Opening the book…
When the measured party also reports the measurement, you are not measuring performance, you are measuring their willingness to report it accurately under pressure. The distortion usually arrives as reclassification and timing rather than outright falsehood, which makes it nearly impossible to audit after the fact. An independent collection path keeps the number tied to the event instead of to the incentive.
Prefer system logs, customer-side records, counterparty confirmations, or a sampling process run by someone with no stake in the result. Where self-report is unavoidable, pin it to an artefact the reporter does not control — a timestamp, a receipt, a third party's acknowledgement. Periodically recount a small sample through a second route and compare; a persistent gap is your gaming signal.
Some things only the practitioner knows — perceived effort, confidence in an estimate, near misses — and forcing an external proxy for them gives you a worse number, not a cleaner one. In those cases, keep the self-report but keep it away from consequences.