Value of Information¶

Origin domain: Decision Science
Subdomain: decision analysis → Decision Science
Also from: Statistics, Economics & Finance, Medicine, Engineering
Aliases: Expected Value of Information, Decision Value of Information, Information Value
Related primes: Expected Value, Uncertainty, Decision, Experimental Design, Opportunity Cost, Prioritization

Core Idea¶

Value of Information measures what evidence is worth before it is obtained by comparing two decision situations. In the first, an actor chooses now using current beliefs. In the second, the actor first observes a possible item of evidence, updates, and then chooses the best action available under that result. The expected improvement in outcome is the gross value of the information; acquisition cost, delay, side effects, and lost flexibility determine whether its net value is positive.

The identity is decision-sensitive rather than knowledge-maximizing. Evidence can greatly reduce uncertainty yet have no value if every possible result leaves the same action optimal. Conversely, a modestly accurate test can be extremely valuable when plausible results fall on opposite sides of a consequential decision threshold. The object being valued is not the data in isolation but the option to condition action on it.

Structural Signature¶

Sig role-phrases:

the pending decision — a choice among actions that can still be altered
the current belief state — uncertainty over relevant states and consequences before new evidence
the baseline best action — the action with the highest expected value under current information
the candidate information source — a test, observation, investigation, or delay that may produce one of several results
the result-conditioned update — the belief revision associated with each possible observation
the contingent best actions — what would be chosen after each result
the expected improvement — the probability-weighted gain from being able to condition action on the result
the acquisition burden — monetary cost, time, harm, attention, disclosure, or option loss caused by obtaining the evidence
the decision window — the latest time at which the evidence can still affect action

What It Is Not¶

Not information quantity. Bits, sample size, or predictive accuracy do not determine value without a decision and payoff structure.
Not uncertainty reduction alone. Resolving a fact is decision-worthless when every possible answer recommends the same action.
Not expected value alone. Expected Value supplies the probability-weighted aggregation; Value of Information adds the counterfactual comparison between acting now and acting after evidence.
Not the realized benefit of lucky news. The quantity is evaluated ex ante across all possible results, including results that confirm the current action.
Not a command to gather more data. The correct result is often zero or negative net value, especially when evidence is costly, slow, or unable to change a commitment.
Not experimentation itself. Experiments are information-producing instruments; this abstraction ranks whether and which one is worth running.

Broad Use¶

Medicine uses the pattern whenever a diagnostic test is ordered because its result may change treatment. Engineering uses it to choose among prototype, load, materials, and reliability tests before committing to a design. Scientific research uses it in experimental design: among competing observations, prefer those expected to discriminate among live hypotheses in a way that changes what should be investigated or built next.

The same comparison appears in intelligence collection, due diligence, environmental monitoring, maintenance inspection, legal discovery, insurance investigation, and product discovery. A military planner asks whether reconnaissance can change route or timing before the operation begins. An investor asks which diligence question can overturn the thesis rather than merely decorate the memo. A product team asks which test can kill or redirect a roadmap while redesign remains cheap. A homeowner asks whether an inspection is worth its fee before buying. The evidence and utilities vary, but each case compares the best current action with the expected best result-conditioned action.

Clarity¶

The abstraction dissolves the common equation "more accurate information is more valuable." Suppose two treatments have the same recommendation across the entire range a test could plausibly reveal. The test may measure the patient exquisitely and still have zero decision value. Now suppose a crude test has only two outcomes, but one outcome makes treatment clearly beneficial and the other makes it harmful. Its discriminatory resolution is low while its decision value is high.

The prime also separates gross and net value. Evidence can improve the choice yet be irrational to acquire because the test costs too much, takes too long, introduces harm, reveals private information, or forces an irreversible pause. A result arriving after the decision window is epistemically interesting and operationally worthless for that decision.

Manages Complexity¶

Real projects contain more unknowns than can be investigated. Without a ranking rule, teams test what is easy, visible, politically safe, or already instrumented. Value of Information compresses that open-ended agenda into a tractable frontier: which uncertainty can change the action, how large is the expected improvement, how cheaply and quickly can it be reduced, and how much downstream commitment remains avoidable?

This shifts attention from confidence-seeking to decision-changing evidence. Questions whose answers cannot alter the plan fall away. Questions far downstream are delayed until upstream assumptions survive. Cheap imperfect signals can dominate expensive precise studies when they are sufficient to route the next action. The abstraction thereby manages both experimentation cost and the hidden cost of continuing under avoidable uncertainty.

Abstract Reasoning¶

Let the current information be I and the available actions be A. Compute the maximum expected outcome achievable now: choose the action whose expectation is highest conditional on I. Then enumerate the possible results x of the candidate evidence. For each x, update beliefs, choose the best action conditional on I and x, and weight that outcome by the probability of x. The difference between this result-conditioned expectation and the current maximum is the gross value of information.

The order of operations matters: update, optimize separately under each result, then average. Averaging the evidence first and choosing once erases the option to respond and understates the value. Perfect information gives an upper bound because no feasible imperfect test can improve the decision more than knowing the relevant state. Net value subtracts the full acquisition burden and is sensitive to timing, reversibility, and whether action can actually be conditioned on the result.

Knowledge Transfer¶

Clinical decision thresholds teach product and policy work to ask whether any plausible result changes the next commitment. Engineering test programs teach research planning to front-load evidence while redesign remains cheap. Intelligence collection teaches organizations that highly interesting questions can rank below less glamorous ones whose answers alter action. Scientific experimental design teaches due diligence to prefer discriminating tests over piles of confirmatory evidence.

The transfer rule is portable: list the decisions still open, name the outcomes that would switch them, price the benefit of switching correctly, and seek the cheapest credible evidence that can separate those outcomes before the window closes. This discipline often recommends a coarse early probe, followed by a more precise study only if the first result moves the case into a region where precision matters.

Example¶

An engineering team must choose between a cheap narrow design and a costly general design. Current evidence slightly favors the narrow design, but uncertainty remains about the operating regime. A two-day prototype test can return one of two results. Under the first, the narrow design remains best; under the second, the general design avoids a large failure cost. The team weights the best result-conditioned choice under each outcome and compares that expectation with choosing the narrow design now.

If the expected avoided loss exceeds two days of cost and delay, the test has positive net value. If both outcomes would leave the narrow design optimal, the same measurements have zero decision value even if they are scientifically accurate. The roles map directly: design choice, current beliefs, baseline action, test, result-conditioned update, contingent designs, expected improvement, acquisition burden, and decision window.

Structural Tensions¶

T1: Precision versus actionability. More precise evidence may not cross a decision boundary, while a coarser test may. Diagnostic: calculate how each possible result changes the optimal action, not merely the posterior variance.

T2: Information gain versus acquisition harm. Tests can consume samples, expose privacy, delay treatment, reveal strategy, or create market movement. Diagnostic: include every burden caused by obtaining or acting on the evidence, not only the test fee.

T3: Early cheap evidence versus late accurate evidence. Accuracy often improves with time just as flexibility disappears. Diagnostic: price evidence jointly with the remaining option to change course.

T4: Perfect-information upper bound versus feasible test. An uncertainty can matter enormously while no available observation resolves it well enough to be worth buying. Diagnostic: distinguish the value of knowing from the value of the actual instrument.

T5: Local decision value versus reusable knowledge. A result may not alter today's action but may benefit future cases. Diagnostic: state whether the valuation scope is one decision, a portfolio, or a durable learning system.

T6: Formal utilities versus heuristic ranking. Exact analysis can be fragile when probabilities and payoffs are poorly known; simple rankings can hide decisive nonlinearities. Diagnostic: use formal calculation where inputs support it and sensitivity-tested heuristics where they do not.

Structural–Framed Character¶

Value of Information is structural with a modest agentive dependence. Its identity is a before-and-after comparison among uncertainty, evidence, conditional actions, and expected outcomes, and that relation is recognized unchanged across domains. It presupposes a decision-maker or rule capable of responding to evidence, but no particular institution, norm, or substantive goal is constitutive.

Substrate Independence¶

The prime is highly substrate-independent across decision substrates. Tests may inspect bodies, materials, markets, environments, hypotheses, or designs, yet the same ordering remains: unresolved state, evidence outcomes, belief update, contingent action, and expected improvement. Its only durable residue is agentive—there must be a rule or actor able to condition action on the result.

Relationships to Other Abstractions¶

Current abstraction Value of Information Prime

Parents (3) — more general patterns this builds on

Value of Information presupposes Decision Prime

Information has decision value only relative to a choice whose selected action could change or improve after evidence arrives.
Value of Information is part of Expected Value Prime

Value of Information averages decision improvement over the possible evidence outcomes and their probabilities.
Value of Information presupposes Uncertainty Prime

The comparison presupposes unresolved uncertainty about states or consequences that additional evidence could reduce.

Children (1) — more specific cases that build on this

Riskiest Assumption Test Domain-specific is a decomposition of Value of Information

Removing lean-startup vocabulary leaves the rule of buying the evidence expected to improve downstream decisions most relative to its cost.

Neighborhood in Abstraction Space¶

Value of Information has no computed distinctiveness yet.

Family — Unclustered & Miscellaneous (429 primes)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-26

Distinction from Neighbors¶

Against Expected Value, this prime contains two optimized expectations—without and with evidence—and values their difference. Against Uncertainty, it asks whether reducing uncertainty changes action rather than measuring uncertainty itself. Against Experimental Design, it supplies an objective for selecting or stopping experiments, while design specifies how evidence is produced credibly. Against Opportunity Cost, it includes the foregone value of acting immediately but adds state-contingent learning. Against Prediction Error, it values evidence before observation rather than measuring a predictor afterward. Against Prioritization, it can generate a ranking of questions, but its ranking criterion is specifically expected decision improvement net of evidence cost.

Solution Archetypes¶

No catalogued solution archetypes reference this prime yet.

Notes¶

(New prime surfaced directly as the structural parent of Riskiest Assumption Test; queued for Claude house-style re-authoring, decision-analysis sourcing, and final formal review.)

References¶

(Citation set to be normalized during Claude re-authoring.)