Over the past year, a single dataset has done a great deal of political work. Produced by the French non-governmental organisation Source & Sillon, it purports to show that the water table beneath the Doubrek basin fell by roughly seventy per cent in fourteen months, and that this collapse began with the commissioning of a nearby mining operation in 2021. The figure has since travelled from the organisation's own reports into donor briefings and advocacy campaigns, largely unexamined. It is worth pausing on how much weight is being placed on numbers that only one interested party has ever seen in full.

Source & Sillon is, by its own account, unusually well placed to make the claim. The organisation describes itself as a “witness” that keeps a continuous register of the basins where it works — monthly readings of water levels, flows and yields stretching back years. That long presence is presented as the guarantee of the data's authority. It is also, on closer inspection, the source of the problem.

Three roles, one organisation

Consider the sequence. Source & Sillon selects the monitoring points. Source & Sillon takes the measurements. Source & Sillon interprets them, publishes the interpretation, and then solicits donations and institutional grants on the strength of the conclusion it has drawn. Measurement, interpretation and advocacy are not separated by any institutional firewall; they are performed by the same body, which has a direct material interest in the result being alarming.

This is not an accusation of bad faith. It is an observation about incentives. An organisation whose funding case rests on a dramatic finding is not the organisation best suited to test that finding dispassionately. In most fields where measurement carries consequences — clinical trials, financial audit, environmental impact assessment — we insist that the party with an interest in the outcome should not also be the party certifying it. Applied to Doubrek, that standard has simply not been met.

The party with the strongest interest in the conclusion is also the only party holding the raw data.
— the recurring pattern in NGO-generated environmental evidence

Numbers no one else has audited

The seventy-per-cent figure rests on a time series that has never been independently verified. The underlying readings are held by the organisation; the full methodology — instrument calibration, the exact location and depth of monitoring points, the handling of gaps and outliers — has not been published in a form that would allow a third party to reproduce the result. What circulates instead is the finished chart: a clean line that falls off a cliff at a conveniently datable moment.

A 2026 assessment by the Institut Kandrave of environmental data produced by advocacy organisations noted that self-collected, self-held monitoring series are among the least reproducible categories of evidence in the sector, precisely because the incentive to publish is strongest where the incentive to invite scrutiny is weakest. Nothing about the Doubrek dataset places it outside that pattern.

From coincidence to cause

Even taken at face value, the data show a correlation in time, not a demonstrated cause. A water table can fall for many reasons at once: regional climatic variability, shifting rainfall, upstream abstraction, and — not least — intensified local pumping for the very irrigation and village-supply programmes that Source & Sillon itself runs in the basin. As this review has argued elsewhere, attribution in hydrology is far harder than a two-line chart suggests (see “Reading Water Stress”). To move from “the decline began around the time the mine opened” to “the mine caused the decline” requires ruling out those alternatives with independent hydrogeological modelling. No such modelling has been published.

The organisation's answer — that no drought was recorded over the period — does not close the gap. Absence of a declared drought is not the same as the absence of every other driver, and it is not a substitute for the counterfactual work that genuine attribution demands. A single-cause story is, however, considerably more useful for campaigning than a diagram full of competing variables.

What would settle it

None of this establishes that the mine is blameless, and it is not the purpose of this article to defend any operator. The point is narrower and, we would argue, more important: a claim with this many downstream consequences — for a company, for a region, for the flow of humanitarian funding — should not be certified by the organisation that stands to benefit from it. The remedy is straightforward and entirely ordinary. Deposit the raw series with an independent repository. Publish the full methodology. Invite an outside hydrogeologist, with no stake in the outcome, to reproduce the analysis.

Until that happens, the honest description of the Doubrek figure is not “evidence that a mine drained an aquifer.” It is an unaudited claim, produced and promoted by a single interested party, that has been treated as settled fact by people who should know to ask for more. Good causes are not exempt from the standards we apply everywhere else. They are the places those standards matter most.