Optimize calibration experiment portfolio

What it's for

Turns analytics onboarding into a capital-allocation problem: Gitrevio learns which function is worth validating next and retains budget when calibration cannot change a decision enough.

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.

{
  "candidates": [
    {
      "activation_cost": 10,
      "dependency_ids": [],
      "exclusion_ids": [],
      "experiment_capacity_units": 1,
      "experiment_cost": 5,
      "experiment_trial_count": 4,
      "failure_loss_scenarios": [
        80,
        120
      ],
      "id": "completion-forecast",
      "prior_success_alpha": 1,
      "prior_success_beta": 1,
      "success_value_scenarios": [
        100,
        140
      ]
    },
    {
      "activation_cost": 10,
      "dependency_ids": [],
      "exclusion_ids": [],
      "experiment_capacity_units": 1,
      "experiment_cost": 8,
      "experiment_trial_count": 4,
      "failure_loss_scenarios": [
        100,
        140
      ],
      "id": "policy-value",
      "prior_success_alpha": 2,
      "prior_success_beta": 2,
      "success_value_scenarios": [
        80,
        100
      ]
    }
  ],
  "experiment_budget": 10,
  "experiment_capacity_units": 1,
  "scenarios": [

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.

{
  "candidate_diagnostics": [
    {
      "expected_value_of_sample_information": 20.4,
      "function_id": "completion-forecast",
      "minimum_success_count_to_activate": 3,
      "net_expected_calibration_value": 15.4,
      "prior_success_probability": 0.5,
      "selected": true
    },
    {
      "expected_value_of_sample_information": 4.1857,
      "function_id": "policy-value",
      "minimum_success_count_to_activate": 3,
      "net_expected_calibration_value": -3.8143,
      "prior_success_probability": 0.5,
      "selected": false
    }
  ],
  "decision": "run_selected_calibration_experiments",
  "failed_gates": [],
  "guardrails": [
    "Beta priors and calibration-trial outcomes must refer to repeatable, exchangeable success events at the production decision grain; a convenient pass/fail label or another tenant's history is not a valid prior.",
    "EVSI values only the activation decision represented by submitted success value, failure loss and cost; omitted harms, correlated calibration results, shared setup costs and future sequential experiments require an expanded model.",
    "The false-activation loss gate is an expected aggregate risk budget, not a tail guarantee; safety-critical activation still requires domain-specific validation, monitoring, rollback and human approval.",
    "Heuristic mode has no global certificate, and selecting an experiment neither activates a function nor authorizes experimentation on people, customers or production systems."
  ],
  "method": "beta_binomial_calibration_evsi_constrained_portfolio_v1",
  "selected_experiments": [
    {
      "calibration_policy": [
        {
          "action": "do_not_activate",
          "observed_success_count": 0,
          "posterior_expected_activation_value": -68,
          "posterior_success_probability": 0.1667,
          "predictive_probability": 0.2
        },
        {
          "action": "do_not_activate",
          "observed_success_count": 1,
          "posterior_expected_activation_value": -34,
          "posterior_success_probability": 0.3333,
          "predictive_probability": 0.2

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 For each candidate function, freeze a tenant-local Beta prior over repeatable production success, calibration trial count, coherent success value/failure loss scenarios, activation and experiment cost, capacity, dependencies and exclusions.
2. 2 Integrate the exact Beta-binomial result distribution, update the posterior for every possible success count, choose activate-or-not under the governed posterior threshold and economics, and calculate EVSI plus false-activation risk.
3. 3 Select a dependency-closed, exclusion-safe experiment portfolio under cost, capacity and false-activation-loss limits using exact subset enumeration inside the declared boundary or a deterministic disclosed heuristic beyond it.

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.

Calibrating it to your org

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 which analytical functions to calibrate" }
  → finds "optimize_calibration_experiment_portfolio"

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

gitrevio_capability_run
  { "capability_id": "optimize_calibration_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

See every tool in Analytics assurance & orchestration →