Audit AI knowledge grounding integrity

What it's for

Turns 'grounded AI' from a demo claim into an auditable chain: the right version, the right permissions, the actual retrieved chunk, every cited claim, and a later real-world outcome.

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.

{
  "answer_traces": [
    {
      "claims": [
        {
          "cited_chunk_ids": [
            "runbook-chunk-1"
          ],
          "id": "claim-1"
        }
      ],
      "evidence_verified": true,
      "grounded_outcome": true,
      "id": "answer-1",
      "occurred_at_ms": 2000,
      "outcome_mature": true,
      "principal_policy_ids": [
        "payments-oncall"
      ],
      "reported_citation_count": 1,
      "reported_retrieval_count": 1,
      "required_source_ids": [
        "payments-runbook"
      ],
      "retrieved_chunk_ids": [
        "runbook-chunk-1"
      ]
    }
  ],
  "indexed_chunks": [
    {
      "effective_policy_ids": [
        "payments-oncall"
      ],
      "evidence_verified": true,
      "id": "runbook-chunk-1",
      "indexed_at_ms": 1100,
      "source_content_hash": "sha256:runbook-v1",
      "source_version_id": "runbook-v1"
    }
  ],
  "source_versions": [
    {
      "authoritative": true,

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": [
    "Source versions are immutable, effective intervals are half-open, hashes come from the authoritative source bytes, and the index preserves the source access policy exactly.",
    "Every claim cites retrieved evidence; required-source coverage is defined prospectively for the answer class and a citation proves lineage, not semantic entailment or factual truth.",
    "Unmatured outcomes remain in coverage denominators but are never coded as failures; only outcomes governed before inspection may label an answer grounded.",
    "The audit is aggregate system assurance, not employee surveillance, individual performance scoring, legal advice, or authorization to expose restricted content."
  ],
  "chunk_diagnostics": [],
  "configuration": {
    "minimum_evidence_coverage": 0.95,
    "minimum_mature_outcome_coverage": 0.8
  },
  "decision": "ai_knowledge_grounding_integrity_supported",
  "failed_gates": [],
  "method": "ai_knowledge_source_index_claim_lineage_audit_v1",
  "source_timeline_diagnostics": [],
  "summary": {
    "answer_trace_count": 1,
    "claim_count": 1,
    "evidence_coverage": 1,
    "grounded_claim_count": 1,
    "grounded_claim_rate": 1,
    "indexed_chunk_count": 1,
    "mature_grounded_outcome_rate": 1,
    "mature_outcome_count": 1,
    "mature_outcome_coverage": 1,
    "source_count": 1,
    "source_version_count": 1,
    "valid_answer_trace_count": 1
  },
  "trace_diagnostics": [],
  "truncated_chunk_count": 0,
  "truncated_source_timeline_count": 0,
  "truncated_trace_count": 0
}

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 Validate consecutive non-overlapping source-version intervals, immutable content hashes and exact source-to-index access-policy projection.
2. 2 At each answer time, require every retrieved chunk to exist, already be indexed, be temporally effective, match source lineage and be authorized for the principal.
3. 3 Require every claim citation to be an eligible retrieved chunk, every prospectively required source to be represented, reported counts to reconcile and immature outcomes to remain unresolved.

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 the complete ai knowledge supply" }
  → finds "audit_ai_knowledge_grounding_integrity"

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

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