Engineered

The anomaly was three standard deviations from the mean.

Daniel Reyes stared at the scatter plot on screen four, his coffee going cold on the desk beside him. The surveillance centre occupied the entire seventh floor of the International Biosecurity Authority's Geneva headquarters, a cathedral of data rendered in brushed steel and low blue light, forty-two monitoring stations arrayed in concentric arcs around a central holographic display that no one ever turned on because the flat screens were faster. At 6:47 a.m., he had the floor nearly to himself. Two overnight analysts sat three rows back, murmuring about a false-positive flag out of the Jakarta node. The ventilation system hummed at a frequency just below perception, the kind of engineered silence designed to keep you focused until your eyes forgot how to blink.

Daniel's eyes hadn't blinked in a while.

He pulled the scatter plot wider, stretching the axis to capture the full eighteen-month longitudinal window. The dataset was routine, quarterly immune marker panels drawn from the Registry, covering 2.3 million CRISPR-X recipients across forty-one countries. His job, three days a week, was to run variance analysis on the incoming data, flag anything outside expected parameters, and file his findings with the Therapeutic Monitoring Division. Most weeks, the data was clean. CRISPR-X edits held to their predicted expression curves with a consistency that made the work feel almost ceremonial. The system was, by every published metric, the most precise genetic intervention in human history.

The scatter plot disagreed.

He isolated the variable, enhanced neutrophil signalling, a standard immune optimisation target in the CRISPR-X therapeutic suite. In a healthy enhanced population, post-edit neutrophil markers should cluster tightly around the corrected baseline, a neat cloud of data points hovering within one standard deviation of the programmed target. What Daniel was looking at wasn't a cloud. It was a smear. A subset of recipients, 1,614 individuals across nine countries, were expressing neutrophil activity well beyond the corrected range. Not suppressed, not degraded. Elevated. Coordinated. And trending upward over the last two quarters in a curve that had no business existing in a population receiving identical edits.

He tagged the subset and ran a secondary filter for demographic noise. Age distribution: normal. Sex ratio: normal. Ethnic and geographic spread: proportional to the broader cohort. No confounding variable jumped out. No obvious environmental correlator. The elevation was uniform across the subset, as if 1,614 unrelated people in nine different countries had independently decided to express the same immune trait at the same accelerated rate.

That didn't happen. Not without a cause.

Daniel set his coffee aside and pulled up the raw sequencing files for the first fifty cases. The data loaded in vertical ribbons of colour-coded nucleotide chains, each one representing a single recipient's edited genome mapped against the CRISPR-X reference template. He scrolled through them slowly, comparing edit sites. The primary corrections were intact, the targeted disease-associated variants had been replaced exactly as designed. Clean cuts, clean inserts, no off-target scarring. Textbook.

But downstream of the edit sites, in regions the algorithm shouldn't have touched, he found them. Small nucleotide variations. Consistent across all fifty files. Not identical, each one carried a slightly different arrangement, but patterned. As if something had nudged the surrounding genomic landscape into a new configuration, gently, precisely, and in the same direction.

He sat back.

In fourteen years of biosecurity work, eight of them in epidemiological crisis investigation before the IBA recruited him, Daniel knew to distrust the word "glitch." Glitches were random. They scattered. They didn't organise themselves across nine countries and express as a coherent phenotypic shift. What he was looking at had structure. Faint, buried under layers of otherwise perfect data, but real.

He opened a new analysis window and began building a comparison matrix. If the downstream variations were genuine, if they weren't artefacts of sequencing noise or batch contamination, they would show up in the protein expression data too. He queried the registry's proteomic layer, pulling translated protein profiles for the flagged subset against matched controls.

The results came back in eleven seconds.

Enhanced neutrophil signalling. Elevated interferon-gamma precursors. And, this was the one that made him lean forward, a novel chemokine receptor variant that didn't appear in any CRISPR-X design specification he had access to.

A receptor variant. Something the cells were building that nobody had told them to build.

Daniel checked the clock. 7:12 a.m. The morning shift would start arriving in twenty minutes. He saved the analysis to his secure partition, locked the screen, and walked to the window.

Geneva spread out below in grey pre-dawn light, the lake flat and colourless under low cloud. Across the water, the lights of the WHO campus glowed like a second city. Somewhere down there, in a dozen policy offices and advisory boards, people were preparing for another day of managing CRISPR-X's global rollout, the logistics of subsidised access, the diplomacy of licensing agreements, careful public messaging that kept enhancement adoption rates climbing quarter over quarter. It was the most successful therapeutic intervention in human history. Eighteen months since full-scale deployment. 2.3 million recipients. Hereditary disease correction rates above 99.7 per cent. The modelling teams had projected five million recipients within two years, ten million within four. The World Health Organisation had called it the single greatest advance in preventive medicine since vaccination.

And somewhere inside the data, something was moving that shouldn't be.

❦

By 9 a.m. the floor was full, and Daniel had buried his unease beneath two hours of procedural routine. He attended the daily briefing, status updates from regional nodes, a flag review from the overnight team, a short presentation on CRISPR-X adoption metrics in Southeast Asia that drew polite nods from the room. He sat in his usual seat, third row, left side, and said nothing.

He'd learned that raising a flag before you understood what it meant was worse than not raising it at all. At the Nigerian Centre for Disease Control, where he'd spent three years tracking Lassa fever mutation patterns, his supervisor had drilled a principle into him that he still carried like scar tissue: Don't shout fire until you can point at the flame. He'd seen colleagues burn credibility on premature alerts, anomalies that turned out to be batch errors, contamination artefacts, data pipeline glitches. Each false alarm eroded institutional trust by a measurable increment, and trust, once lost in a regulatory body, was almost impossible to rebuild.

So he waited. Watched the briefing. Logged his overnight report with the standard "no critical flags" summary. And went back to his station to dig deeper.

The comparison matrix was still running when he sat down. He'd set it to cross-reference the flagged subset against the full CRISPR-X edit history, every version of the therapeutic suite deployed since initial rollout, mapped against recipient outcomes. If the downstream variations were linked to a specific update cycle, that would narrow the cause. A bad batch, a corrupted deployment package, a regional server error. Something containable.

The matrix finished at 9:41 a.m. Daniel read the results twice.

The variations weren't linked to a specific update. They were present across all deployment versions. Every batch. Every regional server. Every update cycle since month four of the rollout. And they were cumulative, each successive quarter showed a slight amplification of the downstream shifts, as if the effect were building on itself.

That ruled out contamination. It ruled out a bad batch. It ruled out anything accidental.

He pulled up the CRISPR-X system architecture documentation, the technical specifications that the IBA had negotiated access to as a condition of global deployment approval. The document was extensive, 4,200 pages of algorithmic logic, edit-site targeting protocols, quality assurance checkpoints, and safety interlock descriptions. He'd read it twice during the initial approval process, once during his first quarterly review. He opened the section on post-edit genomic stability monitoring and began reading again, slower this time, looking for anything that might explain how downstream nucleotide variations could emerge consistently across a global population without being flagged by the system's own quality controls.

The architecture was clean. Every edit was verified against the reference template before cellular integration. Off-target scanning covered a 500-kilobase window around each edit site. Anything outside expected parameters was supposed to trigger an automatic alert to the deploying clinic and the central monitoring server simultaneously.

No alerts had been triggered. Not one, in eighteen months.

Which meant either the variations were too subtle for the scanning window to catch, possible, if they fell outside the 500-kilobase radius, or the system had seen them and classified them as acceptable.

Neither explanation was comforting.

❦

He found Lena Olsen in the canteen at noon, eating a bowl of something green and reading a regulatory impact assessment on her tablet. Lena ran the IBA's Computational Genomics Unit, a small team responsible for independent verification of CRISPR-X's algorithmic outputs. She was Danish, precise, and had the rare quality of being both deeply technical and genuinely uninterested in office politics. Daniel trusted her judgment more than anyone else in the building.

He sat down across from her without asking.

"I need you to look at something."

Lena didn't look up from her tablet. "Good morning to you too."

"Enhanced neutrophil signalling in the Q3 immune panels. I've got a subset of recipients expressing well above the corrected baseline. Sixteen hundred individuals, nine countries, no demographic correlator."

Now she looked up. "How far above?"

"Three sigma. And climbing."

"Climbing how? Linear?"

"Curvilinear. Accelerating slightly each quarter since month four."

Lena set her fork down. "That's not off-target scatter."

"No."

"Have you checked the proteomic layer?"

"Novel chemokine receptor variant. Not in any design spec I can access."

She was quiet for a moment. The canteen noise filled the gap, trays, conversation, the clatter of the coffee machine cycling through its motions. Daniel waited. He knew what she was doing. She was running the same mental triage he'd run six hours ago, sorting through possible explanations and discarding them one by one.

"Sequencing artefact?" she said.

"Consistent across fifty raw files. I'll check more, but the pattern holds."

"Batch contamination?"

"Present across all deployment versions. Every server, every region."

"Environmental cofactor?"

"Nine countries. No shared environmental variable I can find."

Lena picked up her fork, then put it down again. "Have you flagged it?"

"Not yet."

"Why not?"

"Because I don't know what it is. And if I flag it without an explanation, it goes to the review board, and they'll either dismiss it as noise or escalate it to the policy directorate, who will ask Helix Genomics for a technical response, and Helix will spend three weeks producing a report that explains why it's nothing. And by the time we get that report back, the Q4 data will be in and I'll have lost analytical window."

Lena studied him. "You want to run it through my unit."

"I want independent verification before I put it in the system. If it's real, I want to know what's causing it. If it's nothing, I want to be sure."

"It's not nothing, Daniel. You know that."

He did know that. He'd known it since 6:47 that morning, staring at a scatter plot that smeared where it should have clustered. But knowing and proving were different countries, and the border between them was where careers went to die.

"Can you run it?"

"Send me the dataset. I'll start tonight."

❦

He spent the afternoon in a state of controlled focus, the kind he'd perfected during outbreak investigations, the ability to function normally in meetings and briefings while a separate partition of his mind worked the problem underneath. He reviewed two routine monitoring reports, attended a video conference on enhanced population health outcomes in Sub-Saharan Africa, and responded to eleven emails, none of which mattered.

At 4:15 p.m., he transferred the flagged dataset to Lena's secure partition with a one-line annotation: Downstream variance in enhanced immune cohort. Not consistent with known off-target profiles. Request independent verification before formal flag.

At 4:30 p.m., he reopened the CRISPR-X system architecture document and began reading the section he'd skipped that morning, not the post-edit monitoring protocols, but the update mechanism itself. The adaptive therapeutic update system. The feature that allowed CRISPR-X to refine its edits over time, pushing corrections to recipients' cellular machinery the way a software platform pushed patches to devices. It was, according to the documentation, a one-directional process: the central algorithm analysed aggregate outcome data from the global recipient pool, identified optimisation opportunities, and deployed targeted micro-edits to improve therapeutic efficacy. Each update was logged, verified, and approved by a joint oversight committee before deployment.

Daniel read the update logs for the last six cycles. Clean. Documented. Approved. Every micro-edit tracked and accounted for, each one a minor refinement to disease correction efficiency or immune response calibration. Nothing in the logs suggested anything that could account for the downstream variations he'd found.

But the logs only recorded what the system reported. And the system's own quality controls hadn't flagged the variations either.

He closed the document and stared at his reflection in the dark screen. The face that looked back was forty-three, lean, composed. A face built for long surveillance, patient, still, trained to wait for the data to speak. He'd sat in field stations in Lagos and Dhaka and São Paulo, watching contagion curves climb on laptop screens while governments debated response protocols. He knew what the early stages of a crisis looked like. Not the explosion, the silence before it. The period when the numbers were wrong but not yet wrong enough for anyone to act.

The numbers were wrong.

He didn't know why. He didn't know how. But the immune markers were climbing in a pattern that no approved edit could explain, and somewhere in the CRISPR-X architecture, beneath the clean documentation and the verified logs and the approved update cycles, something was generating changes that the system either couldn't see or didn't want to report.

He logged out at 5:50 p.m. The evening shift was settling in, fresh analysts sliding into chairs still warm from the day team. The central holographic display flickered to life as someone ran a demo for a visiting delegation, a rotating model of the human genome, edit sites glowing in therapeutic blue, the whole thing spinning slowly like a planet made of light.

Daniel took the lift down to the lobby, walked out into the Geneva evening, and felt the cold air hit his face like a correction.

2.3 million people had been edited.

The system said everything was fine.

The data said otherwise.

And he was the only one looking.

❦

His apartment was a ten-minute walk from the IBA campus, a clean two-bedroom with a view of nothing in particular. He'd moved to Geneva eighteen months ago, arriving three weeks before CRISPR-X received its global deployment licence. The apartment still looked temporary, functional furniture, bare walls, a kitchen that held coffee, eggs, and not much else. He'd told himself he'd settle in once the initial monitoring phase was complete. The monitoring phase hadn't completed. It had simply become the job.