Marketing runs on claims, and claims run on evidence. When they don't, they're just expensive opinions. Evidence discipline starts with vocabulary. Something is either Observed (usable evidence found) or Unknown (could not verify enough). Unknown is not zero, not a failing grade, and not an invitation to guess. Every claim carries its source and its confidence, and every diagnosis states the evidence basis behind it. A claim without a source is a draft; treat it like one.
Causation language is where discipline matters most, because the temptation to overclaim is constant. A result observed after a task is not proof the task caused it. Say "observed after," "coincided with," "based on," "informed by." Reserve causal claims for evidence that earns them. This isn't pedantry; it's what keeps you from doubling down on tactics that merely happened near good news. The business that says "the email coincided with a lift in sessions" can investigate further. The business that says "the email drove revenue" stops thinking.
Measurement discipline is the other half. Every result names what was measured, which provider supplied it, the data-as-of time, whether it's task-linked or workspace-wide, and whether it's interim or final. Distribution evidence (reach, views, impressions) is real evidence of distribution, and it is never automatically evidence of business impact. Verdicts are exceeded, met, partial, missed, or inconclusive; pending means the measurement is still waiting, and a completed task with no proof is unmeasured, which never displays as met. These rules feel strict until the first time they save you from reporting a win that wasn't one.
Finally, protect the empty states. A system that admits "No target yet.", "No completed tasks yet.", or "No decisions recorded yet." is telling you the truth about its own limits. A team that respects those empty states instead of filling them with guesses builds the only marketing asset that compounds: a record of what actually happened, written down while it was still true.