Construct alternatives only where their consequences matter

Suppose the recording check leaves a real obstacle to using the information, and the team asks whether restoring the earlier handoff rule would reduce planning delay. PSD.5 retains the concern account and the timeline for their own questions. Use an adequate existing causal model directly. When the causal relations are still missing, C.28.CM constructs the needed contribution.

Fix the outcome and horizon before comparing stories. Let X be the rule actually used for a handoff, M whether the receiving information is usable at planning time, Y the delay in assigning that work, and Z the load during the same period. R records what the register says about completeness; S records whether the row was selected for this investigation. Do not replace M with R merely because R is easier to obtain.

One account proposes X→M→Y: the new rule changed the identifier correspondence, which obstructed use and delayed assignment. Load may also have prompted the rule change and delayed work, giving Z→X and Z→Y. The common cause does not erase the proposed mediated path. Under this model, a correctly applied correspondence repair can remove one source of delay while leaving the load-related delay.

A rival account makes load the cause of both information arriving late and assignment delay: Z→M and Z→Y, with Z→X explaining why the rule changed at the same time. It omits X→M. Returning to the earlier rule then leaves those delays unless it changes another mechanism. These accounts may be combined if both mechanisms operate. Their arrows require process knowledge and evidence; the temporal coincidence of a template change and delay selects neither account.

The useful consequence is now explicit: restoring the rule can change M in the first account but does not do so in the second under their stated mechanisms. Trace an affected identifier and the availability times to challenge that difference. If the disagreement is only about how to present known facts, causal construction adds no result. If the necessary process fact is missing, return that fact request instead of completing a graph by intuition.

Selection also matters. If rows enter the investigation because their information is unusable or their assignment is delayed, both M and Y affect S. Restricting attention to S=1 can create dependence between M and Y even without a causal effect between them. Its direction and size do not follow from the three-node shape alone. Use the whole-path rules of C.28.CM for the chosen graph. Address this selection before using those rows to infer a relation across all handoffs.

Separate the start, persistence and amplification of the episode. A correspondence change may start it; a continuing stream of unresolved rows with no correction owner may sustain it; repeated assignment changes may amplify the rework. These are hypotheses about different outcomes and times, not three interchangeable names for one root cause. When delay feeds future load, represent Y(t)→Z(t+1) with the relevant earlier influences. Do not call an unqualified directed cycle a DAG.