Expected-utility information pricing

Expected-utility information pricing values evidence for a declared decision-maker utility, wealth basis, stakeholder scope, and information structure. It is an experimental, Rust-authoritative frontier method. Its contracts may change before stable promotion.

Use this family when risk attitude is part of the decision model. It does not replace monetary EVPI or EVSI for risk-neutral net-benefit models, and it does not make different stakeholders or decision problems numerically comparable.

Measures and units

Let EU0 be optimized expected utility without the information and EU1 the expected utility under the signal-contingent informed policy.

Measure	Interpretation	Unit
EUI	EU1 - EU0	Declared utility scale
CEI	Difference between the informed and current certainty equivalents	Payoff unit
BPI	Largest ex-ante sure transfer paid to acquire information while preserving current expected utility	Payoff unit
SPI	Smallest ex-ante sure transfer received to surrender information while preserving informed expected utility	Payoff unit
PPI	EUI divided by the informed-utility gain above a declared terminal-outcome floor	Dimensionless

EUI values depend on utility normalization. CEI, BPI, and SPI are expressed in the declared payoff unit, but they need not be equal under nonlinear utility. PPI is available only when the declared floor is in the utility domain and creates a positive denominator. Consult the result’s pairwise comparability matrix before ranking alternatives or comparing analyses.

Required decision contract

Every request explicitly declares:

• action and state identifiers, terminal payoff matrix, and state probabilities;
• initial wealth, payoff unit, three-letter currency, price date, and stakeholder scope;
• an affine, exponential, logarithmic, or power utility descriptor;
• either clairvoyance or a finite signal through joint signal-state probabilities;
• the ex_ante_sure_transfer information-cost location; and
• bounded root-solver tolerances and evaluation limits.

The implementation optimizes both current and signal-conditional policies. For bounded-domain utilities, infeasible action/state combinations appear as structured domain_exclusions; they are not silently represented as infinite or missing payoffs.

Root assurance

BPI and SPI are indifference-price roots. Read their complete bpi_root and spi_root records before using an estimate. A usable root reports its status, bracket, final width, residual, iteration and evaluation counts, evaluated policies, and policy transitions. Failure states such as utility_domain, not_bracketed, discontinuous_no_root, and exhausted iteration or evaluation limits do not fabricate a price.

Value of Clairvoyance presentation

Value of Clairvoyance (VoC) is a presentation of this family’s clairvoyant policy result. It is not a separate numerical method or kernel. The Python helper value_of_clairvoyance(...) therefore requires information.kind="clairvoyant" and returns the canonical result with a voc presentation label.

The optional monetary evpi display alias is available only when the request uses a verified positive-affine utility. For exponential, logarithmic, or power utility, request EUI, CEI, BPI, SPI, or PPI instead; the implementation fails closed with affine_reduction_required rather than relabelling a nonlinear result as EVPI. A finite-signal positive-affine reduction is EVSI, not VoC or EVPI.

Python and CLI

from voiage.methods.utility_information import (
    expected_utility_information_value,
    value_of_clairvoyance,
)

result = expected_utility_information_value(request)
voc = value_of_clairvoyance(request, selected_measure="bpi")

The same versioned request object can be supplied as JSON:

voiage calculate-expected-utility-information request.json --measure bpi
voiage calculate-expected-utility-information request.json \
  --presentation voc --measure eui

There is deliberately no calculate-voc command. See the accessible worked example and the frontier contract for a runnable request and the complete result schema.

Maturity and language boundary

Rust and Python execute this experimental family. R and Julia do not expose it, and Mojo remains an external interoperability boundary. Stable promotion requires scientific review, complete assurance evidence, hosted checks, and a separate approval; documentation and deterministic fixtures do not establish those outcomes.