How we work with data

The numbers on these pages are not scraped from someone else’s web page: they live in an archive of our own, fed every day with official exchange closes and with the accounts companies file. Prices, dividends paid, annual and quarterly accounts, share counts: every item enters the archive with its date, and from there it does not move again.

That is why the pages can show a long history — twenty-two financial years on a single company — instead of a snapshot of today. A number that sits in an archive can be compared with the one from the year before; a number collected on the spot cannot.

The lessons of the Trading Encyclopedia are built on the same archive. The charts you click through in a lesson are real sessions of real US stocks over a fixed historical window, not drawings made to fit the text.

The last close available while this page was being written is 18 August 2026. Every page carries its own date, which can differ if an instrument did not trade.

The first public pages went out on 12 August 2026; the per-company pages on dividends and quarterly results followed on 14 August 2026, and the commodity pages on 17 August 2026.

The data pages rewrite themselves every night. They are not documents written once and left alone: they are the output of an automated run that, after the close, reads the archive again and regenerates all 242 public pages one by one — the company pages, their dividend and quarterly-results pages, the dividend calendar, the Borsa Italiana year page, and the commodity pages such as the gold price and the oil price.

The run happens in the early morning, once the prices and the events of the previous session have entered the archive.

A regenerated page is not published merely because it was produced. First it has to pass a set of automated checks: that the numbers are present and consistent with each other, that the dates are there, that the structure of the page is complete.

Above the checks on the single page there is a watch over the whole run, and it has two hard rules. If the pages produced are fewer than expected, the run stops. If they are fewer than in the previous run, the run stops always: a page that disappears is a regression, not a mishap.

When the run stops, nothing goes out and yesterday’s pages stay online. That is a deliberate choice: data one day old, with its date in plain sight, does less harm than data that is new and wrong.

Some things sit in the archive exactly as they are: the closing price, the amount of a dividend paid, the revenue of a financial year. Others we calculate ourselves from those, and when we do, we say so.

Dividend yield, for instance, is the ratio between what a company has indicated for the next twelve months and the closing price: it changes every day because the price changes, not because the dividend changes. Cash coverage compares that same dividend with the cash the business actually generates. The surprise on a quarterly report is the difference between the published result and what analysts expected.

Every measure of this kind is explained on the page where it appears, in the words needed to understand it — not as a formula left lying there.

On these pages you will never find “current price” on its own. You will find the close of a given day. This is not pedantry: a financial number without its date cannot be verified, and a number nobody can verify is worth little.

It works the other way round too. If a measure covers a window — six months of visits, five years of dividends — that window has to be compared with the date on which the thing being measured was born. A window older than the object it measures says nothing, while appearing to say a great deal.

That is why the dates are written so a program can read them too, not just a person.

These pages are composed with the aid of artificial intelligence systems, under human direction and review. We declare it at the foot of every page, as required by Article 50 of EU Regulation 2024/1689, applicable from 2 August 2026.

What matters is what it does and what it does not do. The numbers are not generated: they are read from the archive and shown with their date. Artificial intelligence is there to build the page and to write the explanations around the numbers, not to produce the numbers. A value that is not in the archive does not appear: a dash appears in its place.

Editorial responsibility for what is published remains with the data desk.

We do not give recommendations. These pages are informational and statistical. They are not financial advice, they are not personalised, and they take no account of the reader’s situation. No line on these pages tells anyone what to buy or sell.

We do not publish forecasts of our own. Where expectations appear — for a coming financial year, for example — they are the expectations stated by analysts, and they are presented as such.

We do not quote real-time prices on these pages. What you read is the close: the figure that is final at the end of the day. Anyone who needs the movement during the session finds it in the platform, which is a different thing from these pages.

We do get things wrong. An archive holding twenty-two years of accounts on dozens of companies contains errors, and sometimes the error is ours: a measure built badly, a source read the wrong way, a missing value treated as a zero instead of as a gap.

When we notice, we correct at the source, in the archive, so that the correction reaches every page on the following run. We do not patch the single page by hand: that would be a fix that disappears the next night.

If you find a number that does not add up, tell us. The conditions for reusing our numbers are in the license.

The lessons of the Trading Encyclopedia are free to read, and the platform is where the charts, the screening and the comparisons live. Everything on this page applies to both.