From filings
to insight.

A stock symbol is a company's shorthand identity in the market. Symbology extends that idea to the company's own words — turning the raw SEC filings behind a ticker into clear, structured, comparable intelligence.

Symbology pulls public filings straight from the SEC with the edgartools library and splits each one into its component sections — risk factors, management's discussion, business description, and the rest — often thousands of lines of dense, unstructured text.

From there, large language models distill those sections and compare them across reporting periods to surface what actually changed. That synthesis happens in deliberate stages, described below.

Symbology surfaces what changed — what a company added, dropped, or reworded from one filing to the next. It does not weigh those changes, judge them, or tell you what to do about them. You will not find ratings, price targets, or recommendations anywhere on the site, and nothing here is investment advice. Our Terms of Service set out the specific disclaimers in full.

Because every summary is written by a language model, we make the work checkable rather than ask you to take it on faith. Each piece of synthesis has a source view — its /s/ route — that shows exactly how it was produced: the system prompt, the model configuration, and the source filing text passed in as the user prompt. Any claim can be traced back to the words it came from.

See something off? If a passage looks wrong or misleading, tell us at support@streetfortress.com.

Every piece of generated synthesis carries a synthesis level — labelled L1 through L4 — that records how many steps of generation separate it from the original filing text. Each level is built only from the level beneath it, so the synthesis forms a traceable chain rooted in primary source material.

Our prompting strategy shifts deliberately as the level rises. L1 is direct and fact-oriented — a faithful distillation of a single section. L2 gathers those summaries across years and surfaces what changed and what recurred over time. L3 is editorial, weaving those trends into a readable narrative. L4 is reserved for the short leads that orient a reader before they dive in.

We are deliberately cautious about extending this chain too far. Each additional level moves further from the source text and compounds the risk of hallucination, so we stop well before the synthesis becomes untethered from what the filings actually say. To keep every statement verifiable, the full chain of generation — and the source documents at its root — stays visible, so a higher-level claim can always be traced back through the levels that produced it.

A text diff is a visual aid for highlighting the changes between two documents. By convention, the left column is the old side of the diff and the right column is the new side. Content that was removed from the old document is highlighted in red, and content that was added in the new document is highlighted in green.

On mobile devices we only show the side of the diff that best captures the change. Side-by-side diffs are best experienced on a computer or tablet.

To pair up comparable passages between two filings, we use a technique called word embedding to match sections that discuss similar topics. This isn't an exact science, and it sometimes produces lacklustre diffs — but many are genuinely informative, and the same embeddings help power our advanced search. We aim to improve the signal-to-noise ratio of this feature over time.

SEC filings are among the most valuable sources of information about public companies, and among the least accessible — long, inconsistent, and hard to compare across years without significant manual effort.

Symbology automates that work end to end, from primary sources. No third-party data feeds and no opinions layered on top: the synthesis turns editorial only where it helps a reader, and every statement traces back through its synthesis levels to the filing it came from.