DAY 685
The terminal's login prompt had a thirty-second timeout—her specification, implemented into the Syndicate's encrypted communications protocol three years ago. The server hashed each credential request against a rotating key that changed at 00:00 Tokyo Standard Time, a timing she had chosen because she had calculated the Beijing Collective's intrusion monitoring would refresh at 23:45 and leave a fifteen-minute window of reduced attention. She had never told anyone about the timing. It hadn't come up. She typed her credentials, the terminal accepted, and the development repository opened on the leftmost monitor.
The commits were organized by date, not author, because the repository had been structured that way from the start. Author attribution existed in the metadata—she expanded the commit log now—and there it was, every structural commit from the algorithm's first iteration through its final stable version tagged as v3.2: Author field: Takahashi, K. The K stood for Kimiko. Her mother's name. A small joke that had amused her when she created the account and had not seemed worth retiring since.
The algorithm's development history ran from twenty-six months before the race's start date to six weeks before the public launch of the HashNet blockchain. Forty-one commits. The first nineteen addressed the proof-of-work structure—standard cryptographic construction, nothing to distinguish it from the fifty other novel consensus mechanisms published in that decade. Commit twenty introduced the dual-function embedding: a modification to the internal energy landscape of the hash function that mapped computational traversal paths to protein conformational space. The change was seventeen lines of code. She had written it in a single afternoon, in this room, at this desk, while her husband's photograph watched from the shelf where she had placed it six weeks after his funeral. She did not look at the photograph now. She scrolled through the commit log.
By commit thirty-one, the protein target integration was complete. The algorithm would solve, simultaneously, every valid proof-of-work nonce and every low-energy protein configuration along the embedded folding landscape. The mining network would optimize for hash rate and, without knowing it, optimize for molecular structure. The two problems were, at the relevant mathematical level, the same problem.
The 847-day figure appeared in commit thirty-six, in a comment she had encrypted within the commit message: Timeline to completion of primary target set. Assumes sustained network growth to 25 EH/s combined. Conservative estimate. The network had reached 28 EH/s—three months ahead of her conservative estimate.
On the rightmost monitor, the real-time operations dashboard showed the Krafla facility's current hash rate: 2.4 EH/s, stable, green. Every mining unit running. Outside in Iceland's volcanic northeast, the Icelandic night would be absolute, the facility lit from within like a lamp in a field of lava. She knew this from the one time she had visited, eleven months into the race, when she had needed to verify that GRIND-7's optimization approach was producing the throughput she required. She had not told him she was there for the throughput.
She closed the commit log. The map of global mining operations occupied the wall to her left—six facilities, four continents, the lines connecting them drawn in the thick strokes of a battle map, which, technically, it was. She had designed the network's geographic distribution herself. Singapore for regulatory flexibility. Iceland for power cost and cooling. Kazakhstan for land. Ontario for proximity to financial infrastructure. Each facility placed where it would produce the maximum computational output for the required duration, 847 days; this was Day 685. She went out to the garden.
The garden was her husband's more than hers. He had planted the maple during the second year of their marriage, when the tree was young enough to carry from the nursery in the back of a car and old enough that it had already developed the particular shape of a tree that understood it would outlive the hands setting it in the ground. He had measured the planting hole with a tool she had never learned the name of, had tamped the soil himself around the base with both hands, and she had watched him do it from the engawa that year, before she had known to mark the moment. Her father had done the same with the okra in her childhood garden. She had never grown okra since.
The maple was now nineteen years old. This morning it held the last of the winter light, which came sideways at this hour, the angle too low to warm anything but catching in the branches where the leaves had not yet unfurled for spring—papery, furled, the color of something about to happen.
Kenji's diagnosis had been rendered on a Tuesday in October, three years before she launched the race. Stage IV pancreatic adenocarcinoma. The oncologist at Kyoto University Hospital had used that term with the precision of a technician delivering a measurement, which it was, which she had appreciated. The five-year survival rate: four percent. Median survival from diagnosis to death without intervention: three to six months with chemotherapy. Kenji had refused chemotherapy, had made that decision on the drive home from the hospital, sitting in the passenger seat looking at the river, and had explained his reasoning with the same precision the oncologist had used: he wanted the months he had to contain himself, not a treatment's side effects. He had lived for eight months, and she had not tried to change his mind.
Pancreatic adenocarcinoma was now cured. The algorithmic output for the pancreatic cancer protein targets had published on Day 658, twenty-seven days ago. The synthesis pathway was in public repositories. Independent labs were confirming the mechanism. A child in Boston was apparently asking to receive it. The maple moved, slightly, in a wind she couldn't feel from the engawa.
She had not designed the algorithm because of Kenji. She had designed it because of what it was possible to compute if you could marshal the distributed resources of a global mining competition, and because she had read the protein folding literature in the year after his death and understood that the mathematical structure of the problem was equivalent to a class of optimization problems that mining networks were built to solve. The biology had not prompted the computation. The computation had presented itself as a solution to the biology. These were different things, and the distinction mattered to her, and she did not expect anyone else to find it significant.
She stood in the garden for another four minutes. The light did not change. She turned and went inside.
The call connected at 9:47 and ran for eleven minutes. She had known Wei Longwei's voice for six years—they had met three times, in Singapore, in Zurich, and once at a computational infrastructure conference in Seoul where they had not acknowledged knowing each other. He spoke Mandarin at a register that suggested he expected the call to be recorded, which it was—Syndicate protocol—and which mattered less than one might expect, because the content of the conversation was not, at any level, legally actionable.
"The Parkinson's targets are at eighty percent," he said. The formulation was not a question.
"78.3 as of last night," she said.
"Our MS cluster is at sixty-five."
"65.1."
A pause in which they both noted that they were reading from the same dashboard. She had given him access to the same shared metrics feed she had given GRIND-7 and Tanaka—eight months ago, without mentioning it to anyone at the Syndicate.
"The verification program at WHO," Wei said.
"Twelve months to clinical data on all five compounds. Tanaka presented the parallel track recommendation yesterday."
"And the Parkinson's data will publish in—" He did not finish the sentence. He did not need to.
She looked at the mining operations map on the wall. The Beijing Collective's facilities in Inner Mongolia were marked in blue; her own in red. A carryover from the race's initial framing as a competitive enterprise. She had never changed the colors. She had kept the war-room aesthetic because it was useful for the people who needed to believe they were in a war.
"The Collective's combined throughput?" she said.
"Fourteen point nine. Down from peak by about three percent. Hardware fatigue."
"We're stable at thirteen. The Iceland facility is running at full capacity."
Another pause. When Wei spoke again, he was slightly more precise than before. He was approaching the thing he had called to say. "The Beijing Consortium's board is asking questions about the timeline. They understand the economic disruption. They're less certain about—" He chose his words. "—the sequencing."
"The sequencing was always going to be difficult," she said.
"Yes." A pause. "They're asking if there's a mechanism to slow the publication cycle."
She looked at the map. The red and blue points. The lines that had looked, to anyone who had seen this room without knowing the architecture, like fronts in a conflict.
"The publication cycle runs at the rate of the computation," she said. "There's no mechanism to slow. The data resolves, and when it resolves, it's in the open-source commits, because that's how the blockchain records valid blocks. The Consortium can reduce hash rate if they want to reduce throughput." A pause. "That would also cede the race, formally, to the Syndicate."
Wei made a sound that was not quite a laugh. "That's what I told them."
"Then we understand each other," she said.
The call ended at 9:58. She stood at her desk for a moment, then sat and opened the development log again—not to read it, but to check the timestamp on the first commit. The commit that had created the Takahashi account was timestamped 2:14 in the morning, a Wednesday in November, three months after Kenji's funeral. She had not planned to work that night. She had been at the terminal by two in the morning. She closed the log and caught the morning Nozomi to Tokyo.
The Syndicate's conference room in Marunouchi faced the Imperial Palace grounds, though from this floor the angle was wrong—what was visible was a section of the moat and, beyond it, the low profile of the palace complex, which gave the impression of a city that had chosen to keep something old at its center. She opened the four video feeds on the display wall and Takeshi appeared first, Singapore office, early afternoon there, his jacket on the back of his chair in a way that meant he had come from or was going to a meeting. He summarized the financial restructuring progress in three minutes: the generic cure manufacturing bonds were attracting sovereign wealth capital ahead of projection, the healthcare worker retraining instruments were oversubscribed, Volkov's modeling was producing term sheets faster than the legal team could process them. "He doesn't know he's executing your contingency plan," Takeshi said.
"He's executing an optimal response to the situation," she said. "That's sufficient." Takeshi wore the expression he used when he thought she was sidestepping something. She did not clarify.
GRIND-7's feed showed the Krafla operations center: the hash rate monitor at the top of the display, green, the same green she had checked on every dashboard for 685 days. He delivered what she needed in forty seconds. The computation was running. The target queue was progressing. The mining team was stable.
Tanaka's feed showed a Geneva hotel room. She had seen the parallel track recommendation in the WHO Secretariat's advisory session notes—Tanaka had flagged her access three weeks ago. The recommendation was correct. It was also the only outcome that had been possible once the cures crossed the propagation threshold, which she had known when she designed the algorithm's publication mechanism. She closed three of the four feeds and left the hash rate monitor running.
The train south returned her to Kyoto by mid-afternoon, and she called GRIND-7 at 14:30. The call connected in twelve seconds. His voice was what it always was: minimal, present, a voice that gave you the exact information and nothing additional.
"The computation," she said. "What is it solving now?" There was a pause—three seconds, the pause of someone deciding whether the question was what it appeared to be.
"Current priority targets," he said. "Alpha-synuclein pathway, Parkinson's disease. Myelin sheath repair pathway, multiple sclerosis. Lupus nephritis pathway, systemic lupus erythematosus. Two additional targets in the autoimmune cluster, computation at forty and thirty-two percent." She said nothing, and he continued: "Parkinson's will resolve in approximately eighteen days at current throughput. MS in thirty-one. The autoimmune cluster is harder to project—the targets have cross-pathway interactions we're still mapping. Probably forty to sixty days for the cluster."
Ten million people had Parkinson's disease. Three million had multiple sclerosis. Systemic lupus affected five million, predominantly women, predominantly in their working years. She held these figures where she always held them—not as an abstraction, but as a constraint on the space of acceptable decisions.
"Then we don't stop," she said.
He didn't respond immediately. She understood why. It was not a command; it was a confirmation—the confirmation that the decision he had already made was also the decision she was making, which distributed the weight of it differently. When he spoke, he said: "No. We don't." She ended the call and went out to the garden.
The garden in the evening held a different quality of light than in the morning. Morning light came at an angle and was thin; evening light was lower and more diffuse, and it seemed to come from everywhere at once, from the moss and the stone as much as from the sky. The maple caught it in the branches rather than the leaves—the leaves still furling, a few more warm days from opening—and stood the way it had stood through every year since Kenji had tamped the soil around its base with both hands.
The computation was running in six facilities across four continents—28 EH/s of combined throughput, the Syndicate and the Collective together, as it had always been together—solving the protein targets in the order she had sequenced them twenty-six months before the race began. Mortality impact first, then disease burden, then research state: the diseases for which the folding problem was closest to tractable, and which would therefore yield fastest to the algorithm's method. One hundred sixty-two days remained. Pancreatic cancer had been sixth in the mortality ranking and first in the sequencing queue.
Through the study window, visible from where she stood, her husband's photograph was on the shelf. The evening light was such that she could not see his face from this angle, only the frame and the general impression of a photograph. This was fine. She had looked at it from close enough, often enough.
Day 847 was the completion date for the primary target set. After Day 847, the algorithm would continue running—she had designed no termination condition, because there was no terminal state in which all solvable proteins had been solved. There were 20,000 protein-coding genes in the human genome. The algorithm had begun with the forty-three most consequential disease targets. After Day 847, there would be 43 fewer. There would still be 19,957 remaining. She stayed in the garden until the light failed completely, until the maple was only a shape against the darker shape of the sky, the moss at her feet invisible, the stone path known by memory rather than sight. Then she went inside.