ICM · Research & Decision Studio · decision-lab · Model 1.1.0 · Illustrative inputs
Decision science
Decision lab
Discover whether an imperfect test changes the preferred action enough to pay for itself.
Decision / uncertaintyIllustrative composition
The interactive model requires JavaScript. This saved default illustration remains readable if the runtime is unavailable. Values are computed from the canonical model at build time; they describe fictional inputs, with full calculation precision.
ICM · Decision lab · Model 1.1.0 · Default illustrative record
Default illustrative result
Stage the commitment has the highest expected payoff before new evidence. The supplied test adds net expected value after its cost. The information-value ceiling is 178.25 USDk for perfect state knowledge.
| Metric | Value | Unit | Calculation boundary |
|---|---|---|---|
| Best value before information | 55.75 | USDk | |
| Gross value of this test | 94.99 | USDk | Maximum justified test price under these payoffs. |
| Net information value | 76.99 | USDk | Gross information value less test cost. |
| Value of perfect information | 178.25 | USDk | Upper bound for any information about these two states. |
| Action | Success payoffUSDk | Failure payoffUSDk | Expected payoffUSDk | Success regretUSDk | Failure regretUSDk | Expected regretUSDk |
|---|---|---|---|---|---|---|
| Commit capital | 520 | -340 | 47 | 0 | 340 | 187 |
| Stage the commitment | 240 | -95 | 55.75 | 280 | 95 | 178.25 |
| Wait | 0 | 0 | 0 | 520 | 0 | 234 |
| Signal | Branch probability | Posterior success probability | Preferred action | Conditional expected payoffUSDk | Weighted contributionUSDk |
|---|---|---|---|---|---|
| Positive test | 0.49 | 0.753061224489796 | Commit capital | 307.6326530612245 | 150.74 |
| Negative test | 0.51 | 0.1588235294117647 | Wait | 0 | 0 |
| Prior success% | Preferred action | Best expected payoffUSDk |
|---|---|---|
| 0 | Wait | 0 |
| 5 | Wait | 0 |
| 10 | Wait | 0 |
| 15 | Wait | 0 |
| 20 | Wait | 0 |
| 25 | Wait | 0 |
| 30 | Stage the commitment | 5.5 |
| 35 | Stage the commitment | 22.25 |
| 40 | Stage the commitment | 39 |
| 45 | Stage the commitment | 55.75 |
| 50 | Commit capital | 90 |
| 55 | Commit capital | 133 |
| 60 | Commit capital | 176 |
| 65 | Commit capital | 219 |
| 70 | Commit capital | 262 |
| 75 | Commit capital | 305 |
| 80 | Commit capital | 348 |
| 85 | Commit capital | 391 |
| 90 | Commit capital | 434 |
| 95 | Commit capital | 477 |
| 100 | Commit capital | 520 |
| Action A | Action B | Prior at equal value% | Best of all actions at intersection |
|---|---|---|---|
| Commit capital | Stage the commitment | 46.666666666666664 | Commit capital |
| Commit capital | Wait | 39.53488372093023 | Stage the commitment |
| Stage the commitment | Wait | 28.35820895522388 | Stage the commitment |
Inputs behind this result
| Assumption and unit | Exact supplied value |
|---|---|
| Prior probability of success (%) | 45 |
| Test sensitivity (%) | 82 |
| Test specificity (%) | 78 |
| Information cost (USDk) | 18 |
| Action | Payoff if successUSDk | Payoff if failureUSDk |
|---|---|---|
| Commit capital | 520 | -340 |
| Stage the commitment | 240 | -95 |
| Wait | 0 | 0 |
Transparent method
- Expected value = prior × success payoff + (1 − prior) × failure payoff.
- Bayes updates the prior separately after a positive and negative test. Branches of probability zero have no posterior.
- EVSI = expected best payoff after the imperfect test − best payoff before it. EVPI is the analogous value of perfect state knowledge.
- Net information value = EVSI − test cost. Payoffs are assumed to exclude this test cost.
- Joint signal-state masses, action rankings, EVSI, EVPI and the net-value sign use exact rational arithmetic for canonical decimal inputs. Only an exact zero is labelled break-even.
- Tied action values use literal name order independent of browser language for consistent display; this ordering does not establish a strict payoff advantage.
Limits
- Binary states, a single test, user-calibrated probabilities and risk-neutral expected monetary value.
- No repeated sampling, behavioural utility, ambiguity aversion or causal proof.
- Charts and headline values are numeric displays of exact rational results. Decimal input values are assumptions; risk-neutral economic comparison does not establish acceptable action or calibrated probability.
- Numerical input precision: use zero or absolute values of at least 0.000000001. Smaller nonzero inputs are rejected explicitly.