A method you can inspect.
A readout is only as useful as the chain behind it. This page documents the inputs, the steps, the checks, and the failure modes — including the things the model refuses to do.
Data we read
Moratrix reads public documents and public market data. The list is deliberately short, so the chain from source to readout stays inspectable.
- Annual and quarterly reports, including Forms 10-K and 10-Q.
- Current reports and material announcements, including Form 8-K and company news releases.
- Historical price and volume data for U.S.-listed common stocks.
- Public sector and industry classifications used to assemble a peer set.
We do not read paywalled research, private data rooms, non-public information, or material obtained under a confidentiality obligation.
What the model does
The pipeline turns a document into a dated, sourced readout. It is an extraction and comparison tool, not a market oracle.
- Extract. Identify tables, figures and statements in a public filing or announcement.
- Compare. Check the extract against the prior readout and flag changes in figures, language or disclosure.
- Describe. Write a short note on what changed, attach the source reference, and show the relevant price and volume context.
The output says what the document says and where it says it. It does not assign a quality score to a company or a security, and it does not tell you what to do with the information.
What it does not do
These are product boundaries, not disclaimers added at the end. If a feature would cross one of them, it does not ship.
It does
- Read public filings and announcements.
- Show what changed against the last readout.
- Attach the source and the date.
- Place price and volume beside the disclosure.
- Explain a known limitation.
It does not
- Give buy or sell recommendations.
- Set price objectives or ratings.
- Offer personalised advice or manage money.
- Tell you what the market will do next.
- Accept payment from a company to cover it.
Validation
We sample outputs and compare the extract with the source paragraph. A figure without a traceable source is treated as an error, not a design choice.
When a reader reports a mistake, we review the source, correct the readout where the correction is unambiguous, and keep the correction dated. We do not promise an accuracy rate, and a sample being reviewed is not a guarantee that every output is correct.
Report an error at [email protected] with the ticker, the readout date and the passage you are questioning.
Known limitations
These are the cases where a readout can mislead even when the pipeline works as designed.
- Filing basis changes. A company can restate figures, change its fiscal year or redefine a segment; a like-for-like comparison may not exist.
- One-time items. A gain, charge or impairment can make a trend look structural when it is not.
- Halted or low-liquidity securities. Price and volume context can be thin, stale or unrepresentative.
- Delayed or revised market data. Third-party feeds can be late, adjusted or wrong.
- Document parsing. Tables, footnotes and scanned exhibits can be misread by an extraction model.
- Ambiguous language. A filing can describe the same thing two ways; the readout may quote the wrong passage.
The Methodology page is where these stay visible. Hiding them would make the product look more certain than it is.
No performance claims
We do not publish investment performance figures, a track record, or any comparison against a benchmark. We do not present a historical sample as a recommendation that worked.
The sample reports on this site are format demonstrations. They show the shape of a readout — metrics, sources, a dated note — not a record of picks or a claim about what would have happened.
Read the Risk Disclosure for the full set of limitations that apply to your use of this site.
Read the risk disclosure