Productive deferral requires a preserved option, discriminating evidence, and a convergence rule
Type: kb/types/note.md
Productive deferral is a controlled way to postpone a consequential choice. It requires a preserved option, discriminating evidence, and a convergence rule. Without the option, the choice has already been made or lost. Without the evidence, waiting or further work has no route to a different conditional choice. Without convergence, preservation and learning do not say when or how the choice returns to a decision process. These conditions are individually necessary. Together they do not prove that waiting will be worth its cost.
The option must remain materially live
A preserved option is an alternative that present commitment would destroy or make costly to recover, and that remains exercisable through the later decision point. An alternative retained only in a list is not preserved if expiry, missing preparation, or exercise lead time will make it unavailable when the decision returns.
Real-options theory supplies the primary mechanism. When commitment is irreversible but can be delayed, commitment surrenders the opportunity to observe before deciding whether or when to exercise the future choice. The opportunity can expire, and waiting can carry costs of its own. One live future exercise choice is enough for this condition; productive deferral does not require several parallel alternatives.
Evidence must discriminate between possible actions
Evidence is discriminating only relative to the named decision. At least one possible result of a later observation or bounded test must be capable of changing selection, timing, modification, or abandonment. If every possible result leaves the same action preferred, the delay or activity may produce information, but it does not make this decision better informed. The evidence need not remove all uncertainty.
Evidence may arrive without intervention, or work may produce it. In the second case, focused tests can address a named knowledge gap and eliminate or narrow alternatives before a latest convergence point, as in set-based practice. Such work is evidence-producing only when its possible outputs can change the follow-on decision.
Convergence couples a return condition to a response
A convergence rule names both what returns the choice to decision and what happens next. The return condition may be one or a combination of a date, an observed state, a test result, an evidence threshold, or approaching expiry. The response may select, modify, abandon, authorize more bounded work, or transfer the reopened choice to an authorized decision procedure. The rule must fire early enough for the preserved option to remain feasible.
Different source methods realize this function differently. PMI defines rolling-wave planning as detailed near-term planning with later work kept at a higher level as information improves (PMI). Githens adds a fixed replanning date, assessment of learning, and replanning of the next horizon (Githens). Dynamic adaptive policy pathways use monitored signposts, triggers, mapped responses, and lead-time-aware action (Haasnoot et al.). A date without a response merely schedules discussion. A signal without a mapped response still permits an ad hoc choice.
Deferral still has a cost boundary
Even a controlled deferral can be the wrong choice. Foregone benefit, delay, carrying several alternatives, preemption, lead time, or expiry can outweigh the value of retaining a later choice. Commitment on current evidence or deliberate abandonment can then be warranted. The three-part test rejects false deferrals before this domain-specific comparison; it does not supply a portable quantitative threshold for that comparison.
The test distinguishes four postures
These postures apply the test; they are not an exhaustive taxonomy.
| Posture | Evidence route | Required boundary |
|---|---|---|
| Commit on current evidence | No deferred evidence route is needed because no material option can be preserved, the current decision rule is met, or waiting costs dominate. | Select or abandon now. This is not a deferral posture. |
| Passive waiting | Await a named exogenous observation that can change the conditional action. | Preserve the option, specify a review or exercise rule, and check lead time and expiry. |
| Bounded probe | Pay for limited active work whose possible outputs can change a named decision. | Name the knowledge gap, cost, output, stop condition, and follow-on decision. |
| Maintain alternatives | Keep several feasible candidates live while a named observation or bounded test supplies comparison evidence against shared criteria. | Acknowledge carrying cost, eliminate or narrow candidates with evidence, and set a narrowing boundary. |
Scope
The combined rule is a cross-source inference, not a formulation found in any one source. Real-options timing, rolling-wave replanning, dynamic policy triggers, and set-based narrowing differ in where evidence comes from, how many alternatives they preserve, whether uncertainty declines, and how convergence works. They support only the shared option–evidence–convergence mechanism. This narrow transfer follows the rule that borrowed patterns transfer only over shared mechanism.
The supplied sources do not test whether the combined rule improves planning, delegation, or structural-change outcomes in LLM-agent systems. They also do not support a domain-independent numeric threshold that combines information value with reversal, carrying, delay, and expiry costs. The rule can expose an uncontrolled deferral and the estimates a domain-specific decision still needs; it cannot establish target-side effectiveness or decide universally between deferring and committing.
Operationalized into:
- Write an instruction — requires a consequential deferred choice to retain an option, name discriminating evidence, and define its return and invalidation boundary