Almost everything here is computed from a primary source — the company's own filing, the Treasury's own curve, FINRA's own file — rather than taken from a vendor's restatement of it. That is the easy half. The hard half is what the arithmetic is not allowed to say, and that is most of what this page is about.
A tool that always produces a number is easy to build and is wrong in ways nobody can see. Each of these is a case where the arithmetic would happily return something and the honest answer is nothing. Every one is a real correction, with the company that caused it.
A stock split is not a dividend cut
NVIDIA's effective rate reads $0.59, $0.16, $0.02 across its 4-for-1 and 10-for-1 splits. It has never cut its dividend. Splits are detected from the share count and undone.
A merger is not a split
The first version of that fix “undid” Realty Income's 1.5x jump in 2022, which was the VEREIT merger, and rewrote eight years of correct figures. A split now needs a whole-number ratio AND cash that did not move with the count.
No credit rating is guessed
Refinancing cost depends enormously on investment grade versus high yield — 5.8% against 7.7% today — and the filings do not carry a rating. Both ends are shown and the arithmetic is done twice.
A partial debt schedule gets no comparison
Coca-Cola does not tag its “after five years” bucket, so its interest over a short schedule reads 10.4% and the page concluded it would refinance cheaper than it borrows. An overstated rate can only ever reassure.
A grant is not an insider buying
An RSU vesting is payday, not somebody deciding the stock is cheap. Only open-market purchases count, which is the whole reason the insider feed is worth reading.
One purchase filed by two people is not two buyers
A fund and its designated director file the same shares separately. Counting both doubled the money and invented a cluster — the one fact that page exists to detect.
A CUSIP change is not a decision
Carnival's restructuring had seven managers “closing” it and the same seven “opening” it. Across twenty managers, 114 holdings were relabels rather than trades.
Almost nothing is seasonal
Testing twelve months at the usual 5% threshold finds a “significant” month in a stock with no seasonality about half the time. The threshold is divided by the months tested, and on the stocks measured nothing survives.
Negative equity has no return on equity
A company with negative book value produces a ratio that looks spectacular and means nothing. Same for a payout ratio against negative free cash flow, which prints as a comfortable-looking minus sign.
A percentage across a sign change is left blank
Growth from a loss to a profit is arithmetic without a meaning. Printing one would be worse than printing nothing.
Short interest is three weeks old, and says so
It settles twice a month and publishes about eight business days later. Every site prints it as current; here the age is in the panel header.
13F is late and longs only
Positions here are 87 to 178 days old. Short positions are not in the form at all, and an option appears as a holding — so a manager who bought puts shows up as a bull.
One method among several, and the only one on this site that produces a single number. It applies to an earnings event and to nothing else — the statements, the screener, the filings search, the macro radar and the portfolio tools each answer their own question and none of them is scored out of a hundred.
BULLISH
65-100
NEUTRAL
45-64
BEARISH
30-44
AVOID
< 30
Each layer is scored independently on a 1-10 scale, then combined using proprietary weights that reflect the relative predictive power of each signal category. The framework includes automatic adjustments: a Known Headwind Discount reduces the total score when any single layer scores critically low, while a Near-52-Week-Low Snap-Back Amplifier adds points when a beaten-down stock shows strong estimate momentum.
For every ticker scored, we pull up to 3 years of past earnings and cross-reference with actual daily price bars. This reveals how the stock actually reacted in similar past scenarios \u2014 not just whether it beat or missed, but the exact gap at open and full-day move.
Scenario Matching
Finds past quarters with similar EPS estimates and shows how they played out
Price Reaction Data
Actual opening gap and full-day move after each historical earnings report
Pattern Badges
SERIAL BEATER, CONSISTENT, MISS STREAK, BIG MOVER — auto-detected from 12Q of data
AI-Weighted
Historical patterns directly influence the AI scoring model with explicit weighting instructions
Ordered by how much of the site rests on them. The first three are free, keyless and primary: they are the filings themselves, which is why the company pages work for every SEC filer rather than for whatever a vendor plan allows.
SEC EDGAR — XBRL company facts
Every statement, ratio, share count and debt schedule on the company pages. Free, keyless, and the filing itself rather than a restatement of it
SEC EDGAR — full-text search
What companies wrote rather than what they reported: supplier and customer relationships, risk-factor changes, going-concern language
SEC EDGAR — Form 4 and 13F
Insider purchases separated from grants, and what twenty-one discretionary managers opened and closed
FRED
The Treasury curve, credit spreads, corporate borrowing rates, inflation expectations and the economy underneath them
FINRA
Daily short volume, and the twice-monthly short interest position with the age it was measured
Alpaca
Prices, bars, option chains and corporate actions
Finnhub and FMP
Company profiles, quotes and analyst estimates — the things that are not in a filing
Claude
Reads documents a reader uploads. It does not produce any figure on the company pages; those are arithmetic