Evidence, assumptions and confidence register

A risk register earns its value through traceable statements, not neat looking numbers. For each consequential claim, record the statement, original source, collection period, unit of measure, collector, extraction date and whether it is a direct measurement, testimony or assumption. Add a use decision: fit for this scope, fit with conditions, or in need of verification before reliance. A booking system log may reliably time when its form stopped, but it cannot directly count people who would have booked had the form remained available. That second quantity is an estimate that depends on an arrival assumption. A manager's signature does not change the nature of a number: administrative approval does not turn an estimate into an observation.

Separate three kinds of uncertainty. Missingness means a period or channel is absent from the data. Disagreement means two sources report different figures for the same supposed measure. Natural variability means demand changes across days even when measurements are correct. Each calls for a different response: collect the missing interval, reconcile definitions, or use a range and sensitivity check rather than one point. 'Low confidence' alone is not useful; state why confidence is limited, how the gap could affect the decision and what would improve it. A provider's outage clock might include planned maintenance while the service owner's log excludes it. That difference can be definitional rather than a measurement error.

Use a bounded operational confidence scale, not a disguised statistical probability. In this training, 'high' means the measure and period are defined, the original record is available and a relevant second check supports it. 'Medium' means the source is available with a known limitation unlikely to reverse the decision. 'Low' means a gap or conflict could reverse priority. Do not add these ratings arithmetically or translate 'high' into 90% without calibration. Event count may have high confidence while user impact has low confidence; attach the rating to the claim, not to the whole risk.

When sources conflict, retain both versions and their retrieval identifiers. First compare the event definition, extraction dates, time zones and event identifiers. Ask the record owners to reconcile the difference, and show how each interpretation affects the action. If both lead to the same action, an owner can take a precautionary step while the gap is documented. If the action changes, escalate with an interim option and review date. The specialist must not select the number that favors a preferred recommendation. If complaint records contain personal data, use only what is needed for traceability and follow the organization's access and retention policy; these training examples use fictional identifiers.

Connect confidence to action through a simple sensitivity test. If the weekly escalation trigger is 55 minutes and one log shows 60 while the other shows 52, the disagreement matters: one passes the trigger and the other does not. Show both test results to the owner, explain that the exercise's approved criterion calls for the service log, and request reconciliation. If instead the logs show 60 and 58, both trigger escalation; action can proceed while the evidence gap is repaired. Uncertainty is neither an automatic reason to suspend precaution nor a license to choose a convenient number. Keep a link or event identifier rather than pasting unnecessary personal details into the risk card, and specify who can inspect the original. When a documented correction arrives, add a new row version with date and reason instead of silently erasing the old value. The change history matters when another team later asks why an owner made an earlier decision with different information.