Our approach

How a figure
gets onto the page.

These are not aspirations. They are the rules the software enforces, and the reason a report can go in front of a board without a disclaimer slide.

Every figure carries a source

Each source the engine touches is registered once, numbered, and given a publisher, a title, a category and the time it was retrieved. Figures refer to that number. A report's data cutoff is not a date we chose — it is the retrieval time of the sources it used, and Section 21 lists every one of them.

Sources are ranked, not counted

A weaker second source that agrees with the first is not corroboration; it is comfort. When two sources conflict the stronger one wins — a filing over a press release, a press release over a profile page, a company's own disclosure over a third-party summary — and the report shows which one the figure came from, so you can disagree with the ranking rather than with a black box.

What cannot be measured is declared, not estimated

Every report ships a limitations list, generated from what the run actually could and could not do rather than written by hand. If a platform required an authenticated API we did not have, the report says the metric was not measured. It does not scrape an approximation and label it an estimate. The same applies inside the scorecard: a category with no usable evidence is marked N/A, its weight removed and the remaining weights raised.

The caveat goes in the data, not in a sentence

A caveat in prose is a caveat that was skimmed. So an unmeasured category is a null in the data, where it changes the arithmetic and appears in every view of the report; and a Monitor source that needs a platform key says Needs key on the card itself rather than quietly showing nothing.

The same rule governs these pages

Every number on this site names where it was read and when. The listing counts come from the SEC's own ticker-exchange file, counted on 29 October 2026; the report figures come from the sample shipped with this application. Where we wanted a figure and had no source for it — pricing, customers, revenue, analyst-coverage statistics, time saved against a manual baseline — we left it out rather than ship a plausible one. That includes the index in our own positioning: the engine does not detect Russell 2000 membership, and the coverage page says so in as many words.

What the research is not

It is not investment advice and makes no recommendation to buy or sell anything. It is an audit of what the public record says about a company, including what the company itself has published and how findable it is. The valuation section compares multiples against a peer set built from filings; it does not set a price target.

Known limits

What the engine does not do today.

Published because you will find them anyway — and because a product that hides its limits has told you something about the rest of its numbers.

Social platform metrics

Follower counts, posting frequency and engagement on X, LinkedIn, Facebook, Instagram, YouTube and TikTok need authenticated APIs. Reported as not measured.

Press-release bodies

Release analysis uses headlines, publishers and dates. The bodies are not read.

Proxy statements

Board independence, committee membership and director biographies live in the proxy, which is linked but not parsed.

Segment detail

Segment revenue, customer concentration and backlog are not available from SEC company facts and are not reconstructed.

AI platform coverage

Only Claude is queried, and only when credentials are configured. Other AI platforms are listed as not tested rather than assumed.

Index membership

No index file is read. The engine cannot tell you whether a ticker is in the Russell 2000, the Nasdaq-100 or any other index.

Limits as published in Infiniti Signals' own documentation and in the meta.limitations block of its sample reports.