In the last two blog entries we discussed the desire for a definition of accuracy and the need to settle for a “pseudo” definition to adequately define accuracy in relation to an agreed-to standard as represented in a system of record. I can summarize my three suggestions:

• Agreeing to a standard system of record.
• Asserting a set of rules that can be used to minimally ensure the qualification of the system of record as trustworthy (such as timeliness, auditability, synchronization).
• Instituting comparisons against the system of record as a way to measure accuracy.

This is a little bit of a shell game, since measuring accuracy of one data set is really measuring consistency with the system of record - which is also subject to correctness issues. However, what can be said is that if there is an agreement to the system of record, then the percentage of records that are consistent with the system of record is a viable (albeit not perfect) measure of accuracy.

It is worth introducing a subtle difference between what could be defined as a measure and what could be defined as a metric. The measure is the raw percentage of records that are consistent with the system of record. The metric can be defined in terms of that measure coupled with the assurance that the system of record complies with its agreed-to quality rules. Without that assurance, the consistency is irrelevant, so instituting corresponding auditable measures of the quality of the system of record provides an agreeable standard for accuracy, even if everyone knows it is not perfect.