Resources

How our tools work

Last updated: October 14, 2026

Both of our products show you numbers about your own money. You should be able to find out where each number came from before you rely on it. This page sets out how the figures are produced, what the methods assume, and — the part most pages like this leave out — what the tools cannot see.

How it reads a statement

Portfolio Analysis does not connect to your brokerage. There is no account linking, no credentials and no aggregator. You give it the file your broker already gives you: a positions export, or the PDF statement you download.

The file is read on your own device, by a copy of the PDF reader served from our own site rather than fetched from somebody else's. The file itself is never uploaded. The company logos beside your holdings are drawn from an image that ships with the app, not requested one by one, so looking at your portfolio does not tell a third party which companies you own.

Why an import can be refused

A statement reader that quietly mis-reads a number is worse than one that fails, because you have no way of knowing. So the app checks its own work against the statement before it will show you anything.

Every market-value total the statement prints for itself is compared with the total it gets by adding up what it extracted. If they do not agree to the cent, the import is refused and you are shown the section, the column, both figures and the difference. Nothing partial is loaded on the strength of "close enough".

You can see the whole check either way. The review sheet lists every total it compared, what the statement printed, what it extracted, the difference, and one of four verdicts:

The sheet is not editable, and it reads the same result the gate reads, so what you are shown and what the software decided cannot disagree.

Within an accepted import, a figure the app cannot stand behind is left blank rather than filled in. If a row's quantity times its price does not equal the value printed beside it, that figure is blanked and the review sheet tells you how many rows were treated that way. The same rule holds everywhere: a missing number renders as a dash. No figure is ever substituted from another holding or another period to fill a gap.

One known limit: option positions are read from Fidelity statements and by the general reader. An option row on a statement from another broker may cause the import to stop rather than be read.

What it computes

From one import, the app produces twenty-five analyses in seven groups. Each carries its own written note about what that particular analysis is not — twenty-seven separate caveats rather than one disclaimer repeated, because a sentence a reader has seen on every panel stops being read.

Company exposure adds three things that are usually reported separately: shares you hold directly, your share of the same company held inside your funds, and the net market value of any options you hold on it. They are kept apart in the display, so you can see which part of a position is direct and which arrived through a fund.

Fund look-through, and its blind spot

Looking through a fund to the companies inside it is the thing this software exists to do, and it is also the thing most likely to be incomplete. Here is exactly how far it reaches.

There are three ways a fund's holdings are resolved:

1. A set that ships with the app. Fifty-three tickers resolving to forty-three distinct funds once share classes of the same fund are followed through. Six of those are cash and bond funds recorded deliberately as holding no equities. These are point-in-time snapshots, dated January to July 2026, and the app prints that date beside them.
2. The fund's own public filing. For a fund not in that set, the app can look up the most recent portfolio holdings report the fund filed with the SEC and read it. These are filed quarterly, so they describe the fund as it was at a past date, not as it is today. When the lookup fails, it fails with a stated reason rather than silently.
3. You can tell it. Any fund it could not resolve can have its holdings entered by hand, and what you entered takes precedence over both of the above.

What matters more than the coverage is what happens at its edge. A fund it cannot open is named on screen, with its dollar value and its percentage of your portfolio, under a heading saying it will not appear in company look-through until it is taught. It is not quietly dropped, and the portfolio total does not silently shrink.

Coverage inside a resolved fund is also partial, and the two kinds of exposure behave differently because of it. Where a fund's published holdings list covers only part of its assets, the remainder is distributed according to that fund's own sector profile. Sector exposure therefore stays complete; company look-through does not. A company held in the unlisted tail of a fund will not appear by name.

What PulseAlert measures

PulseAlert is in beta and open to a small number of people. Everything below describes what it does today.

Almost everything PulseAlert shows is its own arithmetic over daily price bars. The mathematics is textbook — these are standard measures, not proprietary ones — and the point is not that they are novel but that the exact parameters are published rather than described.

Why 251 sessions and not 250 or 252: it was measured. Against published 52-week changes, a 251-session window differed by 0.56% at the median, against 3.4% at 250 and 4.1% at 252.

Conviction is a tally, not a confidence level and not independent evidence agreeing. Six of its ten inputs are moving-average pairs computed from the same price history, and there is no neutral vote among them, so one steady trend can reach the alerting threshold on its own.

When PulseAlert emails you

PulseAlert re-reads your watchlist seven times each weekday — on the hour from 10:00 to 15:00 US Eastern, and again at 16:30 after the close. The ranked market screens are produced once each market morning, before the open.

Nothing here is real-time. Every figure is computed from completed bars on a schedule, and the schedule is the one printed above.

An email is sent when a reading changes state, not while it stays that way. A stock that is overbought on Monday and still overbought on Thursday produces one message, not four. More than thirty kinds of alert are grouped into twenty categories you can switch on and off independently, and you choose whether to receive them as they happen, as a digest at the start or end of the day, or both.

You hear from it when something changes, and not otherwise. There is no "all clear" message and no daily confirmation that it ran.

A watchlist holds up to fifty symbols.

What the tools refuse to print

The most useful thing we can tell you about either product is where it stops.

• It refuses an import it cannot reconcile against the statement's own totals, and refuses to restore a saved portfolio that does not rebuild to the figure it was saved with.
• Market Mood is not drawn at all when too few stocks have been counted to make it meaningful. An earlier version of that panel once printed a confident market-wide figure computed from a fraction of the index; the fix was to return nothing rather than something.
• Price levels distinguish three different answers — that the calculation could not be run, that it ran and the result was deliberately rejected with a reason, and that levels were found. "We have not worked this out" and "this stock has no levels" are different claims about the world.
• A figure we cannot stand behind renders as a dash. We removed a fallback that had been quietly serving one well-known fund's return in place of every holding we did not recognise, and there is now a test whose only job is to stop it coming back.

Several readings in PulseAlert are still being worked on, and we would rather say so than present them as settled: the way signal strength and direction are ranked, and the fundamentals score, whose comparison points are approximate style-box medians rather than live peer groups.

Where the data comes from

Company filings, share counts and insider transactions are read directly from the SEC's own public system. Consolidated short interest is read directly from FINRA's own public API. The list of listed securities comes from the exchanges' own published directories. Prices, fundamentals and related market data come from third-party market data providers under their respective terms.

Insider activity is deliberately narrow: only open-market purchases and sales, where an insider used or received their own money. Option exercises, share grants, restricted-stock settlements, tax-withholding sales and gifts are excluded, because they say little about what the insider thinks.

Neither product is affiliated with, endorsed by or sponsored by any broker, fund company, exchange or data provider. Where those names appear in our software they identify the source of a figure; they are not partners. Our Disclaimers set this out in full.

The full reference

PulseAlert publishes a section-by-section reference inside the product, covering every measure above with its formula and its thresholds. It sits behind the sign-in because it is part of the product rather than marketing material. It is available to every account at no cost.

Portfolio Analysis does not have a single reference page. Its explanations are written beside the figures they describe: each of the twenty-five analyses carries its own note about what that analysis does not tell you, and individual figures carry footnotes where a figure needs one. The statement review sheet described above is the most detailed of these, and you reach it from any import.