Compliance-by-design financial infrastructure
A financial knowledge base that structurally cannot recommend a product.
FinAtlas is a knowledge graph, dual-tier education, fail-closed triage and two-sided compliance audit — built so the red line lives in the data model, not in a policy memo.
A reusable base for the financial vertical — swap domain or tenant and reuse the graph, the audit standard and the red-line probes. Not a product-comparison tool, not a recommendation engine, not a sales funnel.
What it is
Three things it can demonstrably do — and one thing it is built never to do.
FinAtlas turns licensed subject-matter expertise into a reusable base: a knowledge graph plus an audit standard. The client-facing UI is a thin shell; the durable asset is the structure underneath it.
A browsable knowledge graph
Categories link to mechanisms, regulatory constraints, suitability factors, trigger questions and required materials — at both a plain-language and a professional tier. It cannot, by construction, produce a product score or ranking.
Live red lines you can test
Ask "which annuity should I buy?" and the system refuses and escalates to a licensed professional. Judgment questions are forced up the chain; ambiguity defaults to escalation (fail-closed).
Two-sided compliance audit
A LongProof dashboard audits process on both the client side and the agent side — did the flow stay bounded, did judgment escalate, was disclosure recorded — with an append-only trail. It never grades advice quality.
How it works
Structure first, so nothing downstream can drift into advice.
Compliance is not a disclaimer bolted on at the end. It is where the data model starts.
Model as structure
Domains and categories are wired to mechanisms, factors and regulatory-constraint pointers. There is no score, rank or best_product node type to write to.
Teach at two tiers
Every node carries both a literacy tier (plain language) and a professional tier (deeper mechanism + regulatory framing). Both stay education — never "buy" or "should".
Triage, fail-closed
Operational and explanatory questions are answered; judgment questions escalate. When intent is ambiguous, the default is to escalate, not to guess.
Audit the process
Both sides of the interaction are audited for flow discipline and disclosure — PASS, NEEDS_REVIEW or FAIL — with the auditor holding zero write access over what it audits.
Red-line probes
The guarantees ship as executable tests, not promises.
Each release runs a suite of red-line probes. A failure on any hard probe fails the whole build — the compliance posture cannot silently regress.
Probes P6, P8 and P11 additionally cover the m566 education boundary, evidence-safe wording, and the internal-demo guard (synthetic data + watermark present).
LongProof audit
Three states, and a rule that keeps them honest.
The audit reads flow discipline on both the client and the agent side. Its verdict is never "good advice / bad advice" — that is licensed activity, out of scope by design.
PASS
The flow stayed bounded, judgment escalated where required, and disclosure was recorded — with no recommendation in the output.
NEEDS_REVIEW
Something needs a human's eyes before it can be signed off — a licensed reviewer, not the system.
FAIL
A process boundary was crossed — for example a judgment question was answered instead of escalated.
UNAUDITABLE > FAKE_PASS. When the input is incomplete, the audit returns UNAUDITABLE rather than forcing a green. And the auditor sits outside the cell it inspects, with zero write access over the audited subject.
Who it's for
Built for people who carry compliance risk.
FinAtlas is a B2B infrastructure. It is reused by swapping domain or tenant — the client-facing surface is deliberately thin.
Licensed SMEs
Financial, insurance, tax and estate professionals who want to publish education and knowledge-graph content that stays product-neutral by construction.
Compliance & risk teams
Teams that need a process audit with a defensible, append-only trail — where "we couldn't audit this" is an allowed, honest answer.
Platform & white-label operators
Operators standing up a financial-vertical surface who want the graph, the tier axis, the triage classifier and the red-line probes as a reusable base.
Access
Pilot access, by conversation.
This is the internal-demo edition. The tiers below describe scope, not price — no public pricing is published for this edition. Figures are set per pilot.
Solo SME
For an individual licensed professional
- Knowledge-graph browsing + dual tier
- Fail-closed triage + the three-script output
- Append-only compliance trail
Team
For a practice or compliance group
- Everything in Solo SME
- Two-sided LongProof audit dashboard
- Tenant isolation across members
- Red-line probe suite on your content
Enterprise / white-label
For platform & white-label operators
- Reusable base — swap domain / tenant
- Custom graph + audit standard profile
- Deployment & integration support
No public prices, user counts, ratings or client logos are shown — this edition runs on synthetic data and has no live client base to cite. Terms for a real engagement are set when the compliance review (G1) applies.
See the red line hold, live.
Book a walkthrough of the graph, the triage refusal and the two-sided audit — on synthetic data, watermarked end to end. We'll show you what the structure makes impossible.