Ask a question about any technical report in the database and get an answer drawn from the report itself, with the source passages attached — rather than reading three hundred pages to find one number.
Try asking:
An NI 43-101 technical report is the most authoritative document a mining company produces and the least likely to be read. They run to hundreds of pages, they are written by geologists for regulators, and the handful of numbers an investor actually wants — the resource table, the NPV, the metal price assumption, the capital cost — are buried at unpredictable depths.
The result is that most investors rely on the company's own summary of its own report. That summary is accurate, in the narrow sense that it does not contain falsehoods, and it is also written by people with an interest in which figures you notice.
This tool lets you interrogate the report directly. Ask a specific question and get an answer assembled from the relevant passages, with the source material shown, so you can check what the report says rather than what the press release said it says.
Ask narrow questions. “What gold price did the PEA assume?” retrieves cleanly. “Is this a good project?” does not, because there is no passage in the report that answers it.
Name the company and the report. Many companies have several technical reports across several projects and several years. Specifying which one you mean avoids answers assembled from the wrong document.
Ask for the assumptions, not just the outputs. An NPV is a function of the metal price, discount rate, capital cost and recovery assumed. The headline figure is far less informative than the inputs that produced it, and the inputs are what you should be testing.
Read the cited passages. The answer points at the source text. For anything you intend to act on, read the passage rather than the summary of it.
Questions that reliably repay asking: what cut-off grade was used, what recovery rate was assumed, what the sensitivity table shows at a lower metal price, what proportion of the resource is inferred, and what the report lists under risks.
Technical reports are processed into text, split into passages and indexed. A question triggers a hybrid retrieval — vector similarity for meaning, keyword matching for exact terms, then re-ranking to order what was found. The best-matching passages are passed to Claude as context, and the answer is generated from them rather than from the model's own knowledge.
The limits worth knowing:
Canada's National Instrument 43-101 is the standard governing how mining companies disclose technical information about mineral projects. A technical report prepared under it must be authored by a qualified person and covers geology, drilling, sampling, resource estimation and, where applicable, the economic study. It is the document that turns a company's claims about its deposit into something with professional accountability attached.
Only those whose technical reports have been processed into the database. Coverage is not universal, and a company with no filed technical report — which includes many early-stage explorers — has nothing for the tool to read.
Tables are the hardest part of a technical report to extract reliably, because the information is dense and the layout complex. Figures drawn from resource tables and sensitivity analyses are the ones most worth checking against the original document before you rely on them.
The ones about assumptions rather than outputs. What metal price the economic study assumed, what discount rate produced the NPV, what cut-off grade defined the resource, what recovery rate was modelled, what proportion of the resource is inferred, and what the report itself lists under risks. The headline NPV is a consequence of those inputs, and the inputs are where the judgement lives.
No. It reports what the document says. A technical report is prepared for the company, and an optimistic metal price assumption will be relayed as faithfully as a conservative one. Judging whether the assumptions are reasonable is the part that remains yours.