One graph, where everything means something.
Every module reads and writes one governed knowledge graph, and an ontology disciplines what that graph is allowed to say. This page is the deep end: how the vocabulary is enforced, how meaning becomes executable checks, and how the graph answers for any date in its past.
Declared once. Enforced at write time.
The graph has a declared vocabulary: which kinds of things can exist, and which relationships can connect them, in which direction. That declaration is not documentation. A write that uses an unknown type, or points a relationship the wrong way, is refused on the spot, not logged for later.
Hover or tap any relationship. Every edge is typed, directed, and citable back to the record that created it.
And the vocabulary cannot quietly drift from the code. An automated census walks every place the product writes to the graph and checks both directions: everything declared is actually written somewhere, and everything written is declared. If the two disagree, the build fails.
The graph, as it stood on any date.
Every edge carries the interval it was true for. Nothing is overwritten, so the graph can be replayed exactly as it stood on any past date: which pipeline fed which report, what a column meant, which rule governed it. When a records rule asks you to reproduce the past, this is the native answer, not a restore from backup. Try it below.
On Today: no change since 31 Mar 2025, and the record can prove that too.
6 live edges · every edge carries its interval · illustrative historyBusiness meaning compiles into controls.
An ontology here is not a poster on the wall. Approved statements of meaning become executable data-quality checks through a mechanical path with no model in the middle, and every step of that path defends itself.
pick a stage · the full three-step story continues below
A steward approves an axiom
Trade date precedes settlement date. That is business meaning, written once, versioned like everything else, and shown verbatim wherever it is used.
A registered generator compiles it
The platform turns the axiom into an executable check against columns a person has confirmed. Content packs never ship their own query logic, so a pack cannot smuggle code in.
The pack proves itself before it loads
Every axiom carries its own pass and fail examples, and they run through the real evaluator at load time. A pack whose examples cannot tell right from wrong is refused.
When the AI proposes a rule on top of this, the proposal card shows the axiom verbatim, with citations into the packs and mappings it stands on. Your reviewers see where a suggestion came from, every time.
Terms relate the way your business talks.
Synonyms, broader and narrower terms, and related concepts are typed edges in the same graph, not notes in a glossary field. So the platform can answer a question like: show me every dataset that means drawn exposure, across synonyms and narrower terms, and cite every hop it took to get there.
What this graph refuses to pretend.
Alignment by pointer, never by import
Standards such as FIBO, ISO 20022, and ISDA CDM arrive as compiled, versioned pack subsets, and the full external ontologies are referenced as alignment tags rather than bulk-copied. Your vocabulary stays yours, and every alignment stays auditable.
An unbound system is a gap, never a pass
Coverage is a governed denominator. A system nobody has mapped stays in the count and is reported as a named gap, not quietly dropped from the percentage.
When grounding is missing, the AI says so
AI suggestions cite the ontology, the confirmed mappings, and the sealed record. Where a column has no confirmed meaning, the suggestion is labeled schema-only instead of pretending.
Start with proof, not a promise.
There are three ways in, and each one earns the next. You approve the exit criteria before we spend any more of your time or budget.
A 60-minute working session
Bring one report that is expensive to prove. We walk the mechanism against it. Nothing is required from your side.
A two-week guided sandbox
Your team works hands-on against a synthetic sample bank. It requires zero access to your systems and zero procurement.
A six-week proof on your systems
We use read-only metadata across five systems and one priority report, scored against criteria both sides sign first. The decision it produces is yours.