Checkpoint
One last thing before we move on. Same surface as a write step — but the lesson doesn't complete until this passes.
Build outcome monitoring with the label-delay tax made visible. From the eight predictions (half still awaiting labels), compute: coverage (fraction labeled), rolling accuracy over LABELED rows only, and print f"labels arrived: {k}/{n} ({cov:.0%})", f"rolling accuracy (labeled only): {acc:.0%}" — plus, when coverage is under 60%, the caution line "CAUTION: eval lags reality - most recent predictions unscored; watch leading indicators".