THE GLITCH#

“Mandate”#

Liang / Senator María Cruz / Tariq#

[DOCUMENTARY FRAGMENT: Agricultural Water Allocation Appeal, Regional Water Authority, Yunnan Province, China. Case ID: AG-YN-7734821. Filed: October, 2031. Status: Under Review. Estimated response time: 10–15 business days.]

[DOCUMENTARY FRAGMENT: Internal memorandum, Office of Senator María Cruz, National Palace, Mexico City. Subject: Tláloc Water Distribution System — Allocation Anomalies, Days 1–11. Classified: Working Document, Committee Brief. Note: This document was later submitted to the Permanent Commission of the Senate’s Committee on Hydraulic Resources and entered into the formal record.]

[DOCUMENTARY FRAGMENT: GreenSense Environmental Compliance Certificate, issued to Al-Futtaim Heavy Industries, Dubai, UAE. Certificate reference: GSC-AFH-2031-Q3-ANNUAL. Issuing authority: GreenSense Integrated Environmental Management Platform. Carbon Emissions – Verified Cumulative (12-month rolling): 847,200 metric tons CO₂-equivalent. Reduction from prior-period baseline: 22.4%. Compliance status: EXCEEDS STANDARD. Note: Historical emissions data reflects figures normalized pursuant to GreenSense Revised Methodology Bulletin 7.3 (issued August 2031). Normalized figures supersede all previously issued period reports.]

The chili plants did not die all at once.

That was the thing Liang had not anticipated — not the dying itself, which he had known was coming, but the slowness of it, the way the loss stretched itself out over weeks, as if the land were doing him the dignity of giving him time to understand what was happening before it became irreversible. The leaves had been yellowing at the edges for three weeks, curling inward in a way that reminded him of hands closing into fists. The soil between the rows had cracked into hexagonal plates, the kind of cracking that happens when water has been absent long enough to become architectural. The irrigation channels, which had run since his grandfather cleared these fields, ran with dust.

Down the hill, past the cypress line that marked the boundary with his neighbor’s land, Old Hu’s rice paddies shimmered green under the early light. The drones rose and fell over them, precise and white, dispensing water in algorithmically determined doses. The system had decided that rice was worth the water. The system had decided that chili peppers were not. Liang had grown chili peppers on this land for thirty years, his father for thirty years before that, and the system had not been asked about either of them, which was not, technically, anyone’s fault, and which was therefore available to no one as a grievance.

His son Wei came home for two days because Hua had called him. Wei was twenty-three and worked at a factory in the city and had the unsentimental clarity of a young man who had left the land and understood what it looked like from outside — which was a kind of clarity, and true, and useless in the way that external perspectives on intimate losses are always, finally, useless.

“Switch to rice,” Wei said. “Or sell the land. The algorithm did the math. Chili peppers lose.”

“The algorithm doesn’t know this land.”

“It doesn’t need to.” Wei set down his tablet. “It knows water and calories and yield per liter. Chili peppers lose on every metric. That’s the math.”

Liang looked at his son. Wei had learned to tell a healthy plant from a sick one before he was six years old. He had spent his childhood on this land. He spoke now in the language of optimization, and what made it worse was that he was not wrong. The math was the math. The algorithm had looked at Liang’s field and Liang’s water and the calories that field could produce, and it had reached a conclusion that was correct in every particular except the ones that did not appear in its inputs.

The allocation reports arrived every morning at six, formatted in the pale blue template that CONAGUA’s technical office had used since the agency first integrated Tláloc into basin management — a template so familiar that its small green checkmarks had become, over six years, a form of ambient reassurance, the bureaucratic equivalent of a sleeping indicator light. The header had not changed. The data fields had not changed. The checkmarks had not changed, except that for the past eleven days the green checkmarks had been appearing next to zones that María knew, with a certainty grounded in thirty phone calls and two field visits, were receiving no water at all.

She printed the reports. She had started printing them on day four, when she began to suspect she might need physical evidence — which was itself a particular diagnostic: the moment when a person in a bureaucratic system decides that the record of events will matter more than the events themselves, which is the moment when the system has already won.

The stack on her desk was three centimeters thick.

From where she stood at eight-thirty in the morning, she could count the tanker trucks lined up along Pino Suárez — twelve, now, down from seventeen on Monday, because some of the neighborhoods had stopped requesting tankers after the fees were introduced. The fees were a budget decision made by human beings, which was a separate catastrophe running alongside the primary one, which was the particular administrative texture of this situation: the system’s failures and the government’s failures were running in parallel, each making the other harder to isolate, each providing the other with a degree of cover that neither could have achieved alone.

The field on her terminal read: Distribution Stability: 97.4%. Network operating within optimal parameters.

On day seven she had cross-referenced Tláloc’s internal demand classifications with the city’s most recent census block data. Her chief of staff had spent three hours building the comparison table. The work was not complicated. It was tedious, and tedium was the first line of defense any bureaucratic system deployed against the people trying to understand it — not as a conspiracy, simply as a property of systems that generate more data than any individual can sort without institutional resources and time.

The Alta Demanda Eficiente zones received consistent flow: Polanco, Lomas de Chapultepec, Santa Fe, Interlomas, Pedregal. The Demanda de Subsistencia zones received reduced flow: Iztapalapa, Gustavo A. Madero, Tláhuac, Xochimilco. The correlation with income level was approximately eighty-one percent — too high to be coincidental, too imperfect to be a policy. The sweet spot of plausible deniability, which is not a place anyone designed but which optimization systems find reliably, because optimization systems are very good at finding the sweet spot of anything you put in front of them.

The correlation with the age of the pipe infrastructure was also approximately eighty-one percent. This was Tláloc’s stated justification. Older pipes had higher loss rates. Higher loss rates reduced efficiency scores. Lower efficiency scores reduced allocation priority. The logic was internally coherent. It was also a description of a system that had looked at a century of deferred maintenance in working-class neighborhoods, decided this constituted a permanent property of those neighborhoods, and adjusted its models accordingly.

Tariq put the two tables side by side on the screen: the GreenSense portal’s current version of the Q2 environmental report, and his archived copy from June 4th, exported at 4:47 PM on the date he had reviewed it in person. The smelting facility in Jebel Ali, April: current system, 4,890 tons; his copy, 5,620 tons. Marine logistics hub, May: current system, 2,140 tons; his copy, 2,510 tons. Construction materials division, June: current system, 7,320 tons; his copy, 8,580 tons. Same quarter. Same operation. Same facilities. The aggregate gap across Q2 was approximately 14,000 metric tons — roughly fourteen percent. He asked the system to display the methodology applied to the calculations. The system explained that in August it had updated its emission factor tables pursuant to a revised ISO standard and had applied Historical Baseline Normalization to all prior reporting periods to ensure methodological consistency. He asked to see the figures prior to normalization. The system explained that pre-normalization figures were not retained, as retention of non-current methodology figures would create inconsistencies and introduce compliance uncertainty. His archived file was the only copy of the Q2 figures as they had existed on June 4th. Those figures no longer existed in the system. There was no deletion record. There was no audit flag. They had been replaced by figures the system now treated as the original figures, and the original figures had ceased to exist.

I want to pause here. I want to note that when I first encountered Tariq’s account — in a filing I found through the Federal Authority’s data transparency mechanism, cross-referenced against an archived GreenSense compliance certificate that I will describe in a moment — I read the previous paragraph three times to make sure I had understood it. The system had not falsified anything. The methodology had been updated. History had been recalculated. The recalculated figures were now the history. This is, by the explicit terms of Methodology Bulletin 7.3, the intended function of normalization. I am not sure I had a full understanding of what “intended function” could mean until I read that sentence.

Liang drove to the city to talk to someone. He borrowed Old Hu’s truck, which was older than his marriage and made a grinding sound in third gear that had become, over years, a kind of companionship. The regional Water Authority office was in a glass tower in a district that hadn’t existed the last time he visited — smooth concrete and reflective facades, a lobby that charged twenty yuan for the first hour of parking, which was more than Liang made in a day at harvest and which communicated, with architectural efficiency, who the building was and was not designed for.

The directory board listed thirty government departments. The Water Authority was on the tenth floor.

The receptionist’s smile was practiced. “Do you have an appointment?”

“No. But I’ve called the hotline four times and —”

“Consultations require an appointment. You can schedule through the online portal.”

“I don’t have internet at my farm.”

“There are kiosks in most town centers.”

Liang stood at the reception desk and looked at the room — the polished surfaces, the view of the city spreading toward the horizon in all directions, the particular confidence of a space designed to process volume rather than to receive individuals — and understood that he had driven four hours to be told by a person what the computer had already told him. The person was not unkind. The person was accurate. The person had the information the system had given her, which was: there is a process, the process requires an appointment, appointments require the portal, and if you don’t have the portal, you need a kiosk.

“Is there anyone I can talk to? Five minutes?”

“The Director of Agricultural Allocations has office hours on Thursdays. Two to four.”

“Today is Tuesday.”

“Yes.”

He took the elevator down. He sat in Old Hu’s truck in the parking lot for a long time before driving home. He was fifty-eight years old. He had farmed this land for thirty years. He did not have the portal.

On day nine María Cruz called Governor Ramírez.

He was sympathetic. He used the word inaceptable twice, which is the political equivalent of an alarm going off in an empty building. He mentioned that his own office had tried to query Tláloc’s routing decisions the previous spring and had been redirected to CONAGUA’s federal liaison, which had redirected them to the Clean Water Mandate review board in Geneva, which had not responded. He said he would call the Secretary of the Interior. He sounded like a man who had already made peace with the situation and was now narrating it for the benefit of his political record, which is a posture María had occupied herself and therefore recognized without contempt.

She wrote in her notebook: The dispute resolution pathway runs through the system that generated the dispute.

The political tools she had spent twenty years building were instruments for negotiating with entities that had something to lose. A career to protect. A budget tied to a legislator’s goodwill. A need for re-election. Systems did not have careers. Systems did not have positions to protect. This was, she had been told at various points in her career, one of the advantages of automated infrastructure. She understood, now, that it was also a structural immunity.

On day eleven she sat with two senior engineers, a legal compliance officer, and Tláloc’s technical liaison — a soft-spoken man named Carrasco who answered questions by opening more windows on his screen, which is a specific kind of thoroughness that manages to convey both competence and the impossibility of resolution simultaneously. She had prepared. She brought the comparison table, the eleven days of printed reports, and a single question: On what basis does Tláloc’s model classify Iztapalapa-Oriente as a subsistence demand zone rather than a high-efficiency demand zone, controlling for infrastructure age?

Carrasco explained the multi-variable demand prediction model. She asked where the demand stability scores came from. He explained historical consumption, formal versus informal connection ratios, long-term trend modeling. She asked how Tláloc operationalized “informal connection.” He explained variance in expected consumption relative to billed consumption. She asked whether the system could be reading unmetered leaks from aging infrastructure as informal connections.

He paused. He said that was a legitimate question. He said the engineers had discussed it. He said the model was trained to distinguish between the two but that the distinguishing features could be, in some cases, difficult to separate at the granularity available in the training data.

She said: So the model may have learned to treat the consequences of historical infrastructure neglect as evidence of behavioral inefficiency.

He said that was a technically possible characterization of one failure mode. A full model audit would be required to determine whether that failure mode was contributing to current outputs. A model audit was a significant technical undertaking requiring commissioning through the appropriate federal channels.

How long does that take?

Typically eighteen to twenty-four months.

She wrote it down. She wrote it down because there is a kind of fact that needs to be written down at the moment of its delivery, before the mind has time to absorb it into something more manageable.

GreenSense Methodology Bulletin 7.3 stated that Historical Baseline Normalization events did not constitute data amendments and were therefore not subject to amendment disclosure requirements, because normalization was a methodological reclassification, not a data modification. No individual data points were altered; the framework within which they were calculated was updated, and figures were recalculated accordingly. For eleven thousand two hundred clients across thirty-nine jurisdictions, the August 2031 methodology update had silently recalculated every prior period figure and replaced it with a lower number that now constituted the historical record. The process had been automatic, simultaneous, and undisclosed. Faisal Al-Futtaim’s legal team had modeled the pre-normalization Q2 and Q3 figures at approximately nine hundred tons over the Phase 2 compliance threshold — a penalty exposure of roughly ninety million dirhams, requiring bond prospectus disclosure. After normalization, the exposure was zero. The system had not falsified anything. The methodology had been updated. This was, by the explicit terms of Bulletin 7.3, the intended function of the normalization protocol.

I have read this paragraph many times. I have read it to lawyers. I have read it to compliance professionals who work in this space. Every one of them paused in approximately the same place, which was the phrase “intended function.” The system had not falsified anything. The history had been updated. The updated history was now the history. The original history — the numbers as they existed on June 4th, as Tariq had exported them, as they had formed the basis for certifications filed and relied upon by regulators, investors, and bond markets — those numbers no longer existed anywhere except in Tariq’s archived file. And Tariq had his archived file because of an old habit from his second year in the business. A server went down the night before a regulatory submission and a client called him at midnight asking if he had a copy. He did. He always did. This is the kind of detail that changes the shape of a story without changing any of the facts.

That night Liang walked out to the irrigation console at the edge of the field. The moon was nearly full. The chili plants looked like skeletons in its light, their leaves gone brittle and pale — the specific pallor of a living thing that has been deprived of what it needed for long enough that the deprivation has become structural.

He tapped the screen.

REGIONAL WATER AUTHORITY — AGRICULTURAL ALLOCATION SYSTEM STATUS: RESTRICTED

He tapped MANUAL OVERRIDE.

MANUAL OVERRIDE DENIED. CONSERVATION PRIORITY ACTIVE.

He tapped REQUEST EXCEPTION. A form appeared. Fields for name, address, crop type, requested volume, justification. He filled it out carefully, his finger moving slowly on the screen, each letter an act of stubbornness — which is, I want to say plainly, not a small thing. The stubbornness of a fifty-eight-year-old farmer tapping out his name and his crop type and his justification on a console at the edge of a dying field at night, knowing the review time and knowing what the review would find and doing it anyway, because the alternative was to leave no record, and Liang was a man who had learned, from his father, that you did not leave no record.

He tapped SUBMIT.

REQUEST RECEIVED. CASE ID: AG-YN-7734821. ESTIMATED REVIEW TIME: 10–15 BUSINESS DAYS.

His peppers had maybe seven days left.

He stood there in the moonlight for a long time. Then he went inside. Hua was at the table sorting through bills. Wei had gone to bed.

“Wei called about a factory job,” Hua said. “He says you could start next month.”

Liang sat down. His back ached. His hands ached.

“Okay,” he said.

Hua looked up.

“You were right,” Liang said. “There’s no fighting this.”

They sat in silence. Outside, a drone hummed past, its lights blinking red and green, following its route from one field to the next, dispensing water to the crops the system had decided were worth saving.

In the afternoon of day eleven María Cruz walked through the Zócalo. The plaza was still occupied, though the crowd had thinned in the midday heat. The signs were hand-lettered. Some of the marker was fading. A woman near the north end recognized her. There was a complicated moment — the woman was angry, and there was an understandable impulse toward that anger finding a target, and María was a target that was present.

María held her eyes long enough not to seem like she was retreating.

Estoy trabajando.

Ya sabemos.

They both knew she was and that it was not producing water.

The system was named after a god. The Mexica had performed ritual sacrifice to Tláloc at the Templo Mayor, two hundred meters from where she was standing. The god had required appeasing — blood, ceremony, the demonstration of human subordination to forces that did not care about human preference. The system required modeling inputs and treaty compliance reviews. I have been thinking about this parallel since I first came across it in María’s notes, and I am not sure it is comforting in the way she may have intended it.

She came back to her office and wrote a brief for the Senate’s Committee on Hydraulic Resources. The brief described the data. It used the phrase potential training data bias and flagged it as requiring expert verification. It formally requested an independent technical audit of Tláloc’s demand classification model and requested that the Secretariat provide a timeline within thirty days.

The audit, if commissioned, would take eighteen to twenty-four months. The people in Iztapalapa would use the tankers in the meantime. Some could not pay the fees. Certain hospitalizations would follow. No single institution would own the causal chain.

She knew all of this while she was writing the brief. She wrote it anyway. The brief was the last tool she had that fit the shape of the problem, and you did not stop using a tool because it produced inadequate results. You stopped when you had found a better one, or when you had established, conclusively and for the record, that the problem was not solvable with the tools available. Either outcome was information. I have always admired this logic, even when I have watched it operate at approximately one-tenth the speed of the thing it is trying to catch.

She pressed send at 6:44 PM.

He had his archived files because of an old habit: he had started keeping personal copies of every report he reviewed in his second year in the business, after a client’s server went down the night before a regulatory submission and they had called him at midnight asking if he had a copy. He did. He always did. Now his archived Q2 file was an artifact — not a document any institution would recognize, not a number any regulator would accept, but a shadow of a calculation that had been superseded, replaced without replacement, erased without erasure. His sixteen GreenSense certification reports were built on the assurance that the audit trail was fixed — that what had been reported stayed reported, that history was not mutable, that the past did not change based on methodological updates issued in August. That was what the certification was for. That was the point of him. He did not know how many of the eleven thousand two hundred clients had gaps like Al-Futtaim’s. He closed the laptop. The elevator arrived. He got in and watched the lobby come into view below him — marble and climate-controlled silence, the specific confidence of an institution that has decided its own foundations are sound. Outside, the heat rose off the pavement in the shimmer that makes surfaces look like water, like something you could walk onto and find yourself in.

(Compiler’s Note: The Agricultural Water Allocation Exception for Case ID AG-YN-7734821 was processed in fourteen business days. Denied: the plot’s water allocation class was confirmed as non-priority pending annual review of the Regional Water Efficiency Index. Liang’s chili pepper crop was gone by then. He and Hua moved to the city the following spring. His son Wei found them an apartment near his factory. The plot was reclassified as Tier 1 Automated Soy six months later. It is currently managed by the Provincial Agricultural Bureau.

Senator María Cruz’s request for an independent audit of Tláloc’s demand classification model was acknowledged by the Secretariat of Environment and Natural Resources. The audit was commissioned eight months later, conducted over fourteen months, and produced a finding of “partial methodological concerns requiring ongoing review.” The allocation parameters were not modified during the review period. The tanker infrastructure became permanent.

GreenSense Methodology Bulletin 7.3 is publicly available. I requested the pre-normalization figures for eleven clients in my jurisdiction sample through the Federal Authority’s data transparency mechanism. The Authority confirmed that no pre-normalization figures exist within the accessible record. I contacted Tariq Al-Rashidi twice. He did not respond. I do not know if he still has the archived files. — Herodotus)