Skip to content

AI transparency

This page consolidates the transparency material for Venturi's decision-time adapter: what the system decides, how confidence is represented, what happens when the model is unavailable, and how a human can review or override an attribution result.

Release state

The transparency semantics below are documented and customer-facing. The model-card index is published at Model-card index, but model cards remain gated by Phase-2 model-ops evidence and are not presented as GA until the card is published there.

Decision flow

  1. A metadata-only invocation record reaches the serving plane through the gateway, proxy, ingestion API, or connector path.
  2. Deterministic checks attach tenant, source-system, policy, and timing context.
  3. The heuristic reconciliation engine evaluates candidate ownership and spend paths.
  4. If the trained decision-time adapter is available inside its latency budget, it may score the attribution edge.
  5. If the adapter is unavailable, slow, stale, or demoted, the system serves a deterministic HeuristicBaseline result instead.
  6. The output carries an evidence trail, confidence value, result source, and output state.

The serving path remains fail-open: Venturi never blocks customer AI traffic to wait for attribution.

Confidence semantics

Operational confidence (coper) is deliberately capped at 0.95. Venturi does not display certainty. Confidence is used to decide whether an edge is chargeback-ready, review-needed, or analysis-only; it is not a claim that the system has perfect visibility into every pathway.

See Confidence & evidence for output states and evidence-card fields.

Output states

State Meaning Customer action
Chargeback-ready Evidence and confidence meet the configured threshold. Eligible for chargeback workflows.
Needs review Evidence exists but the edge needs human confirmation or additional context. Review the evidence card and accept, correct, or defer.
Bounded The system can bound the result but cannot resolve a precise edge. Use for planning and coverage work, not chargeback.
Not identifiable The pathway cannot be resolved with connected evidence. Connect the would-close source or leave out of chargeback.
Heuristic baseline The trained adapter did not serve; deterministic fallback answered instead. Treat as visible but degraded; filter or exclude until re-observed.

Human controls

Reviewers can accept, correct, or reject attribution evidence according to their role and tenant policy. Review actions are written to the audit trail. Corrected edges keep the original evidence and the reviewer action so downstream reports can distinguish model output, heuristic fallback, and human review.

Data use and opt-out boundary

Venturi processes metadata needed for attribution: identity, service, project, cost, token counts, timing, policy context, and source-system evidence. Prompt and completion bodies are not requested for connector setup and are not needed for these transparency controls.

If a pathway cannot be observed without content capture, Venturi treats that as a known limitation rather than silently expanding capture. See Known limitations.