Audit human AI decision complementarity

What it's for

Tests whether human plus AI genuinely beats the better of the two alone on your decisions — the claim most AI adoption assumes and few organizations check.

Proves whether human judgment and AI work better together on the company's real decisions, instead of treating human involvement or automation as inherently valuable.

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.

{
  "bootstrap_draws": 300,
  "confidence_level": 0.8,
  "observations": [
    {
      "ai_action": "approve",
      "ai_loss": 10,
      "cluster_id": "team-0",
      "combined_loss": 3,
      "decision_period": 0,
      "human_action": "review",
      "human_loss": 8,
      "id": "human-ai-shadow-000",
      "system_id": "delivery-risk-triage",
      "weight": 1
    },
    {
      "ai_action": "review",
      "ai_loss": 11,
      "cluster_id": "team-0",
      "combined_loss": 4,
      "decision_period": 1,
      "human_action": "approve",
      "human_loss": 9,
      "id": "human-ai-shadow-001",
      "system_id": "delivery-risk-triage",
      "weight": 1
    },
    {
      "ai_action": "approve",
      "ai_loss": 12,
      "cluster_id": "team-0",
      "combined_loss": 5,
      "decision_period": 2,
      "human_action": "approve",
      "human_loss": 10,
      "id": "human-ai-shadow-002",
      "system_id": "delivery-risk-triage",
      "weight": 1
    },
    {
      "ai_action": "review",
      "ai_loss": 10,
      "cluster_id": "team-0",

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.

{
  "configuration": {
    "bonferroni_one_sided_tail_probability": 0.2,
    "bootstrap_draws": 300,
    "confidence_level": 0.8,
    "maximum_combined_mean_loss": null,
    "minimum_cluster_count": 5,
    "minimum_disagreement_count": 10,
    "minimum_mean_loss_improvement": 0,
    "minimum_observation_count": 40,
    "seed": 19,
    "simultaneous_gate_count": 1
  },
  "decision": "validated_human_ai_systems_available",
  "guardrails": [
    "All three decisions must be prospectively frozen for the same eligible cases and scored against one mature outcome under one loss definition. Human-alone review must be blind to the AI output; combined review is a distinct governed policy.",
    "Complementarity means the combined policy beats the better standalone policy out of cluster, not that either participant is globally superior. Disagreement is necessary evidence, not value by itself.",
    "Cluster bootstrap and simultaneous gates do not repair selective shadow review, outcome leakage, shared incentives or changed operating regimes. Results govern decision systems and must never become employee rankings."
  ],
  "method": "paired_cluster_bootstrap_human_ai_complementarity_v1",
  "summary": {
    "supported_system_count": 1,
    "supported_system_ids": [
      "delivery-risk-triage"
    ],
    "system_count": 1
  },
  "system_diagnostics": [
    {
      "ai_mean_loss": 11,
      "cluster_count": 6,
      "combined_mean_loss": 4,
      "combined_mean_loss_upper_bound": 4.0333,
      "decision": "human_ai_complementarity_supported",
      "disagreement_case_improvement_vs_selected_standalone": 5,
      "disagreement_count": 30,
      "failed_gates": [],
      "human_mean_loss": 9,
      "mean_loss_improvement_lower_bound": 5,
      "mean_loss_improvement_vs_best_standalone": 5,
      "observation_count": 60,
      "system_id": "delivery-risk-triage",
      "weighted_disagreement_rate": 0.5
    }

How it works

Constrained optimization — Pick the best feasible option under real limits — budget, headcount, dependencies, capacity — rather than ranking a list and hoping it fits.

1. 1 Freeze eligible cases, blind human-alone decisions, AI-alone decisions and the distinct combined policy before one mature outcome is known; score all three with one locally governed loss.
2. 2 Within each system, compare weighted mean loss with the better standalone policy, measure decision disagreement and disagreement-case improvement, and resample whole operational clusters to preserve dependence.
3. 3 Bonferroni-adjust one-sided bootstrap gates across every screened system and optional absolute-loss gate; support complementarity only with sufficient cases, clusters, disagreements and conservative loss improvement.

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": "audit whether a governed humanai decision" }
  → finds "audit_human_ai_decision_complementarity"

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

gitrevio_capability_run
  { "capability_id": "audit_human_ai_decision_complementarity", "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 AI cost, routing & return →