A Tunnel Has To Breathe. It Has To Put The Dirt Somewhere.
Four quantities a manned underground facility cannot suppress: air, spoil, power and effluent. Each is metered by somebody with no reason to keep a secret. Here is the whole method, with the four ways it fails.
Background written by the news staff. Contains no argument or recommendation.
LOFTUS, England — Three hundred people underground need 141 cubic metres of fresh air every minute.
That is the occupational minimum — 0.47 cubic metres per person per minute — multiplied by three hundred. It works out to about 0.39 square metres of clear opening, or nearer three-quarters of a square metre of grille once you allow for the blades.
It is a thing you can stand in front of and feel on your face.
This desk has run 41 rumoured underground sites through the same four checks since 2019. We have never published the checks. We are publishing them today, in full, with the arithmetic, the two calibration runs and the four ways the method fails, so that readers can stop writing to us and run them themselves.
The four quantities
Air. As above. The grille exists, it is on the surface, and it is usually the easiest thing in this article to find, because a hidden intake is still an intake.
Spoil. Rock bulks by a factor of 1.4 to 1.6 when you break it. A void of 100,000 cubic metres therefore produces roughly 150,000 cubic metres of muck, which is about 10,000 lorry loads, on a public road, under a permit, weighed by somebody paid by the tonne.
Power. Fans and chilled water dominate: 6 to 9 kilowatts per occupant. Three hundred people is 1.8 to 2.7 megawatts. A connection that size appears on a distribution network operator's register, which is published quarterly because planners need it.
Effluent. A hundred and fifty litres per person per day. Forty-five cubic metres a day, arriving at a treatment works that meters it and bills for it.
Calibrating on two places nobody is hiding
A method you have never tested is a preference. We tested this one on two sites that publish their own numbers before we pointed it at anything rumoured.
The first is a working potash and polyhalite mine on the North Yorkshire coast. The second is a metro deep-tube ventilation scheme whose environmental statement gives design airflow, shaft dimensions and excavated volumes broken out by disposal route.
Air came within 9 per cent at both. Spoil came within 4 per cent at the metro scheme, because there are tickets, and tickets are the best evidence in this article.
Power we got wrong by 22 per cent, and we got it wrong in the direction that matters: we under-predicted, because our model does not hoist ore. Effluent we got wrong by 31 per cent at the mine, where the discharge is dominated by groundwater and has almost nothing to do with people.
Both misses are in the published working, at the same size as the hits.
Where the method breaks
There are four ways to defeat it, and readers should know them before they run it.
Tip the spoil into a worked-out void — an old mine, a quarry backfill — and it never touches a road or a permit. Build small enough that the load hides inside an existing building's own services, and the connection register shows a shop. Dump the heat into a river under an abstraction and discharge licence, and the electrical load halves. Sit on a septic tank, and the fourth check does not exist.
Our engineer adds a fifth, which we accept: anything built before about 1962 predates most of the registers we are reading.
There is also a sixth condition the method cannot address at all, which is a drawn but unbuilt level. The mechanical sheets in PX-1402 dimension three of them. No amount of grille arithmetic distinguishes a room from a room that was cancelled in 1993.
What has actually retired sites
Of the 41, the check that has closed the most is not the satellite pass. PX-1912 cost this paper a great deal of money and settled one question. It is not the spoil arithmetic either, which is slow and needs an archive.
It is effluent, and the work is done for free, monthly, by a trade-effluent clerk at a water company, as routine billing.
The commonest finding is not absence
This is the part we did not expect when we started keeping the tally.
Of the 41 sites, 34 had something genuinely underneath them. A switch room. A pump house. A wartime fuel gallery with a bricked arch. A 900-cubic-metre covered reservoir with a hatch, a ladder and a padlock.
The median real volume was about 610 cubic metres. The median volume implied by the story as it was told to us was around 74,000.
The facility is usually real. It is usually one room. Somewhere between the room and the telling, the story has added eleven floors, and nobody involved was lying — a hatch in a hillside is a genuinely strange object, and a person who finds one is entitled to wonder.
We have four checks for the eleven floors. We have none at all for the room.
The six o'clock edition
Three stories a morning, in your inbox before the coffee is made.
Sources & Method
We wrote the method down first, in January, and only then went looking for sites, because the reverse order is how a screen becomes a story. We tested it on two places that publish their own numbers — a working potash mine and a metro ventilation scheme — before applying it to anything rumoured, and we have published the misses from both calibration runs at the same size as the hits.
Who we spoke to
- Fenella Astrup-Boakye, Chartered ventilation engineer; mine and tunnel airflow, 26 years. Commissioned to build the air and heat model and to try to break it; her report and her objection are published entire January–July 2026 Thinks we have oversold a screen as a proof. Her reply is at the foot of this page and it is the strongest thing in the piece.
- Bronwen Achterberg-Ruiz, Trade effluent officer, a water and sewerage undertaker. Interviewed by telephone three times, then for two hours at the works with her consent schedules on the table March and May 2026 Has never heard of any of the sites in this article and did not ask which ones they were.
- Distribution network operator connection registers, Quarterly published registers of accepted and energised connections, four licence areas. Downloaded from the operators' own planning pages; consolidated to 41 sites and republished by us as a spreadsheet 2019–2026 Free. Nobody has to ask for them and nobody has to answer.
- Environmental statement and spoil schedule, a metro deep-tube ventilation scheme, Design airflows, shaft dimensions and excavated volumes by disposal route. Public documents, read against the contractor's monthly haulage returns obtained under the environmental information regulations February 2026 Our second calibration site. Chosen because every number we wanted to guess was already printed.
- A working potash and polyhalite mine, North Yorkshire coast, Main fan duties, hoist schedule, licensed tip consent and half-hourly demand. Two site visits with the operator's permission; underground once, for four hours, with an escort April and June 2026 The only place in this story where the desk has been more than a metre below the surface.
Documents
- PX-1402 — Denver International Airport, Concourse D substructure — mechanical drawings, sheets 1–6 of 19 accepted
- PX-1912 — Four raw satellite radar passes over the Giza plateau, purchased by this newspaper accepted
What we could not confirm
- The 6 to 9 kW per occupant band. It is derived from four deep ventilated facilities whose operators published their loads, all of them designed to hold 26°C dry bulb. A facility content to run at 30°C wet bulb, with people who are used to it, could sit well under our band and we would call it a farm.
- Whether the connection registers are complete. A site generating behind its own meter, on diesel or on a private wire from a neighbouring works, appears nowhere in them. We asked three operators how often that happens and all three said they did not hold the information.
- The 41 sites themselves. They are not a random sample of rumoured facilities. They are the ones readers wrote to us about, in English, mostly in two countries, and a method calibrated on that sample tells you about that sample.
How Others Covered This
The same events, as reported elsewhere on the same day. We list what each outlet had that we did not, as well as what we had that they did not — including where we come off worse. Why we print this.
-
The Meridian TelegraphTHE GRILLE IN THE HILLSIDE: WHAT THEY ARE BREATHING DOWN THERE
Took the air arithmetic and ran it as a discovery, then applied it to a Gloucestershire escarpment without checking the connection register.
Had that we did not
The 0.47 figure, correctly cited to the occupational standard, and better explained than we explained it.
Left out
The four ways the method fails, which occupy a third of our piece and none of theirs.
-
Signal & LedgerA Four-Test Screen For Undeclared Subsurface Capacity
Turned the method into a due-diligence checklist for people buying industrial sites, which is a fair use of it and probably its most valuable one.
Had that we did not
The bulking factors and the connection-register route, laid out as a workflow.
Left out
That three of the four tests depend on the regulatory furniture of one country, and that the effluent test is worthless anywhere on septic.
-
They Buriedthis newspaperA Tunnel Has To Breathe. It Has To Put The Dirt Somewhere.
Published the desk's own working method in full, calibrated it against two sites that are not secret, and named where it breaks.
Had that we did not
The arithmetic, the two calibration runs with their misses, the four failure modes, and the engineer's objection to the whole exercise.
Left out
We have published a method that sends readers to ring water company clerks, and our engineer told us so before we ran it. We have printed her warning and taken none of her advice. — V. Ashcombe-Doyle, standards editor
Right of Reply
They Buried contacted Fenella Astrup-Boakye, chartered ventilation engineer on 3 July 2026 with the full draft, the model as we had used it, and no deadline. Replied 19 July 2026 and asked that it run entire. Printed unedited, including the parts about us.
You have paid me to build a screen and you are about to publish it as a test, and those are not the same object.
A screen tells you where to spend your next four days. A test tells you what is true. What I gave you sorts sites into probably-nothing and worth-looking-at, and it does that quite well, and I am pleased with it. It does not establish absence. Nothing I have ever built establishes absence, because absence is not a thing a fan curve knows about.
Take the air number, since you have put it in your headline. Nought point four seven is a regulatory minimum for a workplace, and it assumes the workplace is trying to comply. It assumes eight-hour shifts, ordinary work, a duty of care and an inspector who might turn up. Change any one of those and the number moves a long way. Submariners have lived for months at carbon dioxide concentrations that would close a factory. I do not say anybody is doing that. I say that your grille shrinks by two-thirds the moment somebody decides to be uncomfortable, and a two-thirds error in the one quantity you can actually stand in front of should make you quieter than you are being.
The cooling arithmetic is the part I would defend hardest and it is the part you have buried. People are 100-watt objects. Three hundred of them in a sealed volume is 30 kilowatts of meat before you switch anything on, and heat does not care about secrecy — it has to go somewhere, and there are only three somewheres: air, water, or rock. Rock saturates. I would rather you had led on that than on the sewage.
And now the thing I actually mind.
You are going to publish a method that ends at a named clerk in a billing office. I have read your paragraph about her and it is affectionate and it is going to make her life worse. Every one of your readers who runs this at home will run out of public documents at step three and will then telephone a water company and ask a person on eleven pounds an hour whether there is an undeclared discharge at a grid reference. She will not be able to say. She will not be allowed to say. And when she says she cannot say, some proportion of your readership will hear that as confirmation, because that is what a refusal sounds like when you have already decided.
You have built a machine for generating exactly the thing you say you are against, and you have built it well, which is worse.
I would still rather you published it than not. A bad screen published with its failure modes is better than a good screen kept in a drawer, and I have read your four failure modes and they are the honest ones — I would have added a fifth, which is a facility built before 1962, when none of this was registered anywhere. But I want my name on the objection as well as on the model, and I want it in the same typeface.
Published unedited under our right-of-reply guarantee.
How was this story?
We publish the result, whatever it is. Reader verdicts appear on the front page and in our newsroom metrics.
Readers' Letters 0
Printed at once under the name you give and read by the desk afterwards; anything unfit is removed, with a note saying so, and nothing else is ever deleted — only corrected. Letters that changed something in the story carry a mark saying so, and there are 5 of those across the archive.