What Would Count as Benefit

Every noun is working overtime and none of them brought a number. Ask not whether the promise is true. Ask what would settle it, and who agreed.

Two episodes of mechanism, and the other half of the ledger is still sitting there. Every large facility proposal arrives with benefits: jobs, skills transfer, downstream industries, catalytic growth for the digital economy. The SAHRC submission's complaint about these claims was that they arrive unsubstantiated, with no number attached. That complaint is correct and stops one step short. The deeper problem is that the claims are not written in a form a number could attach to.

Consider what "downstream industries" would have to mean before it could be true or false. Which industries, over what horizon, attributable how, against what counterfactual? The city grows either way, so what does the facility add to the growth that was coming anyway? The claim as written cannot be tested, and a claim that cannot be tested is a promise-shaped object with no promise inside it.

The asymmetry, stated once

The asymmetry deserves stating exactly once, without indignation, because it is a design property of the plant rather than a moral failing of anyone who builds it. The capital footprint is enormous; the operational employment footprint is small and highly skilled. That is what the technology is. Criticising a facility for not employing people at factory ratios is like criticising a library for not being loud.

The asymmetry becomes a public problem at exactly one point: when the small, skilled, capital-intensive reality is bundled into an application as "jobs and downstream industries," in a register calibrated for labour-intensive development, where proportionality has to be demonstrated rather than asserted. The claim fails review. The plant never had to.

So the useful question is the one this practice asks of every vendor promise: what would count as evidence of this claim, and who agreed to that standard in advance? A benefit claim that belongs in a record arrives with its falsification conditions attached. This many permanent positions at this band, this local procurement commitment, audited on this schedule, with this consequence if it does not materialise. Everything else is a forecast, and forecasts do not bind. Where other sectors in this country were eventually made to convert intentions into conditions, binding local commitments and audited spend and enforceable obligations, the conversion was learned late and expensively. The lesson is available now at a discount.

There is a worked example, and a data centre now sits on it. Wisconsin agreed in 2017 to subsidies approaching three billion dollars against a promise of thirteen thousand manufacturing jobs. About a thousand arrived. The state renegotiated the subsidy down, which is the part usually reported as the system working. The part that is not is what the village and the county had already committed to land, water, sewer, power and roads before anyone renegotiated anything, leaving a municipality of twenty-eight thousand people carrying debt larger than its entire operating budget. The hyperscale facility built on that ground since stands on infrastructure a failed promise paid for.

Which is the second thing a benefit claim has to survive, and it has nothing to do with whether the claim is true. It arrives into a room where the last one was not. Organisers in those communities answer the jobs figures, the grid commitments, the water assurances and the tax projections with the same four words: I don't believe them. Falsification conditions are worth more to the applicant than to anybody else in that room, because a claim that can be checked is the only kind still available to someone who has been here before.

What the pausing jurisdictions actually did

The comparators are more encouraging than the sector's defenders allow and less damning than its critics hope. Singapore paused new data centre construction in 2019 because resource draw had outrun planning assumptions, then used the pause to work out what it would accept, and reopened on conditions. Ireland stopped issuing new grid connections to data centres in 2021, once the load had reached the limits of what the system could promise, and reopened four years later on terms that had not existed before: developers bringing their own generation, and most of their electricity from new domestic renewables inside six years. The Dublin region stays constrained regardless. New South Wales moved large data centres into the assessment track it reserves for state-significant development, triggered on capital investment rather than on what the building resembles, and that track carries a social impact assessment as a matter of course. A classification decision did the work there too, in the opposite direction. Several of these jurisdictions remain, after all of it, exactly the global hubs they were trying to stay.

A pause is a process for deciding what you will accept. Treating it as a prohibition is a way of avoiding the decision it exists to make.

Who Gets an Advocate asked who is resourced to interrogate a claim and who receives it as a fact. Here that question has a specific shape. Benefit claims arrive in documents that only the approving authority and the applicant's own advisers are positioned to test, and neither is adversarial about them, because benefit claims are the part of the file nobody is paid to doubt. The Persuasion arc's finding holds at municipal scale: assertion bandwidth vastly exceeds verification bandwidth. And the ten questions in the vendor-interrogation library gain an eleventh. What evidence would settle this benefit claim, and who agreed to that before approval?

Until a claim has an answer to that question, it is not in the record. It is in the brochure.

These notes come out of Sociable Systems, a practice that reads AI-shaped documents the way a hostile reviewer will, before a lender or a court finds the gap. The argument has an operational form: the Interim Protocol sets out four rules for AI use in environmental and social deliverables, covering disclosure at touch-point grain, evidence custody, the phrases no automated screening may settle, and a hostile read before anything ships. Free, and written to be cited or retired once institutional guidance arrives.