The most expensive architecture decisions often look manageable at approval time.
The first implementation is bounded. Few consumers depend on it. Migration is still described as future work. The decision becomes difficult later, after data, contracts, operating procedures, customer expectations, and several roadmaps form around it.
A useful pre-mortem should expose that future commitment before the organization approves the first version. It will not remove uncertainty. It should change the shape of the decision while options are still available.
For a choice that may become difficult to reverse and broad in consequence, ask five questions.
1. What assumption would force a change in direction?
Name the assumption precisely enough to test.
“The system will scale” is too vague. A useful assumption describes a workload, consistency requirement, consumer behavior, operating constraint, or business condition that the design depends on.
This question identifies the point at which the decision should be reconsidered. Without it, evidence can change while the original commitment continues by inertia.
2. Which dependencies will make reversal difficult?
The component may be replaceable while the surrounding system is not.
List the state, consumers, contracts, procedures, skills, and incentives that will accumulate after adoption. Include the work required to run old and new paths together, backfill data, validate compatibility, change consumer behavior, and retire the previous path.
That work is part of the decision cost. Leaving it outside the review does not make it disappear.
3. Who inherits the consequence without deciding?
An architecture diagram rarely shows the complete blast radius.
Affected parties may include product teams, operators, security or compliance owners, finance, customer support, partners, and customers whose workflows depend on the resulting behavior.
They do not all need approval authority. The decision record should still show whose roadmap, risk, or business outcome can change because of the choice.
4. How will failure appear before full exposure?
Rollback is useful only when the organization can detect the reason to use it.
Define the earliest observable signals, the stop conditions, the observation window, and the person with authority to pause or reverse the rollout. Separate a technical health signal from a product or business consequence. A service can remain available while producing the wrong decision, creating support work, or moving cost into every consumer team.
The detection plan should match the consequence that matters.
5. What would make the first commitment smaller?
The best result of the review may be a different boundary rather than a different technology.
Reduce the first consumer set. Preserve coexistence. Keep a stable contract between the capability and its implementation. Delay the irreversible state transition. Assign an exit owner. Fund the migration path while the current decision still has budget and attention.
These choices purchase information without exposing the complete blast radius.
The answers must change the operating plan
A pre-mortem that produces only a document has not changed the decision.
Use the answers to choose the review level, rollout sequence, ownership model, evidence window, and exit path. If the decision is reversible and contained, keep the process light. If it creates durable state, broad exposure, or coordinated migration, invest more decision effort before the commitment expands.
The purpose is proportionality. Routine choices should keep moving. Scarce leadership attention should be available when a technical decision can remove options for several teams or the business.
When the review needs facilitation
Some decisions stay unresolved because architecture, ownership, product value, and economic consequences are being discussed separately.
The Executive Decision Altitude Review is my paid, fixed-scope diagnostic for one consequential data, platform, distributed-system, or AI capability decision. The work can include a current-state capability and ownership map, a recommended decision altitude, a capability or consumer contract, risks and conditions, and a sequence of next decisions.
The offer is new. It is not presented as prior advisory delivery, a client case study, or a guaranteed outcome. Scope and fee are quoted after a fit conversation.
If one cross-team system decision has become expensive to change, bring the decision, the affected owners, and the evidence currently available. The first conversation should determine whether the problem is specific, consequential, accessible, and suitable for the review.
The five-question pre-mortem is Gevorg's conceptual decision aid. It does not predict a fixed migration time, loss, probability, or outcome and does not report an employer implementation, client engagement, or delivered advisory result.