See the data.
Invest smarter.
GlassBox shows you exactly why every risk score is what it is: the raw institutional data, the formula, and an independent audit, every time you check a stock. No black box. No score you have to just take our word for.
Eight legendary lenses. One audited score.
Each lens is pretrained on a legendary investor's documented strategy and reads your holdings for growth and risk. Every view traces back to the same audited data. Lenses inform; they never advise.
Berkshire Hathaway · Value / Buy & Hold
Fidelity Magellan · GARP
Citadel · Multi-Strategy
Renaissance Tech · Quantitative
Bridgewater · Global Macro
Millennium · Multi-Strategy
D.E. Shaw · Quant
Soros Fund Mgmt · Global Macro
click a lens to hear its story · drag / arrows to slide · filter to spin the desk
Stop guessing.
Start verifying with GlassBox.
Every lens gave you an opinion. GlassBox gives you the raw data, the formula, and an independent audit behind whatever score you're looking at, on your own holdings.
Free, non-commercial research preview · not investment advice
Not AI magic.
AI math. Audited.
Agent A1 pulls institutional data
Debt-to-equity, Altman Z-Score, free cash flow, 30 years of history, directly from Financial Modeling Prep's institutional API. Twelve Data provides real-time technicals. No aggregators, no intermediaries.
Agent A5 translates to plain English
A configured narrative agent generates a clear summary from the underlying risk metrics. Every number in the text traces back to its raw data source. Every claim is auditable, not asserted.
Agent A6 audits before you see it
The Auditor cross-checks the narrative against raw API data. If anything doesn't match, it's flagged and corrected before display. You only see verified output, and we publish the error rate publicly, every week.
What the models learned, and from where
Transparency isn't only about the audit. It's about what went in. Here's the provenance behind every score.
Institutional sources
FMP + Twelve Data: the same feeds funds pay for, never scraped aggregators.
Deterministic core
Risk math (Altman Z, D/E, beta) is formula-based and reproducible, no model guesswork on the numbers.
Narrative models
LLMs explain, they don't decide. Trained to summarize the metrics, constrained to cite every claim.
Audit layer
A6 checks every narrative against raw data. Weekly error rate published, nothing hidden.
Was this verification useful?
Your rating helps us track how useful the evidence trail actually is. It's logged locally in this demo, not yet wired to a live trust-score pipeline.
A free research preview.
GlassBox is a non-commercial research project. There are no paid plans, no payments and no trials. Everything here is simulated research, not investment advice.
Notes on transparency.
The Committee Casebook
All posts →Why we publish our own error rate
Every AI research tool makes mistakes. Most hide them. Here's why showing ours, every week, is the point.
How Agent A6 audits Agent A5
A walkthrough of the verification step that runs before any narrative reaches your screen, and what happens when it disagrees.
Reading a Debt/Equity ratio without the jargon
The raw metrics behind a risk score, explained the way we'd explain them to a client.