Arc companion

The Wrong Variable

Run a hyperscale data centre through a lender's environmental and social screen. Answer honestly, watch it come back clean, and then look at the question the screen never asked. Written for people who set and apply safeguard standards.

Sociable Systems · Standards

The Wrong Variable

Run a hyperscale data centre through a lender's environmental and social screen. Answer every question honestly. It comes back clean, and the answer is correct. Then look at what the screen never asked.

For people who set and apply environmental and social standards. It characterises a facility class and describes no particular transaction.

Safeguard frameworks size social impact by footprint, land take, workforce and receptors at the boundary. This asset is thin on all four.

Screen it yourself

Six questions, in the substance the standards actually ask. Answer as an honest officer would for a hyperscale campus on a serviced industrial lot.

Nothing on that form was answered wrongly, and nothing about it was hidden.

A reviewer reading that file finds no error, because there is none in it. Now ask the question the form does not contain: what will it draw, from a system other people are already queuing on?

The question that is not on the form

What the facility will draw from the municipal systems, and who else is drawing on the same ones. Set the numbers to whatever you think is realistic.

MVA, in plain terms. A connection is sized in megavolt-amperes. It is the most a facility is entitled to pull from the grid at any one moment: the width of the pipe, not the amount that flows through it. How much of that width actually gets used, around the clock, is the load factor, and this class runs high and flat. Everything below is stated in households as well, because a connection size is not a quantity most people can picture.

How big is one facility? (MVA connection)

How many on the same municipal system?

Each approved on its own, each correctly.

What was one application told the system could take? (MVA)

What the six of them want, together

And the same question about water

One variable decides this answer, and it moves it by three orders of magnitude.

Cooling design

Cooling design	Litres per kWh	Megalitres a year	Households' annual water

All six facilities together, at each design. The row in bold is the one selected above. An application that does not state its cooling design leaves every row of this table open.

While that is happening, what are households held to?

The conversion, and every coefficient in it

Power factor

A connection is sized in MVA. The running load in MW is a little less.

Load factor

The share of the connection actually drawn, around the clock. This class runs high and flat.

One household, a year (kWh)

An ordinary urban household. Replace it with your city's figure.

PUE

Total facility energy over IT energy. Water intensity is quoted per kWh of IT energy, so this matters.

One household's water, a year (kilolitres)

An ordinary urban household. Varies widely by city and tariff block. Replace it with your municipality's.

People per household

Used only for the restriction comparison.

Water intensities: closed loop and direct liquid with dry coolers, zero. Best in class hybrid, 0.26 to 0.30 litres per kWh. Adiabatic or hybrid, 0.2 to 0.8. Full evaporative, 1.8 to 4.0. These are published industry ranges, not figures from any application.

Households are a unit of comparison, not a claim that households are cut off to serve a facility. The point is scale: how much of a shared system one building commits, in terms a person can picture. A number you cannot interrogate is not evidence, so all three are yours to change.

Processes, anywhere, that add up this column

Not an oversight by anyone. Each application is assessed on its own site, correctly, against the headroom that one file was shown. No door, no tribunal, no lender and no register is given the total, so the trade-off it implies is never actually made by anybody. The international sector guidance published in June 2026 names cumulative pressure as a consideration and provides no aggregation instrument either.

So what. Electricity on a constrained municipal system is allocated, not merely sold. Every unit committed to one connection is a unit that is not available to the next applicant in the queue, whether that is a factory, a housing development or a hospital, and where the system cannot serve everyone the shortfall is met by building more capacity, by paying more for it, or by everyone getting a less reliable supply. That is a real distributive decision with winners and losers. It may well be the right decision. The finding is only that on this asset class nobody is currently in a position to make it, because the number it would be made on does not exist in any process.

Why this is not the usual complaint

The figures were due later

In the jurisdiction studied here the resource figures were not withheld. They were not yet due, on a schedule written for a building this is not, and the due date fell after the decision they were material to. That is a stranger problem than concealment, and a harder one to fix.

Better reports change nothing

The question set every downstream specialist inherits is fixed at classification, before anyone opens a file. Improve the studies and you improve answers to questions that were never going to reach the draw.

Where the sector instrument sits

The absence of a sector instrument here is not structural. The performance standards are sector-agnostic by design, and sector specificity lives in the environmental, health and safety guidelines, which is where an answer to this would be written. A data centre guideline has been identified as needed, inside a rolling update of roughly thirty of them, and carries no firm date. So the instrument built to give the answer has not been drafted, and the buildout is not waiting for it.

The Equator Principles Association published sector guidance in June 2026, developed with Arup, and it closes a good deal of this. It asks whether cooling is evaporative or adiabatic and, where it is, whether water scarcity has been assessed at the level of the region and the local catchment. It requires water use effectiveness annualised and at peak, read together with power use effectiveness rather than instead of it. Associated facilities are handled properly, down to substations, transmission, water supply and on-site treatment, and the guidance states in its own words that the direct footprint of a data centre is often modest compared with the extensive external systems required, which is the asymmetry at the top of this page, in their words. It names competition for water resources in water-stressed contexts. It names indirect and cumulative pressure on systems with constrained capacity. Anyone financing this class under the Equator Principles now has a document that asks most of the right questions about a single transaction.

Read end to end, it also names grid pressure twice, in the screening criteria and again on the burden falling to municipal systems. Every appearance is a question about the project in front of the reader: a threshold for one transaction, a benchmark set in line with industry and regional expectations. Nothing asks whether the system can still serve the next applicant, which is the question the households on the same feeder are living inside, and which the panel above is an attempt to make answerable.

Three questions the guidance does not ask

Raised here because in an instrument still being drafted these are choices available to the drafters, rather than omissions to be complained about after the fact.

One

Efficiency is not allocation

The guidance names system pressure, in its Category C screen and its community section, but every one of those is a question about one project alone, and its WUE question is a benchmark: in line with industry and regional expectations or targets. Nothing asks whether the system can still serve the next applicant. An EHS Guideline is structurally built around performance benchmarks, so the shape of the instrument may itself not fit an allocation impact.

Two

Nothing aggregates across facilities

Cumulative pressure is named as a consideration, on a context-specific basis, for one project's assessment. No instrument tells anyone to add up the column across facilities on one system. Each transaction is assessed correctly and alone.

Three

The benefit case is not tested

The guidance is risk-side throughout, which is right for its purpose. But employment and skills claims are what these projects are approved on locally, and nothing asks whether those claims are written in a form that could be shown to have been missed.

And a fourth, which is why this is commercial rather than restrictive. A flexibility condition. A load that can yield is worth something to a system operator, and speed to power is worth a great deal to an operator. One jurisdiction has already priced that trade rather than choosing between approving and refusing. There is an instrument between yes and no, and nobody in this market is holding it.

Why it reaches a lender now. Development finance and commercial project finance are both already inside this class in this region, at scale. The benefit case is typically unenforceable, written at a level of abstraction no number attaches to. Nobody holds the aggregate. And sustainability-linked instruments are routine in this market, yet as far as the public record shows, no transaction in this sector in this country publishes a social or water KPI at all. Since June 2026 an Equator-covered transaction has a sector document to answer to on water, which makes that silence easier to name than it was. Each is a defect a sponsor shares an interest in fixing: an approval resting on an untested benefit claim is a fragile asset, and the applicant carries that fragility for the life of the facility.

The working behind all of it

Everything above is derived from a published corpus rather than asserted. These run in a browser, send nothing anywhere, and show their coefficients so the arithmetic can be argued with.

The mechanism

The Two DoorsThe two approval pathways, what each is built to ask, and where the resource figures actually arrive in the sequence. Which Door, Which QuestionsThe classification path walked to its three termini. The question set is fixed at the fork. You Are the DoorThe same exercise from the tribunal seat. Decide on one file, then watch what a correct decision never reaches.

The numbers

What It DrawsStated capacity converted into municipal draw, every coefficient visible and editable. The Cumulative DrawSeveral facilities on one system, and which questions each door can ask about the total. The Ride-ThroughThe uptime promise as a fuel logistics operation: hours of darkness, tanker deliveries, what the fence line hears. The Prompt's AddressThe demand side, converted into the municipal account terms it has never carried.

Diligence and monitoring

The Pre-Close Exposure ScannerTwelve questions about an application file, answerable from the public record, switchable between the lender, practitioner, councillor, utility and press seats. The Diligence SheetThe same twelve questions as a scored worksheet that drops into a committee pack. The Draw LedgerPost-approval monitoring: stated draw beside actual draw, every claim beside the evidence that would show it was kept. The Covenant TestDraft the condition that would actually bind a claim, then watch three people try to get out of it. The Pre-Application Seam ScanWhich questions a chosen door structurally cannot ask, and what is worth scoping in voluntarily.

The record, and the checking of it

The RegisterOne row per project, source URL mandatory, approval route where a public record shows one and unknown where it does not. The exception count is published beside the data. The Verification LedgerEvery claim behind this work, what it was checked against, and what the check found. Including the comparators that were wrong first. The Claims RegisterThe QA tab a deliverable ships with when AI touched the drafting. A gate that will not clear while anything is unsupported. The Hostile ReadOne public document read the way a hostile reviewer will read it, before a lender, journalist or tribunal does it instead. Ten QuestionsTen questions phrased so that following the standard process cannot settle them, because in most cases the process does not require the answer. The Ground AddressThe written argument in full, episode by episode, with sources.

On the limits of this. The structural work is drawn from one jurisdiction's public record, read closely, and it is recent. Nobody has years of expertise in this asset class, because the class is younger than the buildout. What is being applied is a career of impact assessment and resettlement practice under lender standards, on projects that long predate this asset class, meeting a facility type those standards have not yet met. Where a framework already has an answer that has been missed here, being told is the preferred outcome.