Traceable to the line item
Every figure links back to its original source, page, and line — across regulatory filings, earnings transcripts, annual and quarterly reports, and market data.
Verifiable AI Infrastructure
Trust requires Proof
Contentius builds verifiable, traceable AI for industries where being wrong is unacceptable — beginning with finance. Every figure is computed deterministically and traced to its original source, page, and line item.
Interpretation✦Ontology✦Deterministic Data
Backed by prominent technology founders & operators
Scroll
The Problem
A plausible answer is nota provable one.
Large language models are built on probability. Their outputs carry an irreducible randomness, so the same question can return different figures from one moment to the next — frequently with no citation at all.
But the deeper problem is not the occasional error. It is that even a correct answer cannot be proven. It cannot be traced back to where it came from, and it cannot be defended — not in a boardroom, and not in a courtroom.
In finance, that is disqualifying. A number you cannot source is a number you cannot use. Here, trust is not a feeling and not a brand. Trust requires proof.
The Architecture
Separate language
Separate language
from data
Contentius is built on a single discipline: language and data never mix. The model interprets what you mean. Deterministic code retrieves and computes what is true. The two communicate only through structured contracts — typed, validated handoffs — never raw model output. Randomness is confined to interpretation. The numbers are never guessed.
Language — interpretation
Your intent
A question asked in plain language — the analysis you actually need.
01
Personalization
Templates, preferences, and saved workflows shape how the question is read — without touching the data.
02
AI interface
Interprets the question and builds an execution plan. It decides what must be done — but never retrieves data directly.
Data — computation
03
Ontology & semantic layer
Maps concepts, relationships, and lineage so synonyms resolve to one verified value.
04
Data platform
Deterministically retrieves, computes, and cites every figure it returns.
A cited, reproducible result
The same input always yields the same output — every figure carrying its source.
Why it matters · 01
Randomness, contained
The model is allowed to be creative about understanding you — and nothing else. Because it never touches the numbers, the irreducible randomness of language can never leak into a figure. Interpretation can vary; the computation cannot.
Why it matters · 02
Contracts, not guesses
Language and data speak only through structured contracts — typed, validated handoffs with a defined shape. There is no free-text number passed from a model into a report. What crosses the boundary is checked, not trusted.
Why it matters · 03
Point-in-time & reproducible
Every result is anchored to a moment in time and is fully reproducible. Run the same analysis tomorrow, or in an audit two years from now, and you get the same answer, traced to the same source.
The Stack
Four layers.
Four layers.
One verified value.
Templates, preferences, and saved workflows. Contentius remembers how you, your team, and your firm actually work — the analyses you run, the conventions you keep, and the corrections you've made — so the system meets you where you are instead of starting from zero every time.
Because this layer holds preference, not fact, it shapes how a question is read without ever influencing the value that comes back.
Interprets your question and builds an execution plan. This layer reads intent — what you are really asking, in what context, to what end — and turns it into a precise, structured plan of what must be retrieved and computed.
Critically, it never retrieves data directly. It is the boundary that keeps the model away from the numbers — the line between understanding a question and answering it with proof.
Maps concepts, relationships, and data lineage. “Revenue,” “net sales,” and “topline” resolve to the same verified value — because the system knows what each term means, how it connects to others, and where it ultimately comes from.
This is where meaning is made unambiguous. Every concept carries its definition and its lineage, so an answer is never an accident of phrasing.
Deterministically retrieves, computes, and cites every figure. Results are reproducible and point-in-time: the same input always yields the same output, and every output carries its source — down to the page and line item.
This is the foundation that makes the whole system defensible. Nothing here is sampled or guessed; it is retrieved, computed, and shown its work.
Live in production · Investment banking & financial analysis
Every figure,traced to source
Our first product is live and in production, purpose-built for investment banking and financial analysis. It indexes regulatory filings, earnings transcripts, annual and quarterly reports, and market data across global markets — and ties every figure back to its original source, page, and line item.
Total revenue · FY2025
$4.13billion USD
Form 10-K · p.74
Line: “Total revenue”
Audio · earnings call 04:182
Reproduced point-in-time — same input always yields the same output.
Concept Revenue
resolves to Total revenue (net)
FY2025 10-K · p.74
value verified
Comparable companies · illustrativeEV / EBITDA
| Issuer (example) | EV/EBITDA | P / E | Rev. growth |
|---|---|---|---|
| Atlas Materials | 11.4× | 18.2× | +7.1% |
| Meridian Energy | 9.8× | 15.6× | +4.3% |
| Northwind Logistics | 12.7× | 21.0× | +9.4% |
Background agent flagged an outlier. One comparable's margin differs from its prior four quarters by an unusual amount — review the source before it enters the model.
Export to spreadsheet
Export to presentation
Re-run point-in-time
Illustrative interface — every figure carries its source, page, and line item
Verified exports
Send results to spreadsheet and presentation formats with every data point verified — so the file you hand over is as defensible as the screen you built it on.
Memory at every level
Retains preferences and corrections at the individual, team, and firm level — so a fix made once is honored everywhere, and the system reflects how your firm works.
Auditable workflows
Valuation, comparable-company analysis, and more — fully editable and fully auditable, with each step open to inspection and every input shown its work.
Background agents
Agents run continuously in the background to flag outliers and likely errors — catching the figure that doesn't belong before it reaches your model.
Global coverage
Indexes large volumes of filings, transcripts, reports, and market data across many global markets — the breadth a real diligence process demands.
Contentius · Across finance
Applications
- Investment Banking
- Equity Research
- Hedge Funds
- Private Equity
- Compliance
- Tax & Accounting
- Valuation
- Comparable Companies
- Earnings Analysis
- Regulatory Filings
- Audit & Data Lineage
- Point-in-Time Reporting
On the roadmap
Every application carries full data lineage for audit. And next — an industry-agnostic platform: connect your own data, define your own workflows, and deploy verifiable AI agents anywhere being wrong is unacceptable.
Request access
Start with proof.
Tell us what you need to verify, and we'll show you how Contentius traces it to the source. We work with investment banking, equity research, hedge funds, private equity, compliance, and tax & accounting teams.
Request received.
Thank you — your message is ready to send from your email client, and a copy of the details is on its way to our team. We'll be in touch shortly.