STR.12:4.3 - Qualify the signal before changing its receiving conclusion

Identify what was actually observed, for which subject, configuration and interval, by what means and with what uncertainty. Separate an observation from a forecast, a revised model and a proposed policy. Missing data do not mean that the monitored condition remains satisfied.

Ask whether the evidence changes a relied-on premise, its applicability, the comparison basis, an adopted selection policy or only the wording of an account. A corrected link with unchanged meaning need not reopen a strategic choice. A new data-use restriction can block a dependent trial even if the expected revenue remains attractive.

Use an adequate existing interpretation. Seek further evidence only when its attainable contribution to this decision warrants its complete burden and delay, under C.11.DUA. A sufficiently adverse whole-cost result can justify stopping further inquiry; uncertainty does not create an obligation to run an experiment.

If the decision needs to compare what proportion of customers or participants took a defined step, and the existing account is insufficient, construct an interpretable reading from the permitted evidence:

1. Name the event and counted unit. A visit, account, person and paying organization can give different answers. State what counts as the event, such as a first paid order, rather than treating every recorded interaction as success.
2. Group units by a relevant common entry rule, such as registration during a named month; this is a cohort. Choose a common period after each unit’s entry, such as its first twenty-eight days. Preserve material differences in offer, eligibility, channel and exposure. The same reporting date does not give newer and older entrants equal observation time.
3. Recover the numerator and denominator under that definition. For a proportion of customers placing an order, count each qualifying customer once in the numerator and retain all customers required by the denominator, including known non-purchasers. Unfinished follow-up or missing records cannot be silently counted as no order or removed to improve the fraction. Reconcile the counts with their source events and transformations where the present reliance needs that check.
4. Report both counts and the fraction, the observation period and material gaps. Compare like-defined readings; return the limitation where the groups are not comparable. A larger total can coexist with a smaller fraction. Totals still matter when the receiving question concerns workload, cash or an authorized limit.

This gives a descriptive comparison, not an effect attributed to the changed product. Different monthly groups can also differ because of selection or surrounding conditions. C.16 supplies measurement interpretation and comparability; C.28 governs a causal use. If the decision needs that stronger answer, use a qualified existing result or consider a bounded comparison under STR.7. An observed equality or a difference not distinguished from uncertainty does not establish practical sameness: the uncertainty must be small enough for the difference that matters to this decision.

When a changed source requires discovering and revalidating several actual receiving uses, use A.10.1 for that bounded work. For one known dependence, the direct source-reliance and subject Method may suffice. Citation alone does not establish dependence.