Solve budgeted bayesian experiment portfolio

Choose a budget- and resource-feasible portfolio of Bayesian experiments whose correlated observations can change multiple governed deployment decisions.

Sequential Bayesian & bandits Optimize CTOs Tech CEOs IT managers Team leads Investors

What it's for

Finds the experiment portfolio that buys the most decision-changing evidence—recognizing that one well-chosen test can inform several correlated roadmap bets.

What you give it

Inputs split into evidence read from your connected systems, calibration your team owns, and numerical controls that affect precision but never the result's meaning.

Field Type Role Required
budget number ≥ 0 Your calibration Yes
control_allocation_fraction number ≥ 0.05, ≤ 0.95 Your calibration Optional
correlations array of objects (4 fields) Evidence Yes
decisions array of objects (8 fields) ≥ 1 item Evidence Yes
experiment_options array of objects (6 fields) ≥ 1 item Evidence Yes
max_detail_rows integer ≥ 1, ≤ 500 Numerical control Optional
maximum_exact_states integer ≥ 2, ≤ 4194304 Numerical control Optional
maximum_harm_probability number ≥ 0.001, ≤ 0.5 Your calibration Optional
minimum_adoption_probability number ≥ 0.5, ≤ 0.999 Your calibration Optional
minimum_portfolio_net_value number ≥ 0 Your calibration Optional
minimum_positive_portfolio_probability number ≥ 0, ≤ 1 Your calibration Optional
resources array of objects (2 fields) Evidence Yes
risk_aversion number ≥ 0 Your calibration Optional
seed integer ≥ 0, ≤ 4294967295 Numerical control Optional
simulation_confidence_level number ≥ 0.8, ≤ 0.999 Your calibration Optional
simulation_draws integer ≥ 1000, ≤ 200000 Numerical control Optional
tail_probability number ≥ 0.001, ≤ 0.5 Your calibration Optional

Each decisions record

Field Type Required
affected_opportunities integer (≥ 1, ≤ 1000000000) Yes
harm_effect_threshold number Optional
id string (non-empty) Yes
implementation_cost number (≥ 0) Yes
outcome_standard_deviation number (> 0) Yes
prior_effect_mean number Yes
prior_effect_standard_deviation number (> 0) Yes
value_per_effect_unit number (> 0) Yes
Example input 
{
  "budget": 10000,
  "correlations": [
    {
      "correlation": 0.8,
      "decision_a": "pricing",
      "decision_b": "onboarding",
      "id": "shared-demand-response"
    }
  ],
  "decisions": [
    {
      "affected_opportunities": 100000,
      "harm_effect_threshold": 0,
      "id": "pricing",
      "implementation_cost": 10000,
      "outcome_standard_deviation": 1,
      "prior_effect_mean": 0.01,
      "prior_effect_standard_deviation": 0.05,
      "value_per_effect_unit": 100
    },
    {
      "affected_opportunities": 100000,
      "harm_effect_threshold": 0,
      "id": "onboarding",
      "implementation_cost": 10000,
      "outcome_standard_deviation": 1,
      "prior_effect_mean": 0.01,
      "prior_effect_standard_deviation": 0.05,
      "value_per_effect_unit": 100
    }
  ],
  "experiment_options": [
    {
      "decision_id": "pricing",
      "fixed_experiment_cost": 0,
      "id": "test-pricing",
      "resource_requirements": {
        "analyst": 1
      },
      "sample_cost_per_observation": 10,
      "sample_size": 1000
    },
    {

Truncated for display — the full payload is 63 lines.

What you get back

This is the actual output of running the example above — computed by the same function the platform calls, not an illustration.

Example output 
{
  "assumptions": [
    "Decision effects have the submitted multivariate-normal pre-experiment prior; each two-arm estimate is unbiased with known effective outcome variance, independent observation noise, declared allocation, and no unmodeled interference, attrition, sequential peeking, or design failure.",
    "Selected observations update every correlated decision through the joint Gaussian posterior, so cross-decision learning is model-conditional and requires correlations estimated from transportable pre-decision evidence rather than current outcomes or another tenant.",
    "Deployment value is linear over governed opportunities, implementation and experiment costs are complete and commensurable, resources and budget are genuinely shared, and at most one mutually exclusive sample-size option may target each decision.",
    "A recommendation requires a positive Monte Carlo confidence lower bound; heuristic mode has no global certificate, EVPI is an upper bound, and neither result replaces prospective causal-design, safety, ethics, or accountable approval."
  ],
  "configuration": {
    "budget": 10000,
    "budget_remaining": 0,
    "control_allocation_fraction": 0.5,
    "maximum_harm_probability": 0.1,
    "minimum_adoption_probability": 0.8,
    "minimum_portfolio_net_value": 0,
    "minimum_positive_portfolio_probability": 0,
    "risk_aversion": 0,
    "seed": 7,
    "simulation_confidence_level": 0.95,
    "simulation_draws": 5000,
    "tail_probability": 0.1
  },
  "decision": "precision_supported_experiment_portfolio_selected",
  "decision_diagnostics": [
    {
      "baseline_without_experiments_adopts": false,
      "decision_id": "pricing",
      "posterior_effect_standard_deviation": 0.03922323,
      "posterior_variance_reduction_fraction": 0.3846,
      "prior_effect_standard_deviation": 0.05,
      "probability_policy_adopts_after_portfolio": 0.096,
      "selected_experiment_option_id": "test-pricing"
    },
    {
      "baseline_without_experiments_adopts": false,
      "decision_id": "onboarding",
      "posterior_effect_standard_deviation": 0.04341216,
      "posterior_variance_reduction_fraction": 0.2462,
      "prior_effect_standard_deviation": 0.05,
      "probability_policy_adopts_after_portfolio": 0.0368,
      "selected_experiment_option_id": null
    }
  ],
  "experiment_option_diagnostics": [
    {

Truncated for display — the full payload is 103 lines.

How it works

Sequential Bayesian & bandits — Learn while deciding — update beliefs as evidence arrives and choose where the next unit of effort is worth spending.

  1. 1 Freeze up to twenty deployment decisions with a transportable joint normal prior, effective outcome variances, economics and harm thresholds, plus mutually exclusive sample-size options, shared resources, and full experiment costs.
  2. 2 For every exact candidate portfolio—or disclosed coordinate-search candidate—simulate correlated true effects and observations, update all decisions through the joint Gaussian posterior, and apply economic and harm posterior gates.
  3. 3 Subtract experiment cost, quantify EVSI, Monte Carlo error, positive-value probability, CVaR and cross-decision variance reduction, then select only a portfolio clearing precision, value, probability, budget, and resource gates.

Before you trust it

Every tool in the catalog ships with the conditions under which its answer is meaningful — and the conditions under which it should abstain instead of guessing.

Assumptions & guardrails

  • The likelihood or reward model, prior support, action logging, delayed outcomes, and any stationarity assumptions match the deployment process.
  • The prior covariance is positive semidefinite and pre-data, correlations transport across decisions, estimates are unbiased with effective independent noise, deployment value is linear, costs and constraints are complete, and prospective causal designs are separately valid.
  • Posterior probability and adaptive selection are model-conditional; they are not substitutes for randomized propensities or guaranteed safety.
  • Cross-decision learning is model-conditional; positive EVSI cannot repair confounding or invalid experiments, heuristic search has no global certificate, and an unpriced harm threshold must not be traded against financial value.

Minimum evidence

  • decisions: at least 1 rows/items
  • experiment_options: at least 1 rows/items
  • correlations: required and organization-defined
  • resources: required and organization-defined
  • budget: required and organization-defined

How to validate it

Backtest the chosen action against simple feasible baselines on held-out scenarios, sweep costs/constraints/risk tolerance, and require constraint feasibility under adverse inputs.

Calibrating it to your org

Same for everyone

The mathematical kernel, validation rules, method version, and JSON output semantics are organization-independent; no tenant-trained coefficients or company benchmark is embedded in the function.

Specific to you

  • transport-checked joint pre-experiment effect prior, positive-semidefinite covariance, and effective estimator variance after clustering, repeated measures, attrition, and design effects
  • causal design, prior and correlation provenance, population and horizon transport, complete experiment and implementation economics, shared constraints, posterior adoption and harm gates, tail risk, minimum value and probability, simulation precision, and seed

Calibration workflow

  1. 1 Define the management decision, target outcome, aggregate unit, privacy boundary, cadence, and prediction/intervention horizon for this organization.
  2. 2 Build a tenant-scoped historical cohort using only information available before each prediction or decision; preserve zero periods, censoring, assignment probabilities, and unresolved outcomes when the method requires them.
  3. 3 Estimate statistical parameters on training history, but obtain costs, utilities, risk tolerance, practical-effect thresholds, capacity, and policy constraints from accountable decision owners.
  4. 4 Validate on later time windows or held-out aggregate units at the deployment grain, against a simple baseline and the function-specific validation strategy.
  5. 5 Deploy only if the returned decision clears evidence, overlap, calibration, robustness, and guardrail checks; warning, unsupported, schema-gap, and fallback decisions are abstentions.
  6. 6 Monitor realized outcomes, data drift, coverage, and decision regret; recalibrate at a governed cadence or after a detected regime/definition change, never merely because a stakeholder dislikes the result.

Call it from your AI

You don't wire up 388 tools in your MCP client. The GitRevio MCP server exposes 18 tools, three of which let an agent search the catalog, read a tool's schema, and run it — so the assistant finds this one on its own. 

gitrevio_capabilities_search
  { "q": "choose a budget and resourcefeasible portfolio" }
  → finds "solve_budgeted_bayesian_experiment_portfolio"

gitrevio_capability_describe
  { "capability_id": "solve_budgeted_bayesian_experiment_portfolio" }
  → returns the input schema and agent guidance shown on this page

gitrevio_capability_run
  { "capability_id": "solve_budgeted_bayesian_experiment_portfolio", "arguments": { ... } }
  → returns the result shown above

Works in Claude Desktop, Claude Code, Cursor, Cline, Continue.dev, Goose and Aider. See the MCP server.

Related tools

Estimate value of flexibility

Value a strategy's pre-action signal-contingent flexibility against its best fixed action, including enablement cost, tail loss, and perfect-information headroom.

Statistical audit & measurement

Reconcile plan actual variance drivers

Reconcile plan-to-actual value variance with an exact, order-independent Shapley decomposition of a declared multilinear operating model.

Decision analysis

Audit causal claim negative controls

Gate a causal effect claim using prespecified negative outcome/exposure controls, Benjamini-Hochberg multiplicity control, and an omnibus chi-square falsification test.

Causal inference & experiment design

Audit cluster randomization integrity

Audit cluster-randomized experiments for practical baseline imbalance and differential outcome observation, with cluster-size-weighted standardized differences and assignment permutation diagnostics.

Causal inference & experiment design

Audit growth incrementality experiment integrity

Audit aggregate randomized growth experiments before anyone trusts channel incrementality: enforce unique experimental units, nondegenerate logged propensities, both arms, control-spend discipline, baseline balance, spillover and evidence gates; then estimate propensity-weighted baseline-adjusted contribution, cluster-unit bootstrap uncertainty and incremental return on spend.

Causal inference & experiment design

Audit joint outcome network integrity

Audit whether a company-specific Bayesian joint-outcome network is fit for reliance by validating point-in-time lineage, DAG and CPT completeness, effective support, protected-attribute exclusions, and strictly out-of-time outcome calibration against a baseline.

Sequential Bayesian & bandits

See every tool in Causal evidence & experiments →

Ready to See Your Engineering work clearly?

Request a free demo