The method, in the open.

The model works from documents and data that any reader can request or reconstruct.

• Annual and quarterly reports, current reports and registration statements filed with U.S. regulators.
• Company announcements, press releases and investor materials published by the issuer.
• Historical price and volume series used only for context, not for a direction call.
• Public sector and industry classifications used to assemble peer groups.

We do not use paywalled research, private data or non-public information.

The work is extraction, comparison and presentation. The reader keeps the decision.

• Detects changes between a document and the prior dated reading.
• Flags unusual data points and links each one to its passage and accession number.
• Produces a short digest of the change, the context and the source trail.
• Keeps a dated version so a later reader can inspect what the tool saw at the time.

We do not state an accuracy rate or imply that a model output is complete.

These boundaries are part of the product, not temporary gaps.

• It does not give a price objective or a buy or sell call.
• It does not create a personalized plan or account-specific guidance.
• It does not state where a share price or market will go.
• It does not score management, rate a company or publish a recommendation.

If you need advice for your circumstances, speak with a licensed financial professional.

A sampled check keeps the extraction accountable without pretending the check is exhaustive.

• A sample of outputs is read against the underlying document by a person.
• A flagged discrepancy is corrected or withdrawn and the reading is reissued with a new date.
• Every correction keeps a note of what changed, so the archive is honest about its revisions.
• Readers can report an error to the address in the footer; we review reports during business hours.

Sampling cannot prove that every output is correct. See “Known limitations” below.

The method has edges. They are listed here rather than discovered by a reader after the fact.

• A change in accounting presentation can make two periods look different when the underlying business did not change.
• One-off gains, losses, impairments and restatements can distort a simple comparison.
• Trading halts, low liquidity and thin peer sets reduce the usefulness of market context.
• Public data can be delayed, revised, incomplete or wrong; a document may also be filed late.
• Language-model extraction can miss a qualifier, misread a table or attach a number to the wrong passage.

Check the underlying filing before relying on any extracted figure.

We do not sell a track record because the product is not a strategy.

• We do not publish return figures, success rates or a comparison with a benchmark.
• We do not present a historical example as proof that the method will repeat.
• The illustrative report on this site exists to show the format, not the results of a live account.

Past performance does not guarantee future results. Investing involves risk.