Financial statements, checked against themselves.
Assay reads the numbers a company signs in its SEC filings and pins the contradictions between them, quarter by quarter: margins that move without cash following, receivables that outrun revenue, restatements that quietly rewrite last year. Every finding traces back to an EDGAR accession number.
Built on SEC XBRL filings, which are public domain. Runs on your own machine. Sells no signals.
Counts from the public status endpoint on September 11, 2026. Aggregates only; no single company's figures are exposed there.
Three instruments, one set of numbers.
Everything is derived from the same filings, so the cockpit, the doctor and the forecast never disagree about what a company reported.
Company Doctor
Twelve anchored vital signs per quarter: revenue, gross and net margin, operating and free cash flow, cash conversion, receivable and inventory days, contract liabilities. Each is judged against the company's own history and its sector.
- Findings split into new versus ongoing, so a chronic issue does not ring the alarm every quarter
- Attribution and a written prescription for each new alert
- Year-over-year comparisons demoted when a restatement rewrote the base
Cockpit and Ledger
The whole coverage on one screen: which companies developed new anomalies this quarter, which are chronic, which are clean. Drill from a name to its statements and, for 45 companies, to a reconstructed operating ledger that reconciles to the disclosed totals.
- Insider transactions and congressional trades alongside the fundamentals
- Market weather: price and volume read as regimes, not predictions
Forecasts that keep score
Every daily forecast is issued once, sealed, and settled against a benchmark pool frozen on the day it was issued. The scorecard is public, including the misses. Contributors stake a fixed budget of points on their own calls.
- Kelly-sized stakes: no skill, zero expected gain
- A daily evidence file, hashed and pushed to a public repository
We would rather be checkable than impressive.
Most research tools ask you to trust a score. Assay is built so that a stranger with the same public filings can reproduce every number, and catch us if a record was touched after the fact.
01Forward records are sealed
Once a forecast is issued it cannot be edited or withdrawn. Each day the live forecasts, the benchmark pool and the diagnosis snapshot are hashed into an evidence file that chains to the previous one and is published to a public git repository. Anyone can clone it and run the verifier.
02Failures are labeled, not implied
The collapse library holds 226 events, each anchored to the 8-K or enforcement release that made it public, with the exact filing date. Criteria are evaluated against these labels in the open, and the results are published whether or not they flatter the method.
03We publish the negatives
We back-tested a composite "verdict" score and found it had no predictive power for returns. We kept the score as a description and say so in the interface. A tool that sells signals cannot afford that sentence; we can.
04Your machine does the work
The desktop and mobile apps ship with the data and run every query locally. Nothing about what you look at leaves your device. The server exists for one purpose: to keep score in a place you cannot edit.
"Pin the contradictions between the numbers a company signed, quarter by quarter, in plain sight." The whole product, in one sentence. We do not claim to predict collapses; we claim to make the disclosed record legible.
Public sources, stated licenses, downloadable in full.
Every dataset carries its source and license status. What we are allowed to redistribute, we do, as a versioned package with a checksum.
| Dataset | Source | License | Rows |
|---|---|---|---|
| Quarterly financial anchors | SEC XBRL companyfacts | Public domain | 11,344 |
| Filing dates and point-in-time views | SEC EDGAR submissions | Public domain | 10,746 |
| Insider transactions | SEC Form 4 | Public domain | 434,631 |
| Congressional trades | Periodic Transaction Reports | Public domain | 8,931 |
| Collapse labels | 8-K item 1.03, SEC AAER | Public domain | 226 |
| Diagnoses, attributions, prescriptions | Derived by Assay | Open | 187,292 |
| Daily market prices | Market data vendor | Not redistributed | 1,695,116 |
Market prices and anything derived from them stay on the server and out of the package. The alert criteria do not depend on prices; they read filings only.
assay-public-2026-09-11.zip
22 datasets in Parquet, with a manifest that lists what was excluded and why, and a SHA-256 file to verify the download.
The desktop app checks this index automatically and verifies the checksum before swapping data.
The record is free. The instruments are not.
The public scorecard, the data package and the coverage tools stay free for everyone, because a record you must pay to inspect is not a record. Membership pays for the diagnostic layer.
- Cockpit and ledger-versus-ticker views
- Coverage lookup and data health
- Public forecast scorecard
- Versioned data package with checksums
- Company Doctor: vital signs, attribution, prescriptions
- Market weather and stance
- Insider and congressional transaction views
- Daily forecast subscription and submissions
- Everything in Pro
- Strategy Lab: trend and diagnosis-state back-tests
- API keys for programmatic access
Membership is managed inside the app and billed through Stripe. Access ends the moment a subscription lapses; there is no grace period and no dark pattern on the way out. Prices in US dollars.
Assay 0.2.0
Desktop builds ship with the current data package and work offline. Updates to the data are fetched and verified in-app.
Browser
The full app, running in your browser. Nothing to install; the data package (about 25 MB) and the query engine load once and are cached.
Open the web appMobile
The same instruments, offline on iOS and Android. In testing now; the query layer is shared with the desktop build byte for byte.
All releases