Audit multivariate metric drift

What it's for

Detects when the data underneath a dashboard has shifted enough to invalidate comparisons — across many metrics at once, with false-discovery control.

Audits whether engineering metrics changed distribution or data regime before an agent treats a shifted mean as an operational fact.

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.

{
  "current_rows": [
    {
      "id": "current-0",
      "metrics": {
        "cycle_time": 0.8,
        "failure_rate": 0,
        "review_time": 0
      }
    },
    {
      "id": "current-1",
      "metrics": {
        "cycle_time": 0.9,
        "failure_rate": 0.01,
        "review_time": 0.1
      }
    },
    {
      "id": "current-2",
      "metrics": {
        "cycle_time": 1,
        "failure_rate": 0.02,
        "review_time": 0.2
      }
    },
    {
      "id": "current-3",
      "metrics": {
        "cycle_time": 1.1,
        "failure_rate": 0.03,
        "review_time": 0.3
      }
    },
    {
      "id": "current-4",
      "metrics": {
        "cycle_time": 1.2000000000000002,
        "failure_rate": 0.04,
        "review_time": 0.4
      }
    },
    {
      "id": "current-5",

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.

{
  "assumptions": [
    "Reference and current rows are independent representative samples from comparable entity definitions and measurement pipelines.",
    "Reference quantile bins are fixed before current observations are inspected; changing bins after drift appears invalidates the permutation interpretation.",
    "Benjamini-Hochberg controls multiplicity across the declared metrics while the PSI floor requires practical as well as statistical drift.",
    "Distribution drift is a data or operating-regime diagnostic, not evidence of causality, misconduct, or individual performance change."
  ],
  "decision": "material_distribution_drift_detected",
  "drifted_metrics": [
    "cycle_time"
  ],
  "method": "fixed_reference_psi_js_permutation_fdr_drift_v1",
  "metrics": [
    {
      "current_mean": 1.75,
      "fdr_discovery": true,
      "jensen_shannon_divergence": 0.17443,
      "material_drift": true,
      "metric": "cycle_time",
      "permutation_p_value": 0.004975,
      "population_stability_index": 1.758141,
      "reference_bin_count": 10,
      "reference_mean": 0.95,
      "standardized_wasserstein_distance": 1.136
    },
    {
      "current_mean": 0.0495,
      "fdr_discovery": false,
      "jensen_shannon_divergence": 0,
      "material_drift": false,
      "metric": "failure_rate",
      "permutation_p_value": 1,
      "population_stability_index": 0,
      "reference_bin_count": 10,
      "reference_mean": 0.0495,
      "standardized_wasserstein_distance": 0
    },
    {
      "current_mean": 0.785,
      "fdr_discovery": false,
      "jensen_shannon_divergence": 0,
      "material_drift": false,
      "metric": "review_time",
      "permutation_p_value": 1,

How it works

Statistical audit & measurement — Check whether a number is fit to decide on: coverage, timing, reconciliation, and the gaps a dashboard hides.

1. 1 Detect material distribution shifts with reference-fixed quantile bins, PSI, Jensen-Shannon divergence, standardized Wasserstein distance, permutation tests, and FDR control.
2. 2 Evaluate the method-specific diagnostics and gates returned by the function, then abstain unless the declared decision clears them under locally governed thresholds.

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": "detect material distribution shifts with referencefixed" }
  → finds "audit_multivariate_metric_drift"

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

gitrevio_capability_run
  { "capability_id": "audit_multivariate_metric_drift", "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 Measurement integrity →