Mineral prospectivity mapping: Sustainability peer-review checks, CHF 2,400—Verify or Skip

This guide evaluates whether a CHF 2,400 sustainability peer-review check for mineral prospectivity mapping merits payment. It explains the evidence required for a reproducible technical review rather than a narrative review.

The decision rule requires 3 of 3 checks: documented data provenance, an independent same-method rerun with fixed seeds and container, and reproduction of a pre-specified held-out deposit result with greenfield uncertainty.

How a CHF 2,400 MPM review should work

The CHF 2,400 should purchase a three-part mineral prospectivity mapping (MPM) audit: data provenance, model reproducibility, and sustainability-layer validity. The contract should require a written report identifying the exact files, code paths, environmental configuration, and outputs inspected. A general narrative assessment of the map, or a visual review of high-probability zones, would not satisfy this audit definition.

First, the reviewer must distinguish repeatability from reproducibility. iSixSigma’s operational distinction treats repeatability as consistency when the same analyst, data, and conditions are retained, while reproducibility—as the Wikipedia entry on reproducibility also describes—requires a different researcher to obtain the result by following the documented method. Accordingly, rerunning the author’s notebook in the author’s environment may test repeatability, but it is not an independent peer review. The review should require a different analyst to implement the same method from the delivered instructions and compare the resulting outputs with the submitted results.

Second, the reproducibility check must be mapped explicitly to the MPM entities rather than to the final map as a whole. The contract should state whether the audit inspects the occurrence labels, negative-sample construction, evidence layers, training and test split, uncertainty raster, and greenfield polygons. It should also specify how the held-out deposit assessment is performed and how uncertainty in greenfield areas is evaluated. The machine-learning MPM study by EarthByte supports examining both the spatial correspondence between modeled probability and known occurrences and the potential greenfield zones; the geochemical prospectivity-mapping work in Geological Society, London, Special Publications provides a relevant example of the kind of mapping workflow that can be audited.

Third, the sustainability layers must be traceable from delivered data and code. The reviewer should verify layer provenance, transformations, units, spatial alignment, and the exact role of each layer in training or scoring. The Nature Medicine review on reproducibility frames trustworthy research as depending on practices that are consistently implemented, which is why a verbal assurance about sustainability data is insufficient here. “Sustainable” or “environmentally weighted” labels should be accepted only when the reviewer can connect them to inspectable inputs, code, and outputs.

The final report should preserve the three-part structure and mark each component as passed or failed, with reproducible commands, environment details, inspected artifacts, and observed discrepancies. The reviewer should document the data and code needed for the rerun, identify any departures from the submitted workflow, and show how those departures affect the held-out and greenfield assessments. This makes the check falsifiable: the buyer can verify not merely whether the reviewer described the work, but whether each required MPM component was independently examined.

Evidence that separates audit from narrative

An audit must be evidenced by artifacts and reruns, not by a persuasive narrative. The distinction matters because the relevant sources describe different evidentiary thresholds. Wikipedia defines replication as different researchers using the same methodology and obtaining the results again with a high degree of reliability. A machine-learning prospectivity mapping study from EarthByte adds a domain-specific test: high-probability areas should correlate spatially with known mineral occurrences, while prospective greenfield zones are identified for further exploration. The former establishes who should rerun the work; the latter establishes what the resulting map must show.

Nature Medicine is not a mineral-prospectivity source and should not be presented as one. Its narrow role is to reinforce the documentation expectation: reproducibility is foundational, but claims of reproducibility are weak when the materials and practices needed to reproduce a result are not consistently exposed. In this context, documentation should be operationally judged: can an independent analyst identify the delivered inputs, locate the code paths that produce the prospectivity output, and determine the conditions under which the result was generated?

The convergence is therefore concrete. The delivered code and data support inspection; the independent rerun tests whether another researcher can recover the stated result using the same methodology; and comparison with known occurrences and greenfield zones tests the map’s geological relevance. A prose account of model choices, feature engineering, or environmental considerations cannot substitute for any of that evidence. If those artifacts are missing or cannot be independently exercised, the review’s conclusion is narrative rather than an audit.

Options compared: what to buy

When comparing what to buy, treat the choice as an evidentiary decision rather than a branding decision. This section compares three CHF 2,400 purchase options and names the reproducibility audit as the only acceptable buy. The decisive distinction is whether the deliverable lets an independent reviewer test the result rather than merely judge the author’s account of it. As the Wikipedia entry on reproducibility explains, replication typically involves different researchers applying the same methodology and obtaining the result again reliably.

The audit should be accepted only if the reviewer receives enough material to perform the same method independently: a documented data dictionary, the scripts and execution environment, and access to the delivered data. The rerun must preserve the review protocol, including its fixed execution settings, and the prospectivity output must be checked against both a held-out deposit and the uncertainty associated with greenfield areas. This matters because the machine-learning prospectivity-mapping study summarized by EarthByte distinguishes known-deposit performance from prospective greenfield zones; a persuasive map alone does not establish that distinction has survived reproduction.

Reject the narrative-review option when the requested deliverable is only a memo. Without the code, data, and environment, the buyer cannot determine whether the result follows from the reported procedure or from undocumented choices, manual adjustments, or incomplete inputs.

The repeatability/reproducibility distinction introduced above reinforces this point: only the audit package permits the independent reproduction test, because a memo without code, data, and environment cannot be rerun by another analyst.

Do not treat a full sustainability certificate as a substitute without identifying the governing standard, its scope, and what was actually assessed. None of the sources cited in this article names a certification standard or establishes that a certificate would cover mineral prospectivity mapping, its sustainability layers, or an independent rerun, so a reader should verify the standard, scope, and assessed scope in the certificate documentation before paying.

The purchase order should therefore specify the audit artifacts and the three-part pass condition; anything short of that leaves the buyer with an assurance product whose scope has not been established.

Costs, limits, and when the rule breaks

For this review, CHF 2,400 is a capped audit budget, not an open-ended professional-services engagement. It buys a fixed-scope pass/fail review with three deliverables: a data provenance table, a rerun log, and a held-out/greenfield prospectivity map. The contract should list those artifacts, the files and code to be examined, and the required output format. If the quote is hourly, omits the deliverables, or reserves time for recommendations beyond the audit, clarify the scope and price before paying.

Payment should depend on the rerun log being delivered as a milestone. A report that merely states that the model “was reproduced” is not enough; the log should show the run and make the evidence auditable. If none of the three checks passes, the CHF 2,400 is spent but the decision remains unverified, so another audit may still be necessary. This is the cost of a wrong verify: paying for assurance without receiving evidence that supports it. A milestone structure limits that exposure by withholding final payment until the reviewer has supplied the rerun record.

The cost of a wrong skip is different. Skipping the check leaves the prospectivity decision exposed to missing provenance, an unreliable random seed, or a sustainability layer that cannot be regenerated from the delivered materials. That risk is especially relevant where the map affects exploration priorities, because a polished output can conceal an untraceable input or a non-reproducible processing step. The practical comparison is therefore not “audit versus no audit,” but “audited evidence versus an unresolved verification risk.”

The pass/fail rule stops working in several identifiable circumstances. It breaks if the reviewer cannot obtain the delivered code and data, cannot rerun the same method with fixed seeds and a controlled environment, or cannot evaluate a held-out deposit and associated greenfield uncertainty. It also breaks if the quote is hourly, if payment is required before the rerun log is produced, or if “audit” is used to describe a narrative review without testable artifacts. In each case, the scope has drifted from a decision-grade verification.

Before authorizing payment, ask: are all three named deliverables attached to the contract? Is the full remaining payment conditional on delivery of the rerun log? Can an independent party inspect the provenance table, reproduce the run, and inspect the held-out and greenfield map from the delivered materials? If the answer to any question is no, the CHF 2,400 is better treated as an unresolved procurement question than as completed assurance.

Worked audit checklist and if/then rules

The worked audit materials below—the CHF 2,400 inputs, a copy-usable checklist, and corresponding if/then rules—are the section’s specific contribution. The worked example is method-labeled rather than a claim about any particular mapping project: assemble one MPM raster, one held-out deposit list, one greenfield polygon, one code repository, one data dictionary, and the CHF 2,400 quote. Keep each item as delivered; do not replace a missing file with a screenshot, verbal assurance, or reconstructed substitute.

Checkpoint 1 — provenance: PASS or FAIL. Inspect every evidence layer used to produce or interpret the MPM and verify that its source, date, and license are recorded. The data dictionary should connect each raster or tabular layer to its file, description, provenance, and permitted use; the repository should identify the code path that consumes it. FAIL means at least one layer lacks a traceable source, date, or license. If FAIL: do not release the payment. Require the missing provenance record and rerun Checkpoint 1; if the supplier cannot supply it, treat that layer as unverified and return the engagement for nonpayment or a documented correction.

Checkpoint 2 — independent rerun: PASS or FAIL. Have a second analyst, or the original analyst in a clean container, reproduce the map using the delivered code and data, the same method, fixed seeds, and the specified environment. Record the commands, versioned environment description, seed, output filenames, and a canonical hash of the output raster.

Apply the repeatability/reproducibility distinction defined earlier in this guide: the rerun must be an independent reproduction by a second analyst or clean container, not the original team re-executing unchanged conditions. FAIL if the map hash or seed cannot be reproduced.

If FAIL: withhold payment and require a corrected rerun package. A similar-looking map, narrative explanation, or manually adjusted output is not a reproduced hash.

Checkpoint 3 — held-out result and uncertainty: PASS or FAIL. Confirm that the held-out deposit was excluded from training and that it falls within the reviewer’s pre-specified high-probability zone. Then inspect the greenfield polygon for an explicit uncertainty statement rather than treating it as a confirmed discovery. The EarthByte MPM description reports spatial correlation between high-probability areas and known occurrences while also identifying potential greenfield zones, so the two outputs require different tests. FAIL if the deposit test was not pre-specified, the deposit falls outside the stated zone, or the greenfield polygon has no uncertainty qualification. If FAIL: return the checkpoint evidence and withhold payment pending a compliant rerun and corrected review.

Release the CHF 2,400 only when all three checkpoints are marked PASS. Record each result as PASS or FAIL, attach the supporting artifact, and avoid publishing an accuracy percentage that the delivered inputs do not establish. Any FAIL, missing artifact, or post hoc change to the held-out criterion is a nonpayment rule; three recorded PASSes are the release condition.

What to do next

Frequently Asked Questions

What minimum evidence is required before paying for the CHF 2,400 sustainability peer-review check?

The reviewer must pass all three required checks: documented data provenance, an independent rerun using the same method with fixed seeds and container, and reproduction of a pre-specified held-out prospectivity result.

What counts as an independent rerun for the mineral prospectivity mapping review?

An independent rerun must use the same method with fixed seeds and container.

Why is a narrative review not sufficient for this sustainability peer-review check?

A narrative review is insufficient unless the reviewer reproduces the specified result and audits the sustainability layers from the delivered code and data.

What must be traceable in the sustainability assessment?

The sustainability assessment must be traceable to the supplied code and data rather than rely on unsupported narrative claims.

What should the held-out evaluation test besides the held-out deposit?

It must also include an uncertainty assessment for greenfield areas, following the stated reader rule.

Which sustainability layers does the reviewer need to audit?

The reviewer must audit the sustainability layers from the delivered code and data.

Quick answers

Also worth reading: The best books for mastering spatial statistics and geospatial mapping: best books for mastering spatial · How to choose a truly pure mineral supplement: How to choose a truly · Reimagining the mineral discovery life cycle for a sustainable future in global mining: Reimagining the mineral discovery life