Audit point in time model integrity

What it's for

Turns trustworthy AI from a claim into an auditable pre-deployment gate: Gitrevio can show that every feature and label was actually knowable when the model would have acted and that holdout evidence was not trained on indirectly.

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.

{
  "embargo_time": 5,
  "feature_names": [
    "cycle_time",
    "incident_rate"
  ],
  "observations": [
    {
      "entity_id": "entity-0",
      "feature_available_times": {
        "cycle_time": -2,
        "incident_rate": -1
      },
      "id": "integrity-train-0",
      "outcome_resolved_time": 10,
      "outcome_start_time": 0,
      "prediction_time": 0,
      "snapshot_created_time": -1,
      "source_record_id": "source-train-0",
      "split": "train"
    },
    {
      "entity_id": "entity-1",
      "feature_available_times": {
        "cycle_time": -1,
        "incident_rate": 0
      },
      "id": "integrity-train-1",
      "outcome_resolved_time": 11,
      "outcome_start_time": 1,
      "prediction_time": 1,
      "snapshot_created_time": 0,
      "source_record_id": "source-train-1",
      "split": "train"
    },
    {
      "entity_id": "entity-2",
      "feature_available_times": {
        "cycle_time": 0,
        "incident_rate": 1
      },
      "id": "integrity-train-2",
      "outcome_resolved_time": 12,
      "outcome_start_time": 2,

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": [
    "prediction_time is the instant the model or policy would have acted, and feature_available_times record actual system availability rather than event occurrence or later warehouse ingestion guesses.",
    "Each earlier split was trained or calibrated using all rows assigned to it, so its latest outcome resolution must clear the next split's earliest prediction by the declared embargo.",
    "source_record_id identifies the underlying labeled decision unit; reuse across splits is contamination even when row ids differ."
  ],
  "configuration": {
    "embargo_time": 5,
    "feature_names": [
      "cycle_time",
      "incident_rate"
    ],
    "maximum_row_violation_rate": 0,
    "split_order": [
      "train",
      "calibration",
      "test"
    ]
  },
  "decision": "point_in_time_integrity_passed",
  "detail_counts": {
    "duplicated_source_record_ids": 0,
    "violating_rows": 0
  },
  "duplicated_source_record_ids": [],
  "executive_summary": {
    "contaminated_split_boundaries": 0,
    "observations": 60,
    "row_violation_rate": 0,
    "source_records_reused_across_splits": 0,
    "violated_rows": 0,
    "worst_feature_violation_rate": 0
  },
  "feature_diagnostics": [
    {
      "feature_name": "cycle_time",
      "post_prediction_availability_rows": 0,
      "violation_rate": 0,
      "wilson_95_interval": [
        0,
        0.0602
      ]
    },
    {

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 Require a frozen split manifest in chronological deployment order and an exact availability timestamp for every declared feature on every observation; occurrence time is not accepted as a substitute for system availability.
2. 2 Flag features and snapshots created after prediction, target windows that begin before prediction, impossible target-resolution order, and one underlying source record reused across different splits.
3. 3 For every adjacent split boundary, require the latest outcome resolution in the earlier split plus the governed embargo to precede the earliest prediction in the later split, matching a batch model that consumed all earlier-split labels.
4. 4 Aggregate leakage by row, violation type, feature, and split boundary; use the practical row tolerance only for row-level exceptions while any contaminated boundary or cross-split source reuse remains a hard failure.

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": "gate an analytical or ai model" }
  → finds "audit_point_in_time_model_integrity"

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

gitrevio_capability_run
  { "capability_id": "audit_point_in_time_model_integrity", "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 →