How this data is built

Almost everything in the database is extracted from primary documents that companies themselves published: annual and quarterly reports, technical reports and studies, resource statements, and exchange announcements on the ASX, TSX, AIM and NYSE.

We do not re-model or re-estimate anything. If a company reports a resource of 12.4Mt at 1.22 g/t, that is what we store, on the basis the company reported it. Where a figure is a conversion rather than a restatement, such as a tonnage shown in your chosen unit system, the stored value stays the reported one and the conversion happens at display time.

Market data such as prices, market capitalisation and shares outstanding comes from a third-party market feed rather than filings, and is refreshed on its own schedule. It carries the date it was last updated.

This is the part that matters most, and it is the reason we do not publish a headline accuracy figure. Resource estimates, study economics, production figures, filings and news carry a link to the source document. Where a row has one, you can open the original and check the number yourself.

A single figure you have verified against the filing is worth more than a percentage we quote about the database as a whole. So rather than ask you to trust an average, we try to make each number checkable on its own.

Every source label means one specific thing:

• ▸Primary - the link opens the exchange or regulator’s own archive (ASX announcements, SEC, SEDAR+). The definitive copy.
• ▸Direct - the link opens the exact document, hosted outside the exchange archive - usually the company’s own website or its investor-relations provider.
• ▸Filings - the row has a source, but its stored link is an exchange template that misfiles, so the chip opens the company's filings page instead; find the announcement there by title and date. Archive - the link is an archived or legacy-host copy: it opens, but may have moved. Neither says anything about the figure itself, which was extracted and checked like every other row.
• ▸Via filings - we have not linked the exact document for that row yet. Clicking explains this and points you at the company’s own filings page and the documents we track.
• ▸Unsourced - we cannot yet point anywhere for that row, and the label says so rather than pretending otherwise.
• ▸Vendor - the value comes from a market data feed, not a filing, and carries the date it was refreshed.

Rows also carry the date behind them. When a figure is old, we label it as old rather than hiding it, because a three-year-old resource estimate is still the most recent one a company has published and that itself is worth knowing.

Extraction is automated, and automated extraction makes mistakes. We audit it by pulling samples of companies and checking every extracted field against the source document by hand, then fixing both the individual records and, where a pattern shows up, the extraction itself.

Those audits tell us about the sample, not the whole database. They are how we find classes of error, not a measurement of the database as a whole, and we would rather say that plainly than convert a sample into a headline.

Some parts of the database are more reliable than others, and it is more useful to say so than to average it away:

• ▸Coverage is not uniform. Larger and more actively covered companies have deeper, fresher records than small explorers with thin disclosure.
• ▸Resource estimates are the most carefully checked numbers we hold. Reporting conventions vary between companies, so basis (100% versus attributable) and whether a resource is inclusive of reserves are recorded as their own fields rather than assumed.
• ▸Project coordinates are approximate for some assets. Where a location was derived rather than stated in a filing, the record says so.
• ▸Insider transaction data comes from a third-party feed with known quality issues, including inconsistent currency handling. Treat it as indicative.
• ▸Some source links are still being backfilled. A row without one is not less checked, but you cannot yet click straight through to the document.

Our subscribers usually know one or two companies far better than we do. If you spot a wrong number, there is a Report button on the row: on company and project pages, on resource, economics and production tables, and on every screener row.

It captures where you were and which record you meant, so all you have to type is what is wrong and, if you have it to hand, a link to the filing that shows it. A report without a link is still worth sending.

Reports are reviewed against the source, and you receive the outcome in your Inbox. A completed review can identify a needed correction or explain why the value was left unchanged. Changes to the underlying data are tracked separately.

If something needs more than a row-level fix, or you would rather write to a person, email hello@stream-markets.com.

We intend to publish a running log of the corrections applied, including what was wrong and what changed. That log is not up yet. When it is, it will live on this page, and it will include the reports we rejected as well as the ones we accepted.