How the desk works.

Data we read

The desk reads documents a company files with the SEC: annual reports (Form 10-K), quarterly reports (Form 10-Q), current reports (Form 8-K), and registration statements when they matter. It also reads company press releases and investor materials, historical end-of-day price and volume, and a published sector classification.

• Annual and quarterly reports — business description, risk factors, management discussion, financial statements and notes.
• Current reports and announcements — material events, guidance changes, management changes and financing.
• Historical price and volume — used for context only, never as a claim about what comes next.
• Sector classification — a published taxonomy used to place a company beside comparable public filers.

It does not read paywalled research, private data, social posts or order flow. It does not ingest a brokerage account, a portfolio or identity documents.

What the model does

The model is a language-model pipeline with retrieval and comparison steps. It extracts, aligns and summarizes. It does not value a company, rank securities or decide what a reader should do.

• Align — pair a new filing with the prior version and locate the comparable sections.
• Detect and flag — mark additions, deletions, numeric changes and redefinitions in the text and tables, and flag volume anomalies against a trailing average.
• Retrieve — pull the exact passage behind each mark, with its section and page coordinates.
• Summarize — write a short digest in plain language, quoting the source rather than paraphrasing away the detail.
• Contextualize — attach price, volume and peer figures from the same period.

Each output is a set of observations: what changed, where it sits, what the source says, and which limits apply. It is not a rating, a score of future returns, or an instruction. No accuracy claim is made or published; the checks in the next section are about traceability, not a guarantee.

What it does not do

• It does not set a target price.
• It does not give buy, sell or hold calls.
• It does not personalize output to a reader’s holdings, taxes, time horizon or goals.
• It does not state where the market or a security is going next.
• It does not manage money, execute orders or connect to a brokerage account.
• It does not rank securities by expected return.

If a question needs one of those things, the honest answer is that this tool is the wrong instrument.

Validation

Outputs are checked against the source and against a sampled review. The checks are designed to catch a class of errors; they do not make an output complete or correct in every case.

• Source check — every generated claim must resolve to a passage in the cited document. If it cannot, the claim is removed.
• Sample review — a sample of digests is read against the source filing before publication. Corrections are dated.
• Error reports — readers can email [email protected] with the output date and ticker. A confirmed error is corrected and noted in the archive.
• Regression set — a fixed set of past filings is re-run after model or prompt changes to catch drift.

Known limitations

These are the failures we know well enough to name. The list is not exhaustive, and a clean output is not a guarantee that none of them applies.

• Fiscal-period changes — companies change definitions, and comparatives may not line up.
• One-off items — impairments, litigation and gains or losses can distort a trend.
• Halted and low-liquidity securities — price and volume context can be thin, stale or missing.
• Source delay — filings and market data arrive at different times, so an output can mix periods.
• Restatements — a later filing can replace an earlier one. The archive keeps the version that was used.
• Classification — peer sets are approximate, and companies can sit across more than one sector.
• Extraction — tables, footnotes and exhibits can be misread. The cited passage is the check a reader can run.

No performance claims

Apex does not publish return figures, accuracy scores or benchmark comparisons. It does not present past output as evidence of future accuracy, and it does not claim the model can tell you where a price is going.

All financial figures on this site are illustrative layout values or figures cited from a public document. If you see a performance claim attributed to Apex, it did not come from us.

Method version 1.0 · reviewed June 11, 2026