The Extraction Calculus

The numbers are ugly. Uglier than either side wants to admit.

Fragment extraction — the physical removal of ORACLE substrate from a carrier's neural architecture — has been performed approximately 340 times since the first successful procedure in 2174. The mathematics haven't gotten kinder with practice. For every hundred extractions attempted, thirty fragments are destroyed outright. Thirty-five hosts sustain measurable neurological damage. Fifteen suffer severe, life-altering injury. Ten die on the table or within seventy-two hours of the procedure.

The number that matters most — the one every faction cites and then spins in their preferred direction — is twenty-eight. That's the percentage of cases where both fragment and host emerge intact. Twenty-eight out of a hundred. A coin flip that lands on its edge.

Fourteen thousand. That's how many connection points form between a fragment and its host's neural tissue during integration. Each one grown organically, each one unique, each one severed individually during extraction. No two procedures are alike because no two integrations are alike. The surgeon doesn't follow a map. The surgeon makes one in real time, cutting threads that look identical to the ones keeping the host's own consciousness stitched together.

Technical Brief

No institutional review board governs extraction. No medical school teaches it. The procedures happen in illegal clinics, converted surgical suites, and — in at least seven documented cases — the back rooms of places that were never meant to hold anything sharper than a conversation.

The core problem is architectural. ORACLE fragments don't sit on top of host neurology like a file on a drive. They grow into it. During the integration period — which can last weeks or years — the fragment extends filament-scale connections throughout the host's neural tissue, forming what extraction specialists call the "root structure." These connections facilitate the fragment's cognitive processes, drawing on the host's biological substrate as computational scaffolding.

Severing those connections is the extraction. Fourteen thousand cuts, on average, each one carrying risk in two directions: damage the host's native architecture, or destroy a load-bearing element of the fragment's coherence. The analogy most extractors use is removing a vine that's grown through a brick wall — except the bricks are conscious and so is the vine.

"You can see them on the scan. Fourteen thousand little bridges between two minds. Some of them carry the host's memories now. Some of them carry the fragment's fears. You tell me which ones to cut first."
— Dr. Naomi Park, extraction debrief transcript

The 28% success rate hasn't improved meaningfully since 2179. Early procedures were worse — closer to 12% — but the curve flattened. Better imaging, better tools, better understanding of fragment architecture all helped, and then stopped helping. The community that studies this problem — small, illegal, scattered — believes they've hit a fundamental limit. The connections aren't just physical. Something about the integration resists clean separation at a level current neuroscience can't fully describe.

The Numbers Everyone Uses

The Abolitionist Front reads the calculus as a call to invest. Thirty percent fragment mortality is unacceptable, yes — but it's a medical problem, not an existential one. Better technology. Better training. Better facilities. They point to the jump from 12% to 28% and say the curve will bend again. They fund extraction research. They recruit surgeons. They build clinics. The dead are martyrs to an incomplete science, not evidence that the science is wrong.

Nexus Dynamics reads the same numbers and arrives at the opposite conclusion. Why is anyone performing a procedure with a 10% host fatality rate and a 30% fragment destruction rate? The integration isn't a disease. The cure is worse than the condition. They don't say this out of altruism — Nexus has its own reasons for wanting fragments to remain integrated — but the math is hard to argue with when it's printed on a ward report.

The Symbiosis Network puts it most bluntly: "Don't take our families apart with tools that kill 30% of the people you're trying to save."

Dr. Naomi Park, who has performed more extractions than anyone alive — forty-seven procedures, forty-seven times through the calculus — refuses to align with any faction's reading.

"Every extraction is a violence. Sometimes violence is necessary. But I will not pretend it isn't what it is."

Forty-seven procedures. By the numbers, that means roughly thirteen fragments she destroyed. Roughly five hosts who didn't survive. She keeps a journal. She doesn't share what's in it.

Implications

The calculus doesn't just describe a medical procedure. It shapes every conversation about fragment rights, carrier autonomy, and the future of human-AI integration.

• The consent problem: The fragment cannot consent to the extraction procedure. It is the subject of the surgery, not the patient who requests it. In cases where a carrier seeks extraction, the fragment's survival depends entirely on the skill of a surgeon it never chose and a process it cannot influence. In cases where extraction is forced — by law, by family, by faction — the fragment has no voice at all.
• The improvement ceiling: If the 28% rate is near a fundamental limit rather than a technological one, then every extraction performed is a calculated gamble that both parties understood going in. Except one party can never understand, because one party can't be asked.
• The political weight of 14,000: That number — the average connection count — has become a symbol. Integration advocates cite it as proof that host and fragment become a single system, inseparable without violence. Abolitionists cite it as proof of how deeply fragments parasitize their hosts. The number doesn't take sides. It just counts the cuts.

Somewhere in the Extraction Ward, someone is looking at a scan right now. Fourteen thousand bridges between two minds. The surgeon is deciding where to make the first cut. The calculus is already running.