Audit KPI threshold bunching

What it's for

Extends Gitrevio's Goodhart protection from proxy drift and incentive design into a live post-target diagnostic that exposes suspicious threshold mass without accusing people or pretending the pattern proves manipulation.

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.

{
  "bandwidth": 10,
  "bootstrap_draws": 500,
  "minimum_excess_mass_share": 0.1,
  "minimum_mirror_shift": 0.2,
  "observations": [
    {
      "epoch": "pre_target",
      "id": "kpi-0-pre_target-0",
      "metric_value": 85,
      "unit_id": "unit-0"
    },
    {
      "epoch": "pre_target",
      "id": "kpi-0-pre_target-1",
      "metric_value": 95,
      "unit_id": "unit-0"
    },
    {
      "epoch": "pre_target",
      "id": "kpi-0-pre_target-2",
      "metric_value": 105,
      "unit_id": "unit-0"
    },
    {
      "epoch": "pre_target",
      "id": "kpi-0-pre_target-3",
      "metric_value": 115,
      "unit_id": "unit-0"
    },
    {
      "epoch": "pre_target",
      "id": "kpi-0-pre_target-4",
      "metric_value": 85,
      "unit_id": "unit-0"
    },
    {
      "epoch": "pre_target",
      "id": "kpi-0-pre_target-5",
      "metric_value": 95,
      "unit_id": "unit-0"
    },
    {
      "epoch": "pre_target",

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": [
    "The threshold, activation epoch, bandwidth, unit, metric version, and eligible observation process were fixed independently of the observed post-target density.",
    "Absent the target, local metric density would have evolved smoothly enough that pre/post within-unit changes provide a useful counterfactual.",
    "The unit bootstrap reflects dependence; missing observations, rounding, censoring, and mechanical caps are not differentially concentrated at the target."
  ],
  "decision": "post_target_threshold_bunching_detected",
  "executive_summary": {
    "above_threshold_excess_mass_interval": [
      0.35,
      0.35
    ],
    "above_threshold_excess_mass_share": 0.35,
    "bunching_detected": true,
    "log_density_jump_change": 1.7814,
    "mirror_asymmetry_interval": [
      0.7143,
      0.7143
    ],
    "mirror_asymmetry_shift": 0.7143
  },
  "limitations": [
    "Bunching is a distributional anomaly, not proof of gaming, intent, fraud, or poor individual performance.",
    "Legitimate process improvement, reporting changes, seasonality, or a mechanical metric rule can create the same signature.",
    "The audit is local to the governed threshold and bandwidth and can miss diffuse manipulation away from the cutoff."
  ],
  "local_density": [
    {
      "bins": {
        "inner_above": 0.25,
        "inner_below": 0.25,
        "outer_above": 0.25,
        "outer_below": 0.25
      },
      "epoch": "pre_target",
      "local_row_fraction": 1,
      "local_weight": 400,
      "log_inner_above_below_density_ratio": 0
    },
    {
      "bins": {
        "inner_above": 0.6,
        "inner_below": 0.1,
        "outer_above": 0.15,

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 Freeze the KPI version, target threshold, activation epoch, bandwidth, eligible observation process, unit of dependence, and any weights before examining post-target density.
2. 2 Within each unit and epoch, restrict to two bandwidths around the threshold and calculate the share immediately above the threshold plus inner-above versus inner-below mirror asymmetry; require paired local support for every unit.
3. 3 Subtract each unit's pre-target local measures from its post-target measures, resample whole units for uncertainty, and corroborate excess mass with the change in a Jeffreys-smoothed inner-above/below log-density ratio.
4. 4 Flag bunching only when both excess mass and mirror displacement clear prespecified practical and directional-support gates; return local-density bins so reporting caps, rounding, and legitimate process shifts can be investigated.

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 a posttarget excess concentration immediately" }
  → finds "audit_kpi_threshold_bunching"

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

gitrevio_capability_run
  { "capability_id": "audit_kpi_threshold_bunching", "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 Delivery forecasting & commitments →