Separate the decision from the outcome

What the paper develops

Getting home is not the whole outcome

Suppose a drunk driver arrives home uninjured and in record time. If the scorecard records only his arrival time and his own condition, the trip looks successful.

But on the way, he hits a cat, breaks several traffic laws, and drives through someone's yard. His decision to drive was reckless. The way he drove broke laws and damaged property. The trip harmed others. Once the whole trip is counted, "made it home" is not a successful outcome.

Portfolio reviews make a quieter version of the same mistake when they treat a delivered benefit or an early finish as proof of good judgment. They collapse three questions: What was achieved? How was it achieved? Was the original decision sound given what was known at the time?

Keep the result, the execution, and the decision separate

First score the full outcome. Did the program deliver the promised benefit within the approved cost, schedule, and risk limits? What collateral costs or harms accompanied it? A narrow target does not define the whole outcome.

Then inspect execution. Was the approved plan followed? Were controls bypassed, dependencies ignored, or risks shifted elsewhere? The route matters even when the destination is reached.

Finally grade the original decision. Was the problem framed correctly? Were realistic alternatives compared? Were assumptions, estimate ranges, dependencies, and accepted risks explicit? Was the choice reasonable given the evidence available at approval?

To make that comparison possible, preserve a short record at approval: the problem, alternatives, assumptions, estimate ranges, accepted risks, and evidence that would change the decision. At review, keep anything learned later in a separate column.

The decision grade and full-outcome score produce four combinations. Sound/good work gets celebrated; weak/bad work gets investigated. The mixed cases carry more learning. A sound decision with a bad result tests whether leaders will protect disciplined risk-taking. A weak decision with a good result tests whether they can question a success before its tactics become mandatory. The approved-versus-executed route and the variance pattern inform the attribution.

Attribute the gap before assigning credit

Start with the declared range. A result inside a well-calibrated range is uncertainty resolving. Next compare the approved plan with what was executed: changed scope, missed controls, or ignored dependencies belong to delivery governance. Finally examine comparable decisions. Repeated misses in the same direction are evidence that the estimating model or gate criteria need correction.

These categories overlap, so the review should record the attribution and the evidence behind it rather than force false certainty. The full-outcome score never edits the decision-quality grade. Both grades and the attribution record should appear in the committee packet.

One result should not settle a sponsor's reputation or an estimating model. Calibration accumulates across comparable decisions. Scattered errors around the estimate may indicate an imprecise but roughly centered process; repeated misses in the same direction are a reason to investigate systematic optimism, omitted dependencies, or benefit definitions that do not survive delivery.

Change the incentive inside the review

Review successful programs as well as failures. Report calibration across comparable decisions, not sponsor rankings from isolated outcomes. Protect a sponsor whose documented decision was sound and penalize any attempt to retrofit the record after results arrive.

The practical shift is small: add a decision-quality grade beside every full-outcome score, preserve the approval-time record, and close every material variance with one attribution record and a named update to the model, delivery system, or accountability record. That separation turns retrospective review from a verdict into an operating-control loop.