How It Works

A five-stage research pipeline.

From raw data to verified recommendation in under a minute. Every stage designed for one thing: eliminating confident-but-wrong answers.

Stage 01

Ingest

Real data. Not training data.

When you ask about a ticker, we don’t query some cached summary. We pull it live: the most recent 10-Q from SEC EDGAR, today’s Treasury yields from FRED, real-time prices from multiple market feeds, the latest employment and inflation prints from BLS, and nine other authoritative sources. Everything fresh. Everything traceable.

12+ primary sources per query
Stage 02

Analyze

Three reasoning engines. Same evidence. Independent analysis.

The gathered data is sent — in parallel — to three distinct reasoning engines: Claude, GPT, and Gemini. Each works in isolation. Each produces a structured analysis: recommendation, confidence level, supporting signals with cited data points, and explicit risk factors. No chain-of-thought contamination between models.

3 independent perspectives
Stage 03

Verify

Every claim, cross-checked against the source data.

A supervisor reviews all three analyses. It compares conclusions and flags genuine disagreements. Then it verifies every factual claim actually appears in the source data block. Any claim that can’t be verified is flagged as a red flag — and the overall confidence is downgraded accordingly.

Zero unverified claims
Stage 04

Calibrate

Consensus strength determines confidence.

Unanimous across all three models with verified data → HIGH confidence. Majority (2 of 3) → downgraded one level. Split decision → defaults to HOLD with LOW confidence. Any model returning INSUFFICIENT_DATA triggers an honest escalation: we tell you what we need to produce a more confident call.

Honest confidence calibration
Stage 05

Deliver

A brief you can act on. Or show your advisor.

The final output isn’t a paragraph of hedged copy. It’s a structured brief: the recommendation, the consensus strength, the agreed-upon signals, any disagreements between models, any red-flagged claims, and a plain-language summary. Every data point links back to its source.

Transparent, traceable, decisive
Our commitments

What we will never do.

Every number cited traces to a source

If a figure appears in an analysis, you can click through to exactly where it came from — SEC document, Fed series, or market feed.

We tell you when we don’t know

If the data is sparse or the models disagree, you see a LOW-confidence HOLD with an explicit list of what we’d need to be more certain — not a confident guess.

You see disagreement, not hidden consensus

If Claude says BUY but GPT says SELL, that disagreement shows up in your brief. Other tools hide it. We surface it.

See it in action.

Request access to the private beta.