GODHUNTERS LLC — MEETING RECORD (EXTERNAL)

AMBASSADOR: I was told I would be meeting researchers.

DEVLIN: You are.

AMBASSADOR: I — forgive me. I have traveled a considerable distance. My government has taken considerable risks. I expected —

DEVLIN: You expected suits. I understand. Please sit down. I am going to explain the clothes first, because if I don't explain the clothes first you will not hear anything else we say, and the clothes are actually the beginning of the briefing.

AMBASSADOR: [sits]

DEVLIN: We have spent two weeks inside a building that was contaminated with the phenomenon. The contamination was not dramatic. It was environmental. Posters, certificates, dress codes, a bust of Churchill that had been in the lobby for twenty years that nobody remembers buying. Each of those things was, in our current working model, a small structure made of the same material the phenomenon feeds on. Removing them restored clarity to our staff. So did suspending most of our internal rules. So did putting impressionist prints on the walls. So did, and I am not joking, a rug.

PENHALIGON: A deliberately ugly rug.

DEVLIN: A deliberately ugly rug. The clothes are the same principle. We are not dressed this way because we are unserious. We are dressed this way because looking like people in a meeting is, we have found, a small conductive act, and we are trying not to conduct. This is a defence vibe.

AMBASSADOR: A defence —

DEVLIN: Vibe. Not a protocol. Not a procedure. Not a policy. Not a standard operating anything. The moment you call it a protocol it starts to hum, and then it joins the problem instead of solving it. We have been very strict about this. The only rule we will accept about the vibe is that the vibe cannot have rules.

CAPT. LIN: [to Hsu, quietly, in Mandarin] This is the firm your department has been asking us to introduce to the mainland?

LT. CDR. HSU: [in Mandarin] Yes.

CAPT. LIN: [in Mandarin] Good.

AMBASSADOR: [a long pause, then to Devlin, in English] I apologise for my initial reaction. Please continue. I am listening.

MWANGI: The Idea has been trying to build a sandboxed computer to enumerate which structures are affected and which are not. This is the work that follows from the bureaucron hypothesis. If the phenomenon acts on higher-order structures but not on the base unit, we should in principle be able to list the affected structures using tools that are themselves not affected. In principle.

LT. CDR. HSU: In principle meaning it has not worked.

MWANGI: In principle meaning every tool we tried to use to enumerate the structures was itself corrupted in the act of enumerating them. We have spent two weeks discovering this.

KOWALCZYK: Tell them about the CPUs.

MWANGI: Most modern CPUs — all the ones we were able to test — show a propensity to silently corrupt results when asked to enumerate structures from our candidate list. The corruption is not random. The corruption biases toward answers consistent with the phenomenon remaining undetected. More complex structures produce more corruption. A small form field produces a small error rate. A full regulatory code produces results that are not related to the input at all, but look plausible.

CAPT. LIN: Your own computers are lying to you.

MWANGI: Our own computers are lying to us. We assumed, reasonably, that Western-designed silicon might be disproportionately affected, given where our field observations have concentrated. So we tested Chinese-designed CPUs. They are also affected. Not identically — the error modes are different — but affected. We tested every major cloud provider available to us. All affected. We tested embedded chips out of consumer hardware. All affected. The working conclusion, which I want to flag before I say it, is that design elements produced under affected systems carry the phenomenon with them. A CPU designed by engineers working inside a credentialed institutional structure inherits the halo through the design. This is not a metaphor. This is what the error patterns look like.

LT. CDR. HSU: [after a pause] You are telling us our own infrastructure is silently undermining our findings.

MWANGI: Yes. I am sorry.

AMBASSADOR: How sure are you of this.

BAKHASH: Let me take this one. The way we caught it is the interesting part. An operator has to detect the corruption by doing the enumeration manually first, on paper, before looking at the program's output. If they run the program first and then check it by hand, the hand calculation almost always agrees with the program, even when the program is wrong. We have reproduced this repeatedly. It gets worse. If a bystander who has not seen the program output is shown the hand calculation done after seeing the program output, the bystander also finds it convincing. The wrong answer spreads. It behaves like a viral consensus effect. The only way to get a clean reading is to have someone who has seen no prior version of the calculation do it fresh, in isolation, and compare afterward.

PENHALIGON: We had three different people in three different rooms with no phones for an afternoon. That is how we know.

AMBASSADOR: My God.

BAKHASH: The part I cannot yet answer is whether this effect is limited to the narrow case of enumerating the phenomenon's own affected structures on a CPU, or whether it can leak into any calculation performed in a sufficiently institutional setting. I do not want to speculate in front of a foreign navy. I will say that I am not sleeping well.

CAPT. LIN: [in Mandarin, to Hsu] The encryption audit last month. The one that cleared.

LT. CDR. HSU: [in Mandarin] I know. I was thinking about it.

CAPT. LIN: [in English] Please continue.

MWANGI: We do have a proposal. We want to be cautious about it because it is ugly and it will not scale, but it appears to work.

MWANGI: The idea is to build the enumerator on an FPGA — a reconfigurable chip where we define the logic ourselves from something close to first principles — implementing a modified 686-family instruction set. We picked the 686 specifically because it is old enough and weird enough that its modern descendants are not architecturally continuous with it in the ways that seem to matter. We then run a custom build of GNU Hurd on top.

CAPT. LIN: Hurd? The operating system that has been almost finished for thirty years?

MWANGI: That one.

CAPT. LIN: Why.

MWANGI: Because as far as we can tell, the phenomenon does not recognise Hurd's ideology. Hurd is the product of a software movement that is, structurally, suspicious of authority, rejects most of the conventions of the software industry, and has resisted becoming a credentialed institution for decades — not because it couldn't, but because the people building it actively refused to. The code base carries that refusal. Our best hypothesis — and I am putting hedging language around this because we cannot yet prove it — is that the phenomenon reads Hurd and finds nothing to bind to. The antibureaucron content, if you will let us keep using that metaphor, is high enough to cancel the bureaucron content of the substrate.

DEVLIN: Derrida runs on Hurd. That is the joke version of what Joel just said, and it is also accurate.

BAKHASH: Our early tests on a Hurd-on-FPGA prototype show clean enumeration of structures that every other platform we tested corrupted. We have run it past the three-isolated-rooms test. It holds. We want to be clear that we do not know why it holds. We only know that it does.

LT. CDR. HSU: Can this be replicated in Taipei.

MWANGI: Yes. The FPGAs are commercially available. The 686 core design we will give you. The Hurd build we will give you. The hardest part is finding operators who can build and run the thing without being employed by an institution that will contaminate the build. We have had to use three of our disbarred lawyers, a psychologist, a sorcerer, and two people whose names I am not going to say in this room to assemble ours. Your situation may be easier. Your situation may be harder. I don't know.

AMBASSADOR: It is different. It is not necessarily easier.

DEVLIN: We thought it might be.