Trust & Safety

What this system is, what it is not, and precisely what its results mean. This page is part of the product, not a footnote.

What Platrixa is

Platrixa is financial semantic validation infrastructure — a validation assistant. It takes AI-generated interpretations of financial language and documents, validates them against explicit schemas and grounding requirements, and routes them to deterministic, registry-registered authorities that produce the accounting result.

AI understands. Platrixa validates. Deterministic authorities calculate and execute.

What Platrixa is not

Not a financial adviser: Platrixa does not recommend transactions, strategies, or financial products.
Not an investment adviser: Nothing here is a solicitation, recommendation, or forecast of any security or market.
Not a tax adviser: Classification follows the documented FYJC framework only; it is not tax advice for any jurisdiction or situation.
Not a legal adviser: Interpretations are not legal opinions and carry no legal standing.
Not an autonomous accountant: The runtime validates and executes deterministic accounting for inputs it can prove; it does not run a business's books or act on its own.
Not a guarantee of financial correctness: Validation is only as complete as the implemented authorities and rules it checks against.

What VERIFIED means — and does not

VERIFIED means

The input passed the deterministic validation and authority requirements implemented by Platrixa.
Every claim was grounded in the supplied input under the grounding gate's implemented rules.
A registered deterministic authority produced the accounting result.

VERIFIED does not mean

Everything about this financial situation is guaranteed true.
The classification is the only defensible one under a different framework.
Future inputs will behave the same way.

Operating principles

Model output is untrusted: The language model only proposes an interpretation. It never decides a status, never computes accounting, and its output is discarded whenever validation or grounding fails.
Grounding matters: Interpretations must be tied to explicit input evidence. Inferred or missing evidence fails closed to REVIEW_REQUIRED or rejection — ambiguity is surfaced, never guessed away.
Unsupported inputs are rejected: When no deterministic authority accepts an input, the answer is UNSUPPORTED — recorded as a boundary, not silently converted into something else.
Ambiguous inputs may require review: REVIEW_REQUIRED is a legitimate outcome: the system says what it could not prove instead of producing a confident guess.
Authorities are deterministic: Journal entries and formulas come from versioned, test-referenced authorities — the same input and registry state produce the same result.
The system can be wrong: Validation is bounded by the rules and data implemented today. Bugs, gaps, and framework mismatches are possible.
Users must independently verify consequential decisions: Platrixa is validation infrastructure, not a professional. Anything consequential deserves review by a qualified human before you rely on it.

Questions about scope or safety? The developer documentation describes the exact six-state contract and error behavior, and the capability registry shows precisely what is supported, partial, unsupported, and planned — unfiltered.