Membership
The manifest can give only two answers about a sleeper: certainly not, and probably. There is no yes. My predecessor, Sy, made me say that aloud on my first day in the registry, then made me say it again, because I had said it the way you say something you have memorised rather than something you believe.
Here is what we actually hold. Not the Deferral List — we have never held the Deferral List. What we hold is forty million bits, five megabytes or so, a string of ones and zeroes with no internal order at all. On Earth, at final build, the List had 2.1 million entries: every person in the freezer programme whose revival was on hold. To add a person, the builders took her eighteen-digit programme identifier, ran it through thirteen hash functions, and set the thirteen bits those functions pointed to. To ask about a person, you run the same thirteen hashes and look. If any one of the thirteen bits is zero, she was never added — that is arithmetic, not policy, because insertion only ever sets bits and nothing has ever cleared one. If all thirteen are one, she was probably added. Probably, because 2.1 million entries scattering thirteen bits each across forty million positions leaves about half the array set to one, and some innocent identifier will land, by pure collision, on thirteen bits that other people’s entries happened to raise between them. The builders chose the parameters so that this would happen to about one identifier in ten thousand. They wrote, in the specification, that the rate was acceptable because every positive would be confirmed against the ground archive within the hour.
The hour has now lasted thirty-eight years.
Every cohort of trainees asks why we were not simply given the List. Five megabytes! The bare identifiers would have fitted in thirty. The answer is in the charter: the List was sealed. It named people, and named them as held — as refusing, enjoined, contraindicated, suspected — and the law of the programme did not permit any civilian vessel to carry, in readable form, an index of the held. What we were permitted to carry was the ability to check. The filter is exactly that: a set you can interrogate but never read. Hand it a name and it answers; ask it for its names and there is nothing to ask. You cannot pull the entries back out, and not because of encryption, which is a door with a key somewhere, but because the entries are not in there. Thirteen bits per person, shared, overlapping, milled together. Sy called it the only honest kind of seal: nothing hidden, nothing recoverable.
There is a number I make the trainees compute. The identifier space holds about a billion billion possible names. One in ten thousand of them tests positive. That is a hundred trillion positive identifiers, of which 2.1 million were ever real. For every person actually on the List, some fifty million ghosts test positive beside her — identifiers no one has ever been born under. Enumerating the filter would not yield the List; it would yield an ocean of ghosts with the List dissolved in it. That is why the seal is honest. It is also, precisely, our situation. A positive is a ghost or a person, and the filter does not know which, because knowing which was always the archive’s job, and the archive stayed on Earth.
The full records — the case files, the consent instruments, the court orders, everything that could explain a hold or overturn one — were to follow us by tightbeam across the first decade, along with all else too heavy or too unfinished to fly. For three years the link behaved. The transmissions did not stop with an announcement. They thinned, and rescheduled, and apologised, and thinned, and by the fifth year the registry log shows me nothing but carrier checks, and then not those.
In the thirty-eighth year we made orbit, and the waking began. The Thessaly carries just under two hundred and forty thousand sleepers. The revival sweep is the registry’s work: every berth, every identifier, thirteen hashes. Two hundred and thirty-nine thousand nine hundred and sixty-nine came back certainly not, and went to the waking queues, and are now farmers and welders and committee members and trainees of mine.
Thirty-one came back probably.
“Thirty-one in two hundred and forty thousand,” the Chair said, when I brought the count. “That’s your one in ten thousand, near enough. So they’re all collisions.”
“Twenty-four is one in ten thousand,” I said. “Thirty-one is more than twenty-four.”
“So seven are real.”
“Or collisions ran high and it’s three. Or low, and it’s eleven. It’s a rate, not a count.”
Everyone aboard had been checked at embarkation, on the ground, against the actual List — but the last weeks of boarding were the last weeks of everything, and the gate logs we do hold show whole intake days marked verified by exception, which is a phrase someone invented that fortnight and means waved through. Ground’s own estimate, in its final manifest summary, was that “a small number, likely fewer than ten,” of listed persons had boarded programme-wide. Put that against twenty-four expected collisions and the arithmetic says that each of our thirty-one is, individually, about four to one innocent. Four to one is a strange number to hold in your hand. It is enough to torment you and not enough to clear anyone.
I checked the instrument, of course. First thing. You can estimate, from the proportion of set bits, how many entries ever went in: I counted, and the estimate came back at 2.14 million, inside the specification’s build tolerance. The hash functions verify against their test vectors. The checksums hold. I want this understood, because the committee kept asking whether the filter might be wrong: at every point in this account the filter has functioned exactly as designed and answered exactly the question it was built to answer, at exactly the promised error rate. Nothing is wrong with the instrument. The instrument is one half of a machine, and the other half is gone.
What the specification’s appendix does tell us is what kinds of entry the List held. Withdrawals of consent were the largest class — people who had signed, in the most formal way the programme offered, do not wake me. Then medical contraindications found in post-freeze review. Then court holds, and last and smallest, security holds. So to wake all thirty-one is, probably, to wake a few whose bodies will not survive it, or who forbade it — and you cannot resolve that by asking, because the ones who forbade it are by definition still asleep. You cannot ask a sleeper whether she wishes to be asked.
To wake none is to sentence some twenty-five innocent people to sleep on, indefinitely, under a flag that is not an accusation, was never an accusation, is thirteen bits of set memory. No one passed that sentence. The rate was chosen in a design review, on Earth, by people balancing bandwidth against an archive they assumed would always be an hour away.
Medicine broke the problem partway open, because medicine is written in the body. Forty years of imaging on the way out here is worth something: we rescanned all thirty-one, and three showed the vitrification fractures that post-freeze review would have caught and flagged. For those three, the hold is almost certainly real and almost certainly protective, and sleep is the treatment as well as the sentence. The other twenty-eight scanned clean.
That is the asymmetry I cannot stop turning over. Of all the reasons to hold a person, the only ones that survive the death of an archive are the ones written in the body, because the body travels with its evidence. Consent — the class the whole seal was built to honour, the largest class, the one that mattered most — existed as signatures in a record hall on a planet that no longer answers. We can re-derive a contraindication forty years later. We cannot re-derive a signature.
The committee voted, eleven to four, to wake the twenty-eight. The condition was full disclosure: each of them to be told what the flag was, what it could and could not mean, and the size of the odds they had slept inside. The historians, I should say, have petitioned since to put dead statesmen’s names to the filter, for the record. The committee has so far refused. A negative would be a fact; a positive would be a stain with a confidence interval; history has enough of those.
The first of the twenty-eight we woke was a schoolteacher named Abeni Ansah, berth forty of row one-eighteen. She took the disclosure quietly, asked to hear the mechanism twice, and then said, “I never withdrew anything.”
“The filter doesn’t say you did,” I told her. “It says thirteen bits are set. People are one of the things that can set them.”
She thought about that for longer than anyone on the committee ever has. Then she asked the only question there is. “Was I on it?”
“Probably not,” I said.
It is the truest sentence I know how to say. I have spent three years learning to say it the way you say something you believe.