First Source of Truth: Where a Number Comes From Is Part of the Number

Knowledge Bank · Shelf: Financial Analytics

A number without a provenance is only half a number.

That sentence is the whole argument, and everything below is just the working-out. When we read "revenue: 4.2 billion," the figure itself is the visible half. The invisible half is the chain that carried it: who first recorded it, under what obligation, on what document, at what moment. Cut that chain and what remains is not a fact with a missing footnote — it is a shape that resembles a fact. It can be copied, reformatted, resold, and re-copied until the tenth copy looks more authoritative than the first simply because it is everywhere. We treat the origin as part of the number, not as metadata attached to it. This article explains how that principle is wired into everything we publish, and what it costs us to keep it.

What "first source" means

By first source we mean the entity that originates the datum and is accountable for it: the regulator that receives the filing, the exchange that runs the market, the central bank that sets the rate, the statistics office that conducts the survey. The first source is not the aggregator that reprints it, the terminal that redistributes it, or the API that wraps it. Those are copies, and copies drift — through restatement, revision, rounding, lag, and plain transcription error. Each hop away from the origin adds a place for the number to change without anyone noticing.

So our research is wired only to first sources. Every dataset we carry is connected to its origin and named on the page where it appears. That naming is not decoration. It is the difference between a number you can audit and a number you can only trust.

The four verdicts

Not every first source is usable by a small, ad-free research operation, and pretending otherwise would be the first dishonesty. Our Source Registry Matrix — built by verifying official regulator, exchange, and statistics-office documentation market by market — assigns every source one of four verdicts, in plain terms:

Wire-now. Free, documented, usable. The source publishes official data with no key and no blocking terms, and we connect to it directly. US company filings are the working example: SEC EDGAR is official, requires no key, publishes public data, and is wired and in production.

Wire-with-key. Free, but requires registration. Japan's EDINET is official, free, and gated by a registration step; the registry's rule is that we register for the key and wire the source directly. The key is an access mechanism, not a paywall, and the source remains the origin.

Hard. Free to read, but with no lawful or documented machine route — or terms that block the use. This is the category most operations quietly paper over, and the one we insist on naming. More on it below.

Out of scope. Paid-only data. Some official feeds are commercial products: the exchange data hub in Israel, Japan's paid TDnet API. Where the first source is paid-only, that market or dataset is simply out of scope for us. We do not substitute an aggregator for it and we do not work around it.

Two further rules keep the registry honest. Aggregators of the yfinance class are development backup only — labelled non-primary wherever they appear, and never production. And anything not established on an official page is marked UNVERIFIED. We never infer a source's status; we verify it or we flag it.

The two hard cases

Lawful but not machine-readable. Canada's SEDAR+ portal is official. It is the first source for Canadian company filings. It is also, by its own terms, closed to automated searches, bots, and scraping. For a human reading a document, the source works. For a machine pipeline, there is no lawful, documented route in — so the registry verdict is hard. The temptation at this point is obvious: find a third-party mirror that has already scraped the filings and pull from there. That would convert a licensing problem into a silent provenance problem, and the page would look complete while resting on a copy of unknown fidelity. We do not do it. The gap stays visible.

Paid-only, and no substitute. When an official feed is a paid product, the honest options are to pay or to go without. We go without. The Israel exchange data hub and Japan's paid TDnet API are out of scope — not because the data is unimportant, but because the only lawful route to the first source runs through a commercial licence we do not hold. Filling that space with an aggregator's version would mean publishing a number whose chain of custody we cannot show. A reader would have no way to tell the difference between our wired data and our borrowed data unless we told them — and the whole point is that they should never have to wonder.

The gap rule

The companion rule for reports is short: if a source does not carry a number, the report shows the gap and names it. Missing stays missing. We do not zero-fill, we do not interpolate, and we do not let a blank quietly inherit a neighbour's value.

Why is an honest gap worth more than a filled one? Because a filled gap is indistinguishable from a fact until it is audited, and by then it has propagated. A named gap, by contrast, tells the reader exactly where the edge of the evidence is — and the edge of the evidence is information. It tells you what could not be established, from which source, and why. That is a finding, not a failure. A report that shows its gaps can be trusted precisely where it does not show them.

What it costs, and what it buys

The cost is coverage we do not claim. There are markets we cannot wire and datasets we cannot carry, and we say so on the page rather than papering over them. A research operation that refuses to substitute aggregators for paid sources, and refuses to scrape sources that forbid it, will simply know less than one that does not care. We accept that. The registry's verdicts exist to make the limits explicit rather than invisible.

What it buys the reader is narrower and, we think, worth more: every number we publish can be walked back to where it was born. Not to a terminal, not to a wrapper, not to the tenth copy — to the regulator, the exchange, the central bank, or the statistics office that first put it on the record. You can follow the chain yourself, check the origin, and decide what the number is worth to you. That is the entire service: not conclusions handed down, but numbers you can trace, gaps you can see, and the provenance kept where it belongs — attached to the number, as part of it.