Ask a due-diligence question about a company and get back the exact passages from its NI 43-101 technical reports that address it, ranked by relevance — the source text itself, not a summary of it.
Select a company first.
A technical report is the primary source for everything material about a mining project, and it is long enough that nobody reads it end to end. The consequence is that most due diligence is conducted on secondary material — the corporate presentation, the press release, the analyst note — each of which is a selection made by someone with a position.
This tool goes to the primary source. You ask about metallurgy, recovery rates, permitting risk, capital cost or anything else, and it returns the passages of the technical report that address the question, ranked by how well they match.
It deliberately does not write an answer for you. What comes back is the report's own text, with its own hedges, qualifications and caveats intact — because in due diligence the qualifications are frequently the point, and they are exactly what a summary removes.
Use the vocabulary of the report. Technical reports say “metallurgical recovery” rather than “how much gold do they actually get out”. Retrieval matches against the document's language, so phrasing your question in it materially improves what comes back.
Ask about one thing at a time. A question combining metallurgy, capital cost and permitting retrieves passages that partly match all three and fully answer none.
Read the ranking as relevance, not importance. The top passage is the closest textual match to your question. It is not necessarily the most significant thing the report says on the subject, and the fourth result is often where the caveat lives.
Questions that consistently repay asking: what recovery rate was assumed and on what test work, what the cut-off grade is and why, what the report says about permitting status, what infrastructure the project depends on, and what appears in the risks section. That last one is the most under-read part of every technical report.
Processed technical reports are split into passages and indexed. A question runs a hybrid retrieval — vector similarity to catch meaning and keyword matching to catch exact terminology — and the results are ranked before being returned. There is no language model summarising the output. What you receive is the document's text, selected and ordered.
This tool retrieves and ranks passages from the report and stops there — you read the document's own words. The Analyzer takes the same retrieved material and has Claude synthesise an answer in prose. Use this one when you want the source text with its qualifications intact; use the Analyzer when you want a direct answer to a specific question.
The ones about assumptions rather than headline outputs. What metallurgical recovery was assumed and on what test work, what cut-off grade defines the resource and why, what the permitting status actually is, what infrastructure the project depends on, what metal price the economics assume, and what the report itself lists under risks. The risks section is the most under-read part of most technical reports.
Usually because the phrasing does not match the report's vocabulary. Technical reports use precise industry language — 'metallurgical recovery', 'strip ratio', 'cut-off grade' — and retrieval matches against that. Rephrasing in the document's terms, and asking about one topic at a time, usually fixes it.
Verify anything material against the original document, particularly figures from tables. Resource tables and sensitivity analyses are the hardest content to extract cleanly, and a passage can also arrive without the qualification that sits in the adjacent paragraph.