<!-- Canonical: https://myoctopus.ai/trust/system-cards/knowledge-qa -->

1.  [Home](/)
2.  /[Trust](/trust)
3.  /[Model System Cards](/trust/system-cards)
4.  /Knowledge Q&A AFI capability

Model System Card

# Model System Card — Knowledge Q&A AFI capability

Version 1.0

Last updated September 2026

Public document

This AFI capability answers plain-language finance questions from the customer's own reconciled data and logged commentary.

Figures are retrieved or computed deterministically and cited to source. The model composes the sentence around them.

## 1\. Purpose and intended use

*   Answer questions about actuals, budget, forecast, and variance for periods the customer has connected.
*   Return the recorded business reason for a movement, attributed to the person who supplied it.
*   Retrieve the definition of a metric as the customer has defined it.
*   Point the user to the underlying line items behind any figure it returns.

## 2\. Out-of-scope uses

*   Answering from public or general internet knowledge as if it were customer data.
*   Legal, tax, audit, or accounting-policy advice.
*   Any question requiring data the requesting user's role is not permitted to see.
*   Predicting outcomes outside the connected data set.

## 3\. Data sources and access scope

| Source | Access |
| --- | --- |
| Reconciled data layer | Figures already reconciled to the general ledger, read-only. |
| Logged commentary | Variance explanations supplied by named budget owners, with author and timestamp. |
| Customer definitions | Chart of accounts mappings and metric definitions maintained by the customer. |
| Access control | Answers are filtered by the requesting user's role. Role-based access control is applied before retrieval, not after. |

## 4\. Computation boundary

| Layer | Responsibility |
| --- | --- |
| Deterministic code | Retrieval, filtering, aggregation, currency conversion, and any arithmetic in the answer. |
| Language model | Understanding the question and composing the explanatory sentence. |
| Boundary rule | If a figure cannot be retrieved or computed deterministically, the AFI capability says so instead of estimating one. |

## 5\. Evaluation approach

*   Citation tests: every figure in an answer must resolve to a source line item.
*   Refusal tests: questions outside scope, or outside the user's permissions, must return a refusal.
*   Consistency tests: the same question asked repeatedly must return the same figure.
*   Human review of sampled answers during onboarding and at intervals thereafter.

## 6\. Known limitations

*   Answers cover only connected periods and mapped accounts.
*   The recorded business reason reflects what a budget owner stated. It is not independent verification.
*   Ambiguous questions can be interpreted narrowly. The AFI capability states its interpretation with the answer.
*   Wording is model-generated and can vary between sessions even when the figures do not.

## 7\. Human approval gates

*   Answers shared outside the platform, and any outbound message generated from an answer, are approval-gated.
*   Metric definitions and permission scopes are set by the customer, not inferred by the model.
*   New data sources require explicit customer authorisation before they are queryable.

## 8\. Logging and lineage

*   Every question and answer is logged with the requesting user, timestamp, and the sources used.
*   Every figure carries lineage back to the general ledger.
*   Refusals are logged with their reason.
*   Query logs are exportable for audit and access review.

## 9\. Escalation path

*   A question the AFI capability cannot answer within scope is routed to a named human.
*   Suspected data-quality problems raise an internal item before an answer is given.
*   Availability incidents are published on the status page.

## Questions and answers

### Does the Knowledge Q&A AFI capability calculate figures with a language model?

No. Arithmetic, consolidation, and currency conversion run as deterministic code against a reconciled data layer. The language model handles wording only.

### What access does the Knowledge Q&A AFI capability have to source systems?

Read-only access, scoped per source system, with role-based access control applied before data is retrieved.

### Is human approval required before the Knowledge Q&A AFI capability acts?

Yes. Every outbound message to a person is approval-gated and a named human signs off before anything sends.

### What is logged when the Knowledge Q&A AFI capability runs?

Every action is logged with who, what, and when, and every figure carries line-item lineage back to the general ledger. Logs are exportable for audit.

This system card describes documented product behaviour and controls. It is not a warranty of outcomes. Figures produced by any AFI capability remain subject to the customer's own review and sign-off.

## Related pages

*   [All Model System Cards](/trust/system-cards)
*   [The Octopus AI Constitution](/trust/constitution)
*   [Transparency Report](/trust/transparency-report)
*   [AI governance in finance (SR 11-7, DORA)](/ai-governance-finance)
*   [Trust Center](/trust)
