Finance training data that is licensed, not scraped

For AI labs, fintech model teams & eval builders

This site publishes a Content-Signal: ai-train=no policy and blocks AI crawlers at the edge — so our corpus is not already in your pre-training mix. If you are building or evaluating a finance model, you can license it properly instead: analytically-verified question/answer pairs, MCP tool call traces, and ground-truth calculation datasets across accounting, valuation, fixed income, actuarial, compliance and 14 more disciplines.

Request a data sample & terms →
Why it is different
Every numeric answer in the set is checked against a closed-form or reference calculation — Black-Scholes, Hamada, MACRS, ASC 842, chain-ladder — not against another model's output. You are licensing verified data, with the provenance documented.

What is available to license

Set A · Instruction pairs
Verified finance Q&A
Task-shaped prompts with worked, checked answers — DCF, LBO, WACC, options Greeks, lease accounting, reserving, credit metrics. Each carries the ground-truth method and tolerance used to verify it.
Set B · Tool-use traces
MCP call / response logs
Structured traces from live finance MCP tools (EDGAR, portfolio risk, valuation, market data): the tool schema, the arguments a model chose, the response, and whether it was correct.
Set C · Eval ground truth
Scorer datasets
The analytically-computed reference values and pass criteria behind the public benchmark — usable as a held-out finance eval for your own model.
Set D · Editorial corpus
Practitioner guides
200+ long-form workflow guides on how finance professionals actually use AI — regulator-aware, domain-specific, written for this site and not published for training.

Sets are licensed individually or bundled. Format is JSONL with a documented schema; delivery is a one-time export or a periodic refresh. See the public benchmark methodology → that Set C is drawn from.

Licensing posture

01
You tell us the use
Pre-training, fine-tuning, RAG index, or internal eval — the grant, attribution, and refresh terms are written to match. We do not sell an unlimited redistribution licence.
02
Sample first
A representative sample (a few hundred rows per set) goes out under a short evaluation licence so you can check fit before committing.
03
Documented provenance
Every licence ships with a datasheet: how each row was produced, how it was verified, and what it must not be used for. Useful when your own model governance asks where the data came from.

Request a sample & terms

Tell us what you are building. We reply within one business day.

Want to embed the content in a product instead of train on it? Content & tooling licensing →
Benchmark methodology →
FEEDBACK