Optimize org health intervention portfolio

Select an anti-Goodhart intervention portfolio using conservative causal lower bounds on real operating loss, never score movement, with design/transport/fidelity shrinkage, negative controls, interference, shared-loss, equity, resource and CVaR constraints.

Causal inference & experiment design Optimize CTOs Tech CEOs IT managers Team leads Investors

What it's for

Gives Gitrevio a rare anti-Goodhart advantage: it recommends only interventions with conservative causal evidence on business outcomes, even when the visible health score moves the other way.

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
beam_width integer ≥ 1 Numerical control Optional
budget number ≥ 0 Your calibration Yes
capacity_hours number ≥ 0 Your calibration Yes
causal_confidence number > 0.5, < 1 Your calibration Optional
detail_limit integer ≥ 1, ≤ 1000 Your calibration Optional
exact_search_limit integer ≥ 1 Your calibration Optional
intervention_options array of objects (23 fields) Evidence Yes
maximum_expected_loss number ≥ 0 Your calibration Optional
maximum_loss_cvar number ≥ 0 Your calibration Optional
minimum_group_loss_reduction_fraction number ≥ 0, ≤ 1 Your calibration Optional
minimum_units_benefiting_fraction number ≥ 0, ≤ 1 Your calibration Optional
operating_units array of objects (4 fields) Evidence Yes
required_control_ids array of string Evidence Yes
risk_aversion number ≥ 0 Your calibration Optional
risk_quantile number ≥ 0.5, < 1 Your calibration Optional
scenarios array of objects (3 fields) Evidence Yes

Each intervention_options record

Field Type Required
capacity_hours number (≥ 0) Yes
control_ids array of string Yes
cost number (≥ 0) Yes
dependency_option_ids array of string Yes
design_reliability number (≥ 0, ≤ 1) Yes
effect_standard_error number (≥ 0) Yes
evidence_verified boolean Optional
excluded_option_ids array of string Yes
id string (non-empty) Yes
identification_verified boolean Optional
implementation_fidelity number (≥ 0, ≤ 1) Yes
interference_handled boolean Optional
interference_scope string (non-empty) Optional
is_baseline boolean Yes
negative_control_passed boolean Optional
org_health_score_delta number Yes
outcome_loss_reduction_fraction number (≥ -1, ≤ 1) Yes
prospectively_registered boolean Optional
scenario_effect_multipliers array of number Yes
shared_loss_effect_ratio number (≥ 0, ≤ 1) Yes
study_design one of "randomized", "cluster_randomized", "regression_discontinuity", "difference_in_differences", "propensity_known" Optional
transport_weight number (≥ 0, ≤ 1) Yes
unit_id string (non-empty) Yes
Example input 
{
  "budget": 300,
  "capacity_hours": 30,
  "intervention_options": [
    {
      "capacity_hours": 0,
      "control_ids": [],
      "cost": 0,
      "dependency_option_ids": [],
      "design_reliability": 1,
      "effect_standard_error": 0,
      "excluded_option_ids": [],
      "id": "baseline-0",
      "implementation_fidelity": 1,
      "is_baseline": true,
      "org_health_score_delta": 0,
      "outcome_loss_reduction_fraction": 0,
      "scenario_effect_multipliers": [
        1,
        1
      ],
      "shared_loss_effect_ratio": 0,
      "transport_weight": 1,
      "unit_id": "unit-0"
    },
    {
      "capacity_hours": 10,
      "control_ids": [
        "change-control"
      ],
      "cost": 100,
      "dependency_option_ids": [],
      "design_reliability": 0.9,
      "effect_standard_error": 0.03,
      "evidence_verified": true,
      "excluded_option_ids": [],
      "id": "action-0",
      "identification_verified": true,
      "implementation_fidelity": 0.9,
      "interference_handled": false,
      "interference_scope": "none",
      "is_baseline": false,
      "negative_control_passed": true,
      "org_health_score_delta": 4,

Truncated for display — the full payload is 149 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 
{
  "decision": "org_health_intervention_portfolio_selected",
  "finding": "outcome_backed_actions_available",
  "limitations": [
    "The objective uses real operating loss, full cost, and tail risk; the composite score delta is diagnostic only.",
    "Causal effects are conservative lower bounds shrunk for design, transport, and implementation reliability.",
    "The output proposes aggregate actions for human approval and cannot rank people or authorize employment actions."
  ],
  "method": "causal_lower_bound_anti_goodhart_stochastic_portfolio_with_shared_loss_and_cvar",
  "pareto_frontier": [
    {
      "benefiting_unit_fraction": 0,
      "capacity_hours": 0,
      "conditional_value_at_risk": 6600,
      "cost": 0,
      "diagnostic_org_health_score_delta_not_optimized": 0,
      "expected_operating_loss": 3600,
      "expected_total_loss": 3600,
      "group_loss_reduction_fraction": {
        "platform": 0,
        "product": 0
      },
      "minimum_group_loss_reduction_fraction": 0,
      "risk_quantile": 0.95,
      "selected_option_ids": [
        "baseline-0",
        "baseline-1"
      ],
      "value_at_risk": 6600
    },
    {
      "benefiting_unit_fraction": 0.5,
      "capacity_hours": 10,
      "conditional_value_at_risk": 5902.69,
      "cost": 100,
      "diagnostic_org_health_score_delta_not_optimized": 4,
      "expected_operating_loss": 3176.97,
      "expected_total_loss": 3276.97,
      "group_loss_reduction_fraction": {
        "platform": 0,
        "product": 0.2226
      },
      "minimum_group_loss_reduction_fraction": 0,
      "risk_quantile": 0.95,

Truncated for display — the full payload is 104 lines.

How it works

Causal inference & experiment design — Separate what a change caused from what merely moved alongside it.

  1. 1 Admit non-baseline actions only when a prospective randomized or identified quasi-experiment, negative control, interference treatment, required controls and verified evidence support a positive lower-confidence reduction in real operating loss.
  2. 2 Shrink that causal lower bound for design reliability, transport and implementation fidelity, retain full cost, and explicitly reject proxy-only options whose health score rises without supported outcome benefit.
  3. 3 Enumerate or deterministically beam-search one option per aggregate unit under common scenarios, count shared loss once, enforce dependencies, exclusions, mandatory actions, budget, capacity, group-benefit, expected-loss and CVaR gates, and expose the cost-loss-tail Pareto set.

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

  • Assignment or adoption timing, interference policy, overlap, attrition, and outcome availability match the estimand encoded by the method.
  • Operating units, outcome-loss baselines, unique shared-loss groups and scenarios are complete; causal studies identify the deployed version; negative controls and interference checks are credible; finance owns full cost and loss.
  • A causal label is warranted only when design and diagnostic requirements pass; otherwise treat the result as descriptive or abstain.
  • This proposes reversible aggregate operating interventions for accountable approval. It never instructs agents to game the score, change production automatically, rank people or make employment decisions.

Minimum evidence

  • operating_units: required and organization-defined
  • intervention_options: required and organization-defined
  • scenarios: required and organization-defined
  • required_control_ids: required and organization-defined
  • budget: required and organization-defined
  • capacity_hours: required and organization-defined

How to validate it

Preserve the assignment/adoption design and validate overlap, pre-period or placebo diagnostics, attrition, interference policy, and cluster-level uncertainty; never tune on the estimated effect.

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

  • versioned unit-option-scenario matrix joined to validated operating-loss forecasts, unique shared-loss groups, causal-study registry, conservative lower-bound effects and immutable resource/approval state
  • eligible aggregate actions, causal design/identification, confidence, negative-control and interference policy, required controls, transport/fidelity, dependencies/exclusions, mandatory coverage, group benefit, budget/capacity, expected loss, CVaR, human authority and explicit prohibition on optimizing the score

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": "select an antigoodhart intervention portfolio using" }
  → finds "optimize_org_health_intervention_portfolio"

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

gitrevio_capability_run
  { "capability_id": "optimize_org_health_intervention_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

Audit org health score integrity

Reconstruct the organization-health composite from frozen component evidence and block publication when source completeness, consent, versioning, construct balance, cross-group measurement invariance, redundancy, privacy or leave-one-component stability fails.

Statistical audit & measurement

Forecast org health operating loss

Prove whether the company-local organization-health score leads later delivery, reliability and capacity losses, then simulate their correlated economic tail only after a latest-whole-period baseline challenge passes.

Forecasting & survival

Analyze info gap robust satisficing

Select a robust-satisficing action under severe uncertainty with Info-Gap Decision Theory: evaluate worst and best payoff across a governed nested uncertainty envelope, maximize the radius before a critical requirement fails, report windfall opportuneness, and use no scenario probabilities.

Decision analysis

Audit attention fragmentation evidence integrity

Audit consented point-in-time contributor identity, availability, privacy-safe calendar metadata and work-session lineage before reporting aggregate meeting load, protected focus blocks or cross-project switching.

Statistical audit & measurement

Audit attrition risk prediction integrity

Audit an attrition model's complete eligible cohort, point-in-time features, supportive-use governance, intervention-contaminated labels, competing outcomes, calibration, false positives and authorized subgroup error before any person-level use.

Forecasting & survival

Audit code knowledge concentration integrity

Audit file, module, service, or repository knowledge concentration from point-in-time substantive changes, reviews, incident response and documentation using identity-confidence filtering, recency decay, Bayesian ownership uncertainty, entropy-effective owners, HHI and leave-top-owner-out resilience—without turning contribution evidence into a person-performance score.

Sequential Bayesian & bandits

See every tool in People, retention & knowledge →

Ready to See Your Engineering work clearly?

Request a free demo