True Copy

The plan came up on the hearing-room screen at four hundred per cent, and it was beautiful. That was the first thing I put in my report, though my instructing solicitor struck it out. Drawn in ink on linen in 1987 by a surveyor who died before the century turned, scanned once in 2031, and there was no grain to it, no fog, none of the softness you learn to expect from old paper pushed through old glass. Every figure stood off the screen as though it had been typeset that morning.

“Exhibit R-4,” said counsel for the acquiring party. “The conveyance plan. I’d invite the tribunal to note the dimension on the western boundary. Sixteen point eight metres.”

Crisp as scripture. And in the bottom corner, the stamp: CERTIFIED A TRUE COPY OF THE ORIGINAL, WHICH HAS BEEN DESTROYED IN ACCORDANCE WITH SCHEDULE 4.

I was twenty-four when the Consolidation started and twenty-nine when it finished, and for three of those years I sat on the verification line at the Registry’s site outside Swindon. Split screen: original on the left, scan on the right. You checked the page count, you checked nothing was folded or skewed or cropped, you checked legibility, you initialled, and the original went into a grey crate, and the crates went to the pulper on Thursdays. Eleven seconds a page, on a good afternoon. We were told, and it was true, that every scan was hashed at ingest, and that the hashes were anchored somewhere incorruptible, and that this made the archive safer than paper ever was. Paper burns. Paper foxes and silverfishes. The archive would be bit-perfect for a thousand years.

Nobody lied to us. Every word of that is still true. The hash of Exhibit R-4 has verified, flawlessly, every night for thirty years. We hashed it after the scanner had finished with it, that’s all. You can keep a thing perfectly and still not have kept the thing.

The scanners were Calder Meridians, and on the archive setting they did something that was in the manual, if you read the manual. A page of text is mostly the same few dozen shapes repeated, so the Meridian built a dictionary as it went: the first time it met a mark, it stored the mark; every time after, it measured each new mark against the dictionary, and if the difference fell under threshold it stored no image at all — just a note saying another one of those, here. That is why the archive fitted in the space it fitted in. And it is why every letter in it is so clean. You are never looking at the ink. You are looking at the dictionary’s specimen, stamped wherever the machine believed it saw a match.

Believed. Under threshold is not the same as identical. When the machine got a match wrong — a worn six against a stencilled eight, at the sizes surveyors used for dimensions — it did not blur. Blur I could work with; blur is a machine being honest about what it doesn’t know. The Meridian substituted. It printed the wrong digit with total confidence, in the house style of the page, and moved on.

Mrs Ansell is eighty-one. Her husband put up the fence along the western boundary himself, in the spring of 1988, and set his gatepost two big strides and a bit in from the corner, because the plan he’d signed the year before gave him eighteen point eight metres and he wanted the van to pass. The acquiring party’s scheme needs the strip his van used to stand on, and their case is short: the register is conclusive, the register incorporates the plan, the plan says sixteen point eight. The fence, on this telling, has been two metres into someone else’s land for fifty-two years, and a dead man’s stride is not evidence against a national archive.

I asked for the raw file, not the viewer’s rendering, and I did what I always do first. I cut out every six on the page and overlaid them. Exclusive-or: zero. Pixel-identical, all four of them, including the one in sixteen point eight. Then the eights: identical, all six. In a true scan of a typed or stencilled page no two characters are alike — ink starves, fibres catch, the hand moves. Identical glyphs don’t prove the figure is wrong. What they prove is worse in a way: the page has stopped being evidence about the ink, and become evidence about the dictionary.

“You’re saying the scan is a forgery,” the chairman said.

“No, sir. A forger knows what he wants you to read. The machine had no intentions at all. It was built to say another one of those, and it was extremely good at it, and the only failure it had in it was to be wrong about which those. I’m saying the scan is not a photograph. It’s a transcription, done by a clerk who never admitted to doubt.”

The Science Museum’s store at Wroughton still has two Meridians, and one of them runs. Under supervision I fed it test sheets — stencil numerals, the sizes off the plan, decent toner, nothing unfair — on the same archive setting recorded in R-4’s metadata. Eleven pages. It swapped a six for an eight three times, an eight for a six once, and a three for an eight once, and every substitution was perfectly formed. The video went in as Exhibit C-9. The chairman watched it twice and did not say anything for a while.

Counsel for the acquiring party did the competent thing and turned my evidence around. “On Dr Hallam’s account, any digit in this archive might be an artefact. The tribunal cannot prefer Mrs Ansell’s eighteen to the register’s sixteen; her figure could as easily be the corrupted one. Her expert has not impeached one number. He has impeached all of them, including his own.”

Which is right. That page, on its own, can no longer tell you anything. If it said eighteen point eight I could not have sworn to that either. So I stopped asking the digits and asked the drawing.

A parcel plan is a closed traverse. Leave the corner post, walk each boundary in turn — this bearing, this length, the figures written along each side — and you must arrive back at the corner post. The errors of a competent 1987 survey close to centimetres. It is, although nobody in 1987 would have used the word, a checksum: the polygon carries enough of itself to check the rest.

I walked R-4’s boundaries as written. With sixteen point eight on the west, the traverse misses its start by one point nine seven metres — a gross error, the kind that gets a surveyor’s letters taken away, sitting in a drawing that is otherwise fastidious. Then I tried every single-digit correction in the pairs the Meridian is known to confuse, every figure on the sheet in turn. One and only one closes the polygon: eighteen point eight on the western boundary. Misclosure, thirty-one millimetres — at plan scale, comfortably inside the width of the ink line the surveyor drew.

“The drawing vouches for itself,” the chairman said. “That’s the submission.”

“It’s the only witness left, sir. We pulped the rest of them on Thursdays.”

They found for Mrs Ansell. The determination runs to sixty paragraphs and the word the tribunal chose was not forgery or error but substitution, which I thought was exact. The gatepost stays where Ted Ansell’s stride put it, and the register now carries eighteen point eight with an annotation, the first of its kind, recording that the figure was recovered from the geometry of the instrument rather than the image of it.

It will not be the last of its kind. The Registry has commissioned the audit I spent a career saying quietly ought to happen, and the early numbers are about what I expected: of the sampled pages that contain their own arithmetic — accounts that must sum, traverses that must close, dates that must land on their stated weekday — roughly one in two thousand disagrees with itself. Most heal the way R-4 healed. A marriage register gave us a Thursday the eighteenth of June, 1953, above the printed word Tuesday; the sixteenth was the Tuesday; the couple got their anniversary back after a hundred years, not that they were using it.

And then there is the rest of the archive: the pages that carry no redundancy at all. A name. A single sum, with nothing it must add up to. A dose on a hospital sheet. Those digits are almost all correct — the substitution rate is genuinely tiny — and each one of them is now a specimen from a dictionary, immaculate, unvouched, and there is no polygon to walk. We can say of the archive as a whole that it lies about one page in two thousand. We cannot say which pages, and we never will, and the pages will go on looking the way they look, which is perfect.

The certification stamp, I should say, comes through the Meridian flawlessly on every one of eleven million pages, bit-identical, and no harm done — a rubber stamp really is the same mark every time. That was the machine’s entire theory of writing: that every mark on paper is a stamp, an instance of a type, another one of those. It was only ever wrong about people.

CERTIFIED A TRUE COPY OF THE ORIGINAL, WHICH HAS BEEN DESTROYED IN ACCORDANCE WITH SCHEDULE 4.