Risk-sensitive constrained VOI

Issue #570’s experimental v1 surface answers a deliberately narrow question: how much does perfect state information improve a decision when the same risk functional and operational constraints govern both the current and informed policy problems?

import json
from pathlib import Path

from voiage import risk_sensitive_constrained_voi

request = json.loads(
    Path("specs/frontier/risk-sensitive-constrained-voi/v1/fixtures/normative/input.json")
    .read_text()
)
result = risk_sensitive_constrained_voi(request).to_contract_dict()
print(result["value"])

The objective can be expected value, already-declared utility, lower-tail CVaR, or minimax regret. A utility transform or regret reference must be fixed before calling the evaluator; VOIAGE does not infer risk preferences from raw outcomes. Constraints declare their unit, direction, limit, enforcement, and provenance. Chance constraints apply to the probability mass on which the selected policy mapping satisfies the limit.

voiage --format json calculate-risk-sensitive-voi \
  specs/frontier/risk-sensitive-constrained-voi/v1/fixtures/normative/input.json

The evaluator exactly enumerates bounded finite mappings. Its constraint-removal results are global discrete diagnostics, not local shadow prices. Continuous optimization, sample information, stable maturity, and cross-language parity remain future gates.