatten first." Without that entry, an on-call engineer unfamiliar with the strategy's internals has to guess whether an unusual position is expected behavior or a bug, under time pressure and often at a bad hour.

Good runbooks are kept next to the code, updated whenever the strategy's logic changes, and tested by having someone other than the author actually follow them during a drill — a runbook nobody but its author can execute is not really a runbook.