From the filings up.
How we underwrite a mining royalty: net asset value rebuilt from the primary filings, an independent read of the feasibility case, and provenance an investment committee can audit.
A royalty is a claim on someone else's mine plan. The holder carries none of the operating cost, the sustaining capital, or the reclamation, and carries the one question that decides the trade: what the attributable cash flow is worth, and whether the case in front of them is honest. The royalty sector answers that question in a single language. We built a system to answer it from the source documents, and to answer it faster and more independently than the study the seller provides.
The sector runs on P/NAV
Royalty and streaming finance is priced in net asset value. A precedent deal is read as the P/NAV paid, meaning the consideration divided by the discounted cash flows of the attributable royalty. Trading comparables are read the same way, and the pattern is stable: the public royalty and streaming companies trade above net asset value, while the operating miners trade below it. Capital is scoped on that difference, and it is measured in net-asset-value-per-share accretion rather than earnings.
A counterparty that works in NAV can be engaged on its own terms. The engine was built to work in NAV from the first line, against the standard the strongest desks hold rather than an easier one.
Primary sources, not a black box
Net asset value is rebuilt from the filings rather than bought from a vendor as a finished number. The engine reads the authoritative record directly: technical reports, financial statements, corporate and ownership filings, permit and regulatory records, and the sector's published price decks. From a technical report that runs several hundred pages it extracts the resource, economics, and risk data a NAV depends on, in minutes. Every external figure carries its primary-source citation. Any figure that is not technical-grade is labeled as modeled or indicative rather than presented as precise.
The discount rate is applied the way the sector applies it. A royalty carries no operating cost, no sustaining capital, and no closure liability, so its cash flow is discounted at a lower real rate than the mine that produces it, and precious-metal streams are held to a tighter rate than base metals. Optionality such as area-of-interest rights, mine-life extension, and exploration upside is valued on a separate line and kept out of the headline P/NAV. Streams are normalized to gold-equivalent terms. Comparables that mix deal structures are marked as such rather than blended into one figure.
An independent read of the case
A feasibility study is prepared to secure financing. An underwriter has to form a view against it. The engine does that. It ranks candidates and states a position rather than returning a summary. On a single transaction it confirms the structural strengths that make a deal bankable and surfaces the analytical dimensions a deal team rarely prices at speed, and it does so in the time it takes to read the executive summary.
What one underwrite returns
- Six structural strengths scored against the record, as categories every underwrite covers: financing-stack quality, sponsor capital at risk, construction status, legal and permitting posture, demand structure, and policy exposure.
- Seven analytical dimensions surfaced that a deal team rarely prices at speed: network de-risking value, a channel and sourcing premium, cross-portfolio coherence, downstream integration optionality, a regulatory hedge, an option value on any embedded timing right in the deal terms, and a sponsor default probability read across macro regimes. The last is offered as a reconciliation and sensitivity surface for the committee, not as a forecast.
- A ranked position, with the reasoning attached, so the committee argues with a view rather than assembling one from scratch.
Why the figure holds up
No language model is used in any valuation. Each valuation engine is deterministic, so identical inputs return an identical result. Each one is versioned and records a hash of both its own code and its inputs, so any figure in a memo can be reproduced and traced to the version that produced it. Fabricated identifiers and numbers are prohibited by rule. When the audit trail catches an error, the correction is recorded rather than quietly overwritten.
The purpose is defensibility. A figure that can be reproduced and sourced is one a board can stand behind. A figure that cannot is an opinion in a suit.
The validation program
The engine was tested before any claim was made for it, and the program was built to be falsifiable rather than to demonstrate well.
How a claim is earned
- Pre-registration. The hypothesis and its scoring were fixed and locked before any evaluation ran, so nothing could be fitted to a known answer after the fact.
- Leakage control. A capable model already knows how well-known deals ended. The test set is therefore transactions that closed after the model's knowledge cutoff, which is the cleanest out-of-sample available, and every case is checked for re-identification before it counts.
- Calibration first. Before it was trusted on live deals, it had to tell genuine bargains from value traps on controls, to establish that it discriminates rather than simply declines everything.
- The right yardstick. Scored on risk-adjusted return against a hurdle rate over an observed window rather than raw return, and benchmarked against the best human judgment available.
Across that out-of-sample set, which spanned the range from cheap to expensive, the underwrite fell within or above the human benchmark on each case, and the leakage controls held on every one. It approved genuine value and declined the rich and the operator-favorable for reasons that did not depend on the prevailing price.
The character of the output
The result reads as a consistent kind of analyst: a value underwriter that works with a margin of safety. It will not extend a temporary price spike across the life of an asset, and it asks for a cushion before it commits. It reports the drivers that actually move the answer, among them resource confidence, by-product dependency, jurisdiction, step-downs, permitting conditions, and price leverage. It is not reflexively cautious either. When the net asset value is genuinely low, it turns more constructive than the desk.
The compounding part, and the actual moat
The engine values a deal and keeps reading. Each day it monitors the peer companies' filings, extracts every new royalty and streaming transaction along with its multiples, and refreshes the comparable set the next underwrite draws on. Every evaluation is written to a decision log with outcomes tracked forward at set horizons.
The defensibility follows from that, and we are precise about where it sits. The finance itself is standard, and we say so plainly. The advantage is the assembled system: direct primary-source ingestion, deterministic computation, a citation on every figure, and a calibration and audit history that accumulates with use and cannot be reconstructed after the fact. It is produced by running, which is the one part a competitor cannot compress.
Honest about the frontier
A figure is trustworthy in proportion to how clearly its limits are stated. The engine separates fact from hypothesis in its own output, labels what is modeled, measures itself against the standard the strongest desks hold, and names where it is still developing. The current frontier is reading authoritative resource statements straight into net asset value at the fidelity established firms reach by hand. That work is underway on the same terms as everything above it.
The royalty sector answers one question, in one language. We answer it from the source documents, state a position, and show each step from the filing to the figure.
See it run on your mandate.
ICHOR deploys one industry at a time, forward-deployed and tied to outcomes. Every inquiry is read personally.
Request access