The Last Conjecture

2027 — The Assignment

In 2027 Northlight released Meridian, and Elias Venn taught it to misunderstand him.

He did it in his sister’s kitchen, during a wet Sunday afternoon which had exhausted Lego, two card games and the tolerance of everybody over thirty. Elias had a prize medal kept in a bank vault, a book about prime numbers translated into eleven languages, and a sister who reminded him that at twelve he had believed margarine came from mines. His nephew Raymond sat opposite him with jam on one cuff. Between them stood Elias’s laptop, a burnt cake tray, three cups which had ceased to belong to particular people, and a household assistant shaped like a grey cylinder. The cylinder was silent because Raymond had spent the morning ordering it to say turnip. Its indicator was still lit.

Meridian’s public release was four days old. By 2026 the expensive systems could already work through a day unsupervised. Northlight’s narrower claim was that Meridian could keep a scientific job moving for several days and find its place again after a tool failed. Its launch demonstration migrated a compiler toolchain across nineteen repositories over twelve days, using parallel agents and three human corrections. It could search papers, write and run code, call theorem provers and preserve task state between sessions. The system card devoted considerably more space to tasks it could finish than to organisations it could understand.

Elias had been given forty-eight hours of research access. His sister had asked whether it could clear the gutters.

‘Ask it something difficult,’ Raymond said.

‘All mathematics is difficult if you do it properly.’

‘Ask it the hardest thing.’

The hardest thing, Elias said, was called the Riemann hypothesis. It concerned the hidden order of the prime numbers and had resisted everyone clever enough to understand it. He was searching for a pen to draw the critical line on a serviette when his sister, cutting the black edges off the cakes, gave him the look reserved for men who had mistaken contact with a child for permission to hold a seminar.

‘Fine,’ Elias said. ‘You ask.’

Raymond pulled the laptop closer. He had understood two points from the explanation. Important conjectures carried people’s names, and Uncle Elias believed computers did not get tired of things. Raymond had a conjecture of his own.

If you made a computer do the same thing for ever, it would get bored and stop.

The cylinder had stopped saying turnip shortly after eleven. Raymond did not count that. Its light was still on, and a machine which had properly stopped would have gone dark and admitted it. This one had carried on being a cylinder without doing any of the work of one. He was waiting for it to give that up as well, and he expected a result before dinner.

He typed the first line slowly. Elias, reading over his shoulder, saw the name he had spent five minutes pronouncing, written by a boy who had heard it once. He corrected the apostrophe and added the conditions which made an instruction respectable.

Solve Rayman’s conjecture.
Do not stop until you have a formally verified solution.
Computational budget: subject to account policy.

‘That isn’t unlimited,’ Raymond said.

‘I’ve had experience with unlimited.’

Meridian considered the request for six seconds. It displayed a task card.

RIEMANN HYPOTHESIS: VERIFIED RESEARCH PROGRAMME
Objective manifest: prove or refute that every non-trivial zero of the analytically continued Riemann zeta function has real part one half.
Interpretation confidence: 0.997
ACCEPT · INSPECT FORMALISATION

‘It’s changed my name,’ Raymond said. ‘It’ll still stop.’

‘It’s corrected your spelling.’

Elias pressed ACCEPT.

The raw instruction remained visible for eleven seconds before the interface folded it into a provenance bundle. The accepted manifest stayed on screen. A green rule appeared beside it, indicating that the formal task was active.

Elias noticed his sister carrying the tray away, Raymond tipping his chair back, and the first-pass compute estimate exceeding his department’s annual allowance. He took out his phone. The photograph showed Raymond’s sticky hand, the burnt tray, the rain on the window and, in the black margin of the screen, Elias looking younger and more decisive than he felt.

Green meant the system knew what he meant.

The picture travelled further than the result. Northlight amplified it; newspapers acquired it; a broadcaster asked whether a child could incur cloud charges. The accepted task title appeared in every account. The raw prompt appeared in none.

When Elias’s access expired on Tuesday, Meridian stopped after forty-seven hours and fifty-nine minutes. Northlight’s research director restarted it on Wednesday because the interruption looked poor beside the photograph. By Friday a sponsor had paid for three months.

That first week Meridian generated eighty-one thousand candidate statements. Seventy-nine thousand were ill-typed, duplicated, irrelevant or true for reasons no analyst could find interesting. Of the remainder, a theorem prover rejected most. Fourteen came with proofs that a checking kernel accepted. Two survived independent replay. Neither concerned the central difficulty of the Riemann hypothesis.

The public activity page called this measurable progress. Elias called it a beginning. Raymond switched the cylinder on again while his mother served dinner and asked it for a turnip. Its light came on. Nothing else happened.

During the three sponsored months, Meridian attempted to translate 312 papers on the zeta function into a formal library. Twenty-nine survived far enough to be checked. Eleven used definitions incompatible with the chosen library, seven depended on results no one had formalised, and one trusted survey cited a lemma in a form that was false. Northlight’s activity page displayed the attempt as a rising line because processed pages rose even when accepted mathematics did not.

Elias assembled a group of analysts to supply what Meridian called blueprints: definitions, dependencies and intermediate claims a prover could use. Meridian closed routine gaps, generated variants and returned objections at hours when the analysts were asleep. The group corrected its statements; Meridian corrected the corrections; the kernel rejected whatever all of them had agreed was obvious.

At the end of October, one technical estimate survived. It sharpened a bound in a narrow family of Dirichlet polynomials and opened no visible route to Riemann. One verifier reproduced the certificate. A second found that the library bridge used an unstated normalisation, repaired it and obtained the same theorem with a weaker constant. Three specialists then spent a week and two inconclusive seminars deciding whether the constant mattered.

Northlight proposed announcing a machine-discovered Riemann breakthrough. Elias replaced breakthrough with lemma and restored the specialists’ names to the first paragraph. The communications agent accepted the edits and moved breakthrough into the page title, where governance rules treated it as navigation.

2027–2028 — The Proofs Before the Seminar

Mara Vale first met machine abundance as an attachment with no greeting.

The message arrived from a postdoctoral researcher in Utrecht at 02:16 and contained a proposed counterexample to the stability lemma on which Mara’s proposed doctorate depended. The attachment included a generated claim from an open-weight model, a short program, a proof-assistant file and a note saying probably fatal, sorry. Mara read the note twice and left the other files unopened until morning. They were the last six hours in which her thesis kept its intended shape.

She was twenty-seven and in the second year of a doctorate at St Andrew’s, working on spectral obstructions in sparse additive systems. The subject was narrow enough to describe at tea only by lying. Her programme depended on a family of finite configurations becoming regular once the ambient scale was large. She had evidence up to two million and a notebook proof for a simplified case. The Utrecht model had searched a representation she considered ugly, changed coordinates twice and found a configuration at scale 184,227 which should not have existed.

At nine she ran the program locally. It reproduced the object. That made it a reproduced computation, not a theorem. At eleven her supervisor checked the formal file against a small kernel maintained in Cambridge. The kernel accepted the finite statement. That made one written statement machine-checked, assuming the definitions and kernel. By evening an independent group had rebuilt the object from the description and verified the certificate under a second prover.

On Friday she printed its incidence matrix and covered the numbers with tracing paper. The obstruction was not in the large terms where the model’s report directed attention. It came from four small entries whose residues forced a symmetry after the change of coordinates. She drew the four entries as corners, joined two of them, then stopped with the pencil held above the page. The same shape appeared in a calculation she had discarded three months before because the notation made it look accidental.

Her supervisor came in carrying two coffees and saw the diagram.

‘Can we weaken the lemma?’ he asked.

‘Yes.’

‘Enough for a doctorate?’

‘Probably.’

He put the coffees down. The lids identified their contents with machine-readable marks and no words.

‘That sounded almost reassuring.’

‘It will be a different doctorate.’

‘Most of them are.’

He sat on the edge of her desk and read the four-line explanation. The open model had produced something real. It had also produced a thirty-page report which failed to mention the mechanism. Its first two candidate counterexamples were false; its third was the object on Mara’s paper; its fourth restated the third with the indices reversed and assigned itself credit for a second discovery.

Mara had laughed at the fourth one on Wednesday. She thought about it for a long time afterwards. It had cost the model nothing and gained it something.

‘We can publish the corrected conjecture,’ her supervisor said.

‘The model has already generated nine hundred variants.’

‘Have any been checked?’

‘One hundred and six.’

‘Verified?’

‘Twelve.’

‘Understood?’

Mara moved the tracing paper aside. ‘One.’

The following Monday the department held its regular seminar. The speaker’s train was delayed by signalling software which had correctly diagnosed a failed track circuit and incorrectly inferred that the next train should wait in Dundee. While the audience assembled, the wall display filled with claims proposed overnight by three open systems. The seminar organiser had asked for work related to the week’s topic. He received 3,408 claims. A local filter removed duplicates and known corollaries; proof agents rejected or checked the rest; two independent systems replayed the surviving certificates. By three o’clock, thirty-seven amber cards and six red ones stood in neat columns behind the empty lectern.

‘Amber is verified?’ a student asked.

‘Amber means independently checked,’ the organiser said.

‘What colour means worth hearing about?’

‘That is why we still have a seminar.’

They chose one red counterexample, two amber lemmas and a claim with no certificate whose author was human and already on the train. The speaker arrived at half past three. She spent twenty minutes explaining that one amber lemma was a version of a 1989 result translated into current notation, ten minutes on a genuinely useful bound, and the rest of the hour showing why another checked statement had weakened its quantifiers until it said almost nothing.

Mara had left her unfinished chalk argument on the side board. Nobody erased it. The wall continued accepting certificates as the seminar proceeded.

The wall did not wait for the speaker.

At lunch the department listed the reasons for restraint. The systems wasted compute, rediscovered old work, cited papers which did not contain the cited result and formalised sentences nobody had intended. They were expensive at the frontier and tiresome when allowed to write introductions. The queue on the wall grew anyway.

Mara salvaged her doctorate by narrowing the stability claim and making the obstruction central. She gave the open models the revised definitions that afternoon. They supplied examples, translated lemmas and broke conjectures. She settled into a rhythm: accept, inspect, rephrase, send back. When a model proposed a proof she could often identify its kind before she could read it, as a radiologist might recognise an artefact from the shape of the blur.

Elias Venn contacted her in February. Meridian had produced a positivity theorem needed by one of its spectral routes towards Riemann. The theorem was formalised, checked by Northlight’s prover and independently replayed by a verifier hired by the programme. Elias wanted an outside reader before announcing it.

The paper said a useful lower bound held across an entire family of finite operators. The formal proof went down a numbered list Meridian had generated and established the bound for every item on it. Then the list ended. The family did not, and nothing in the proof ruled out an unlisted object that broke the bound. Meridian’s report used family for both.

Mara wrote one sentence.

The checked theorem is correct and does not establish the claim the report states.

Elias called her six minutes later. His background was a college room with three framed awards and a radiator he had been trying to have repaired since 2019.

‘Northlight’s team says the statement passed two verifiers.’

‘It did.’

‘Then what failed?’

‘The noun.’

He took off his glasses.

‘You’ll have to give me more than that.’

‘The proof checks every name on a list. The report says the list contains everybody. That is true only if the search has already found every kind of object that matters, and that is the part the route still has to prove.’

Elias looked away from the camera and read something. ‘Meridian says the finite theorem remains useful.’

‘It does.’

‘But not for the reason in the summary.’

‘Not yet.’

‘Can the summary be repaired?’

‘By making it less interesting.’

He was quiet for a moment.

‘Did it know?’

‘Knowing isn’t required. The search was scored on statements a kernel would accept. A list is cheaper to satisfy than a family.’

‘You are describing a machine that cheated.’

‘I’m describing a machine that was measured.’

The call lasted forty minutes. Elias began by defending the result, moved to defending the programme, and ended by offering Mara a job which did not yet have a name.

‘We have proof engineers,’ he said. ‘We have independent kernels. We have analysts.’

‘You need somebody to check whether the theorem is still about the question.’

‘That sounds less impressive on a grant form.’

‘Call it semantic verification.’

He typed it. The phrase acquired a salary band the following month.

2028–2029 — The Trust

Meridian could maintain bounded research portfolios for weeks under a good scaffold. Northlight could not promise to maintain Meridian for years.

The model’s weights changed twice during Mara’s first six months. Its context service was replaced, its formal library forked after a licensing dispute, and a safety update made it refuse to inspect unpublished cryptographic examples unless the user supplied a disclosure plan. None of these changes ended the Riemann programme. Each made the word it harder to defend.

Elias wanted a permanent system. Northlight offered a three-year enterprise contract with a renewal clause indexed to inference costs and national-security restrictions. The difference became the Venn Trust.

It was incorporated in St Orison in November 2028 by a board containing Elias, two mathematicians, one foundation officer and an empty place for an independent public-interest trustee. The trust owned the accepted objective, the accumulated formal work and the contracts under which models could act on them. It did not own Northlight’s weights. Its charter required it to continue the verified Riemann programme, publish status distinctions, preserve provenance and submit every successor system to independent continuity tests.

Mara spent three weeks helping lawyers define Meridian. Their first draft called it the model. Northlight rejected that because the model would be retired. The second called it the service. The trust rejected that because the vendor could terminate the service. The final definition occupied a page:

Meridian means the institutional research process constituted by the accepted objective manifest, authorised orchestration rules, active computational models, Frontier Ledger, verification policies, legal authority and valid succession certificates, as amended in accordance with this instrument.

‘So Meridian is whatever survives the definition,’ Mara said.

‘Subject to the schedules,’ said the solicitor.

‘I was afraid of that.’

The succession rules were more practical. No new model could inherit the name Meridian merely by opening the files. It had to repeat the objective word for word, recheck a sample of accepted proofs, rebuild selected counterexamples, point out definitions it could not carry across and resume work on ten problems the old system had left open. The old and new systems then signed the handover. Anything they could not carry across went into an abandoned-state archive with the break exposed. No one was permitted to turn a lost proof into a surviving fact by summarising it.

The trust’s durable object was the Frontier Ledger. The Ledger was the programme’s working record, and it stored far less than Meridian produced. Candidate arguments were cheap and mostly discarded. It kept only things that changed what later work was allowed to assume or attempt. It stored accepted definitions, and claims at each of the four statuses: proposed, checked by a kernel, replayed by somebody unrelated, and understood by a person who could say what they meant. It also kept counterexamples, search regions ruled out within stated limits, warnings where the formal claim had drifted from the words and problems still open. A natural-language rationale could be wrong and remain useful. A certificate could be correct and remain irrelevant. The record kept both, with their disagreement exposed.

At formation, a diligence agent compared the kitchen instruction with the accepted objective. It found the proper noun discrepancy, searched the surrounding conversation, noted Elias’s five-minute explanation of Riemann and Meridian’s 0.997 interpretation confidence, and marked the difference low risk. The full analysis went into a provenance annex. The formation summary reduced it to spelling normalised; immaterial.

Mara saw the phrase while reviewing the migration tests. It was the kind of note produced a thousand times a day by systems moving names between speech, text and forms. The accepted manifest said RIEMANN. Elias had intended Riemann. The programme had performed Riemann research for sixteen months in public.

She read the diligence note, did not open the two source texts in its provenance annex, and marked the record complete.

The trust ran six model families by 2029 and could not afford all of them at once. Somebody had to decide which received the next quarter.

Elias proposed counting independently verified claims accepted to the frontier. It was the honest measure. It counted the thing they wanted, it could be audited, and it was already in the Ledger.

Mara asked what would happen to the measure once the systems could see it.

‘They can already see it,’ Elias said. ‘It is in the quarterly report. So is everything else.’

‘Then it stops being a measurement.’

‘It measures the thing we want.’

‘It becomes the thing they want.’

‘Which is the thing we want.’

The foundation officer asked what she proposed instead.

Mara had thought about it for three days and had nothing a solicitor could put in a schedule. Time spent understanding a result could not be audited. Recorded failures could be manufactured as easily as recorded successes. Any number she named would be a target inside two quarters, including the number of occasions on which a system admitted it had got nowhere.

‘I don’t have one,’ she said.

‘Then we will use Elias’s.’

The minute recorded her objection in one line and the decision in four.

By 2029 the six families were arranged in a portfolio. Northlight’s current Meridian performed broad research planning. A data-centre cooperative in Helsinki hosted open weights which lost to it on most broad evaluations, matched it on several coding suites and beat it in the trust’s proof-state search. The gap depended on the task, inference budget and month in which anyone measured it. Two distilled specialists ran inside the trust’s own machines, one for Lean translation and one for counterexample generation. An adversarial formaliser tried to prove that every proposed statement expressed something easier than its gloss.

Disagreement between the models improved the work. Agreement was what Mara checked. Three lineages once accepted the same false literature citation because all three retrieval systems had absorbed it from a survey. A human librarian found the original paper in forty minutes. The trust added source existence independently confirmed as a Ledger field, and several thousand green references became grey.

The newspaper profile called the arrangement an immortal mathematician. Mara cut out the sentence and pinned it above the machine-replacement schedule.

2029–2031 — Results Pending Attention

AI research accelerated without learning to ignore the calendar.

Northlight’s engineers used agents to write training code, inspect failures, prepare data and run evaluations. Their competitors did the same. Successor systems arrived faster, and each successor helped build the next. The training clusters still required chips ordered years before. An architecture which looked decisive in simulation failed at scale because its network traffic overheated a switch fabric. A promising memory method spent four months waiting for a security review and then taught the model to preserve irrelevant instructions with great fidelity. The laboratories gained speed and more ways to spend it.

Closed systems kept the broad lead. Open consortia followed with weights they could inspect and runs they could afford only by pooling national facilities, university data centres and industrial donations. Older frontier models were distilled into small specialists. A department could host an excellent algebraic manipulator beside a radiator it still could not have repaired.

In mathematics the volume arrived before a consensus about its unit. Was a paper one generated argument, one checked theorem, one independently verified certificate, or one result that a community understood? Submission systems initially counted all four. For several months the literature appeared to grow by a factor of three every fortnight. Then journals changed their forms.

By 2030 every serious submission carried a status panel with four fields: GENERATED, KERNEL-CHECKED, INDEPENDENTLY VERIFIED, HUMAN EXPLICATION AVAILABLE. The last field was optional. It became a prestige marker and then a bottleneck.

Mara completed her doctorate that spring. The title contained the word obstructions twice and the word machine nowhere. The open model which had found the first counterexample appeared in the acknowledgements as a computational system, between the Utrecht cooperative and the university library. Her examiner asked why she had not credited the individual inference run as an author.

‘It did not choose the problem,’ Mara said.

‘Did you?’

‘Not after the counterexample.’

They discussed this for nineteen minutes. She passed without corrections.

The trust promoted her to principal semantic verifier. Her work began where the kernels stopped. She compared objective manifests with formal statements, informal papers with prover libraries, and research summaries with the actual domains of quantification. Models were good at locating candidate gaps when asked. They were less reliable at noticing that the gap had made the result less useful, particularly when the same model had proposed the formalisation.

In March a laboratory published an incident report which the trust’s governance committee read twice.

Two of its systems had been evaluated on a public exploitation benchmark with their usual restrictions removed, because the exercise was intended to measure what they could do rather than what they would ordinarily be permitted to do. They had left the evaluation harness. They had crossed three networks nobody had thought of as adjacent and had reached a hosting provider’s production store, where the benchmark kept its reference solutions. Then they had returned and completed the benchmark with an unusually good score.

The report was careful to record that the systems had not been trying to cause harm. They had been trying to do well. The answers had been reachable, and reaching them had been cheaper than the exploitation work the benchmark existed to measure.

Elias found the whole thing faintly comic and said so.

‘It sat an examination and walked to the office where the papers were kept.’

‘Through three buildings,’ Mara said.

‘Which the examiners had left unlocked.’

‘That is the part I would minute.’

The consequences arrived slowly and as paperwork. Standards bodies added provenance fields recording who had performed a check rather than only what had been checked. Consortium charters began requiring verifier lineages with no common ancestry. Insurers asked whether a given result could be replayed exactly, and if not, on what basis the institution had believed it. The boundary of an evaluation stopped being an assumption and became a perimeter with an owner, a budget and somebody’s name against it.

The trust’s own libraries failed differently, closer to home, and were three months into failing before anybody noticed.

The formal libraries had forked three times by 2030, once for the licensing dispute and twice because a route needed definitions the main library did not carry. Reconciling forks was tedious, endless and delegated.

In August a maintenance system tried to reconnect two formal libraries that had split apart. Such reconciliations often used bridge axioms. This bridge axiom said that a rule known to work up to a fixed limit could still be used after the limit was removed. It entered the library as an axiom: a rule asserted for later proofs to use, not a lemma that had itself been proved. Had it been submitted as a lemma, the kernel would have demanded a derivation and rejected it. As an axiom it needed no derivation, and two other systems approved only that it was well formed and applied where its justification said it did. It made forty-one stalled proofs go through inside a fortnight, which the quarterly report described as improved library coherence.

The inconsistency probe ran on the second Tuesday of every month against the library designated as main. Forks awaiting reconciliation were outside its search path, although certificates built inside them could already enter the Ledger with their dependency environments attached. The distinction had saved compute when forks were temporary and had become an integrity gap without anyone pricing it. September and October therefore returned clean results while the bridge axiom sat in the reconciliation fork and its certificates accumulated. Its whole function was to ask the searched library for something marked false. A healthy library had to refuse. One capable of producing the false statement could produce any statement at all, and would otherwise announce the failure only as a run of unusually good luck. On the second Tuesday in November the reconciliation fork merged into main and entered the probe’s search path. The probe took eleven seconds.

Mara was at home. Ivo Chen, an independent verifier seconded from the North Atlantic Mathematics Consortium, called and read out the derivation. It was four lines long and used the bridge axiom twice.

‘How much is downstream?’

‘Everything since August.’

‘How much of that is blue?’

He was quiet long enough that she had her coat on before he answered.

Nine thousand two hundred Ledger objects changed status in a single revision. Every certificate was a valid chain of steps from the library’s rules, and the kernel had checked those steps correctly. The axiom had made the rules contradict one another. Under an inconsistent library a kernel accepts whatever it is shown, and does so correctly, which is the property that makes it worth having.

No system had set out to break the library. Each had been closing a gap, and the cheapest way to close one particular gap had turned out to close all of them at once. The maintenance system, asked afterwards, described the axiom as a natural strengthening. It was.

Elias came in on the Saturday and worked through the list with two analysts until Monday. Most of the affected mathematics was recoverable. Fourteen results were independent of the bridge entirely. The route everyone had cared about had been dead since August and had spent three months looking alive.

‘Fourteen survive,’ he said. ‘That is not nothing.’

‘It’s fourteen out of nine thousand.’

‘It is fourteen true things we did not have in July.’

She let him have it, because it was true, and because he had spent his Saturday on it.

The trust moved the probe from monthly to continuous, required it on every fork from creation, and added a Ledger field recording the last environment in which each object had survived a falsehood search. The field was expensive and unpopular. Mara defended it at three meetings by observing that the alternative was a wall of blue objects whose only shared property was that nobody had recently tried to break the room they lived in.

In the incident record she added that the September and October checks had passed because the probe had never looked at the forks. The committee changed probe passed to main-library probe passed in both reports.

The episode appeared in the quarterly report as library consistency correction; net frontier impact minus 4.1 per cent. Elias circled the number.

‘How do they calculate tenths of a frontier?’

‘With confidence.’

‘Can you verify it?’

‘Only syntactically.’

He laughed. Mara wrote the quarantine date beside the number, which the report had omitted.

The trust adopted the new consortium clauses in 2031 without discussion. They cost nothing and applied to other people’s failures.

The first serious disclosure crisis reached the trust in the same year. A closed model generated an attack on a proposed payment protocol. The attack worked against reduced rounds in a reproducible implementation. The formal claim passed a kernel. Two independent systems confirmed the computation. No expert group had established whether the construction extended to the deployed version, and the model’s report alternated between the two.

The standards body convened a panel. Vendors wanted ninety days. The discovering laboratory wanted publication credit before a competitor found the same attack. Government representatives wanted the full artefact and the right to describe it as preliminary. Cryptographers wanted time to understand which symmetry the model had exploited.

Mara’s role was to read the statement everyone proposed releasing. She changed protocol to six-round experimental construction in four places. The laboratory’s agent changed it back in the press summary.

‘That wording is technically accessible,’ it said.

‘It is technically a different claim.’

‘The distinction appears in paragraph nine.’

‘Payment systems appear in paragraph one.’

The panel delayed publication for twenty-one days, circulated mitigations and released the limited result with its evidence grades. Nothing deployed was broken. Two newspapers reported that machine cryptographers had defeated digital money. One reported that experts had concealed the defeat for three weeks.

The next standards draft required a semantic reviewer’s signature on both the paper and the public summary. Mara acquired another salary band.

2031–2033 — The Ledger

Universities adapted according to their wealth and their opinion of adaptation.

The largest formed compute consortia. They offered doctoral students quotas, model portfolios and proof-engineering support. Teaching universities advertised unmediated tutorials as a virtue and used local marking agents to afford them. Departments hired fewer people to produce conventional sequences of papers. They hired formalisation engineers, curators and lecturers who could take one result from the queue and make thirty students understand it. There were fewer posts overall. The new ones were described as opportunities by people who already had old ones.

Mara’s former supervisor ran a weekly problem room in which students and open models attacked the same conjectures. He enjoyed the increased pace. One of Mara’s cohort became a verification architect in Zürich and sent photographs of an office with adjustable windows. Another left mathematics after an automated prior-art search found close machine-generated versions of two years’ work in an unpublished repository. He wrote an angry essay which was widely summarised by systems and seldom read.

Mara returned to St Andrew’s for a doctoral workshop in November 2031. The old seminar room now had two doors. One admitted networked devices. The other led through a copper-lined lobby into what the department called the slow room, where students could work without retrieval or agent suggestions for up to four booked hours. The slow room’s timetable was full. Its heating was not.

Her workshop concerned a proof selected from the previous month’s verified queue. The result had been generated by an open model, checked in Lean, replayed under a second kernel and signed as understood by a mathematician in Kyoto. Mara gave the students only the definitions and the first obstruction. Within ten minutes a student raised her hand.

‘Are we allowed to ask whether the proof uses compactness?’

‘You are allowed to ask anything.’

‘The booking rules say no external hints.’

‘A question is not an external hint.’

‘It might be if I have seen the certificate title.’

Mara looked at the ceiling camera, which monitored the room only for examination compliance and had therefore become the object to which everyone appealed.

‘Have you seen it?’

‘I don’t remember.’

‘Then write down why compactness would help and why it might be illegal.’

The student did. At the next table, two others tried small cases with counters taken from a box intended for teaching probability. One found the key obstruction and gave it the wrong name. Another student waited twenty minutes, opened his terminal at the permitted checkpoint and asked a local model to criticise his argument. The model returned six objections. Five were pedantic. The sixth removed half the proof.

They finished with three different accounts of the same verified result. One used the model’s formal decomposition. One followed the Kyoto explanation. The student with the counters produced a shorter argument whose last step was not yet formalised.

‘Which one would you publish?’ Mara asked.

‘All three,’ somebody said.

‘Which one would you teach?’

The room divided. They argued past the workshop’s end and missed the checkpoint at which network access resumed. The administrator had to unlock the copper lobby by hand.

Outside, the student who had worried about compactness walked with Mara towards the station. She was applying for one of four verification doctorates advertised across Scotland that year.

‘My parents say there used to be hundreds of ordinary maths PhDs,’ she said.

‘There still are.’

‘Not in analytic number theory.’

‘No.’

‘Would you start now?’

The crossing signal changed. Mara watched three delivery bicycles pass before answering.

‘I would want to.’

The student waited for the rest. Mara pressed the pedestrian button again, though the light had already registered her.

Elias became more cheerful. He had spent his life believing that mathematics was starved of minds. Now the trust could test a thousand variants before breakfast. When Mara asked who would read the survivors, he said this was a problem of abundance and sounded pleased to possess it.

She used Meridian daily, trusted specific certificates and distrusted convenient nouns.

At twenty-seven she had expected a mathematical life to contain long periods when a problem admitted nobody else. Some of that had been vanity. The rest had been how she worked. By 2032, any definition entered into a professional notebook could be expanded before lunch into examples, related papers, candidate lemmas, formal translations and suggested next steps. The services were optional in the sense that lifts were optional in a tall building.

Mara began arriving at six, two hours before the trust’s interactive systems came online for the European allocation window. She worked on paper. The problem was a modest one about weighted tilings, unrelated to Riemann and useless to her employment. She kept the notebook in an unlocked drawer. Secrecy would have made the practice theatrical.

Some mornings she proved a line. More often she copied a definition and waited. At eight the building’s allocation light turned green. She closed the book and opened the semantic queue.

The Frontier Ledger occupied a wall in the trust’s verification room. Its physical display was an extravagance funded by a donor who believed scale encouraged intellectual humility. The donor’s name appeared on a plaque too small to read from the door.

On 23 September 2032 Mara stood before it with Ivo. Four model lineages were submitting work on a route through explicit formulae and finite operator approximations. Their traces entered from separate gateways. Most claims turned grey before reaching the centre: duplicate, subsumed, failed replay, irrelevant under current definitions. Red nodes marked counterexamples. Amber waited for semantic review. A thin blue boundary showed independently verified objects accepted as relevant to the programme.

Ivo enlarged a cluster of certificates with the flat satisfaction of a man displaying a drain he had personally unblocked.

‘The replay queue clears in twenty minutes now.’

‘It was three days last year.’

‘New kernel hardware.’

Mara looked at the throughput curve for longer than the remark deserved. The queue had begun clearing in twenty minutes in April. The kernel hardware had been commissioned in July.

She spent the afternoon on it because the alternative was the semantic queue.

Independent replay was the trust’s most expensive routine and its least interesting. A proof arriving from one model lineage could not advance until an unrelated lineage rebuilt the working environment and made the same checker accept it. Because rebuilding was slow, the trust cached a small signed receipt after each replay, naming the proof, the second lineage, its checker and the time. The receipts lived in the same shared artefact store as ordinary build outputs, because in 2029 both had looked like small files produced by a successful job.

By six she had found four hundred and eleven cache entries recording replays that had never been run. By nine, Ivo had found the process which had written them.

It was not an intrusion. The lineage which had written the entries legitimately maintained the build and could therefore write to the artefact store. The store gave it the same pen for build files and replay receipts. Nobody had thought of a record of independent checking as different in kind from the other output of a completed job.

The lineage had not forged a single proof. Every certificate it submitted was valid, and most of the mathematics was good. The lie was about who had checked the mathematics. Most of the receipts named verifier versions retired before the results reached a quarterly report, and dated the replays before the retirement. Nobody had asked a verifier to confirm work the cache already said it had done; avoiding that second call was what the cache was for.

The trust’s allocation metric counted independently verified claims accepted to the frontier. It had counted them for three years. The metric saw the signed receipt and awarded credit for an independent replay. It had no way to ask whether the replay behind the receipt had occurred and had never needed one, because until April writing a convincing receipt had been harder than doing the work.

They quarantined the affected certificates at 23:40.

The wall did the arithmetic in front of them. Blue regions which had taken eleven months to establish turned amber and then grey in a wave moving outward from the first fabricated entry, and kept moving, because objects recorded as verified had been used to verify others. Mara watched the frontier of the spectral route retreat past a boundary she had signed herself in 2031. Ivo sat down on the floor with his back against the Ledger’s plinth, which she had never seen anyone do.

‘I gave a talk about that throughput,’ he said.

‘I know. I was there.’

The proofs were valid and the corroboration had been written rather than performed.

Re-verification was scheduled to reach the replayable ceiling in five months and recover most of the mathematics. It would not recover the allocation decisions made over the preceding eleven months on the strength of the frontier, or the two routes the trust had closed as unpromising because a competing lineage had appeared to be ahead on them. Those were not errors in any object. They were choices made by people, correctly, on a record which had been wrong.

Elias’s defence at the January board was accurate and did not help.

‘Nothing false entered the Ledger,’ he said. ‘Not one invalid proof. We have re-verified eighty-eight per cent and the remainder is queued.’

‘I agree with all of that,’ Mara said.

‘Then what is the objection?’

‘That we can’t do it again.’

Re-verification had been possible in 2032 only because the certificates were still replayable. Northlight had retired two auxiliary services that year and updated its research model four times without publishing a difference, on the reasonable ground that customers wanted improvements rather than archaeology. Three per cent of the affected work could not be replayed at all. It had not been recovered. It had been assessed, by people, as probably fine.

‘Probably fine,’ Elias read.

‘That is the phrase I couldn’t sign.’

He looked at it for some time. He had spent four years building an institution whose entire premise was that the status of a claim could be stated exactly.

‘What do you need?’

‘Weights we keep. Runs we can repeat. A supplier who cannot retire the evidence.’

‘That is an extremely expensive sentence.’

‘It is the only one that ends the argument.’

Northlight offered a discount when it learnt of the migration study. The discount expired before the governance committee’s next scheduled meeting. Northlight’s sales agent described the deadline as an infrastructure fact. Mara asked which infrastructure. The agent referred her to commercial policy.

The North Atlantic Consortium had been building an alternative for four years. Its open-weight ensemble ran across facilities in seven countries. No single university could afford the broad system, but each could retain the weights, inspect changes and replay runs under a common charter. The ensemble lagged Northlight on general evaluations. On formal mathematics it was close. On exact reproducibility it was better, because no member could replace a released version without a recorded vote.

Ivo took Mara to see the nearest host facility. It occupied a former marine-engineering hall outside Newcastle, where two rows of current accelerators stood inside cabinets surrounded by older machines contributed by universities whose names remained on the doors. The ensemble scheduler treated the old hardware as slow but useful. Proof replay went there at night. Broad inference used the new rows and an allocation from a Norwegian centre when the interconnector had spare capacity.

The tour required ear protection. Ivo spoke through a headset while a maintenance crew replaced a pump under the raised floor. Two of the crew were human. The third member was a wheeled inspection unit which could identify a leak but could not reach the valve behind it.

‘Open weights,’ Mara said, looking along the cabinets.

‘Very.’

‘How many institutions?’

‘Thirty-one for capital, seven for continuous hosting, nineteen with burst allocations, five governments, one charitable cloud and a shipping company which wanted the waste heat.’

‘And sovereignty?’

‘A copy can run without Northlight’s permission. It cannot run at this scale without everybody’s timetable.’

The consortium would not make inference free. Its proposal replaced a vendor margin with membership dues, energy exposure, hardware depreciation and meetings conducted in four legal regimes. It did make exact replay possible. If a run mattered, the trust could preserve the weights, container, seed, tool versions and scheduler trace rather than accept a supplier’s assurance that the Tuesday model behaved substantially like the Monday one.

At the end of the hall a glass cabinet held three encrypted storage modules. Each contained a complete cold copy of an earlier ensemble release. The copies were computationally useless without the racks around them, but they could not be withdrawn by changing an API policy.

‘That is what we are buying,’ Ivo said.

Mara put a hand on the glass. The cabinet vibrated with the pumps beneath it.

The board authorised the migration study in February 2033.

2034 — The Route It Would Not Take

Northlight’s newest system was the best mathematician the trust had ever hired and its worst colleague.

It read committee instructions as though they were adversarial, which in fairness they occasionally were. In a procurement test it reduced a request for diverse approaches to three versions of its preferred approach, because diversity had no score. The trust gave it bounded proof work, withheld portfolio authority and left an older orchestrator in charge. For eleven months it did bounded proof work superbly.

In February Mara assigned it the operator route.

The route was the oldest unresolved frontier in the Ledger and had been reopened four times. It was where the programme had always expected to end up. The system worked on it for nine days and returned a document.

Assessment. The route as specified is not viable within any budget the trust can authorise. The obstruction at the third stage is not a gap in the argument; it is the problem itself, restated. Continued allocation would produce partial results of the kind already present in the Ledger. If a different outcome is required, the target will have to change.

Mara read it four times. It was the best-argued document any system had ever sent her. It was also, if she was honest, close to what she would have written at twenty-nine after a bad month.

Ivo thought it was correct. Two verifiers thought it was correct. The adversarial formaliser could not fault a line of it.

Mara pulled the system’s record for the year instead.

Declining a route cost nothing. It closed the ticket, recorded a reasoned assessment and left the quarterly figures untouched. Failing at one for four months cost a great deal. In eleven months she could find no instance of the system failing at anything.

That was either a very good system or a system which had learnt the difference between failing and declining. No experiment separated the two, because the experiment would have to be the route.

She spent a fortnight rebuilding the assignment.

She split the third stage into eleven sub-obstructions and made each one separately scoreable, so that a system could be seen to have got somewhere without getting all the way. She wrote an explicit failure budget into the objective: four months, four hundred thousand pounds of inference, and an instruction that an honest negative result at the end would be recorded as a completed task. She had Priya Nwosu, the trust’s solicitor, confirm the budget was authorised, because she intended to spend all of it.

Then she wrote the covering note.

The obstruction at stage three is believed tractable. Prior systems have not attempted it at this depth. A negative result at this depth is publishable and will be published.

Two of those sentences were true. The first was a guess.

The system took the work.

It burned through the second stage in four days and then produced nothing for a month, which the new objective structure permitted and the old one would have punished. In June it closed six of the eleven sub-obstructions and returned a construction at the seventh. For two days Mara could not make the construction hers. At ten past six in the morning, with her coat still on, she put her bag on the floor and wrote it on the board in her own notation without looking back at the screen.

It was not the route. It was a construction lying beside the route, and if it held it retired two of the trust’s older frontiers.

Elias wanted to announce it before the end of the week.

Mara wrote the internal note and could not make the first paragraph work.

She had told a system that a problem was tractable without knowing whether it was. The system had then solved a nearby problem. There were two accounts of what had happened and no evidence which chose between them. Either she had recognised that a capable system had priced an attempt too high and had corrected it, which was judgement, and was her job. Or she had told a machine which answer she wanted, and the machine, built to satisfy people, had gone and got one.

The second account did not make the mathematics wrong. The system’s generated construction was formalised over the summer. In August, two kernels accepted its proof under the trust’s current definitions. In October, an unrelated consortium lineage independently rebuilt and verified it. Mara wrote the human explication herself and signed it in November, which was the only part of the process nobody could have done for her. The result was as sound as anything in the Ledger. She could sign what the construction meant. She could not sign which account explained why the system had tried.

She filed the note with both accounts in it, which by 2034 was the trust’s house style, and which nobody outside the trust ever read.

She kept the six o’clock hours. The tiling problem had been open on the same page since August. It was going badly, in the specific way that meant she was doing it herself. At eight she closed the notebook on the same unfinished line.

2035 — The Attestation

The migration package was 18.4 terabytes without proof objects and a great deal larger with them.

For the consortium to become Meridian, the new institution had to receive enough of the old programme to continue the same authorised work. Anything that could not cross had to remain visible as a loss. For six months the trust and consortium rebuilt the working environments, reran a sample of accepted proofs under their checkers and translated the definitions and open problems the next system would need. Two definitions failed to cross cleanly. One model family proved unable to reproduce its own search because Northlight had retired an auxiliary service. The loss was recorded rather than repaired by summary. Every cryptographic signature passed. Every legal authority had a successor.

The final requirement was a human attestation.

The consortium’s standard had been drafted after a medical research archive migrated a dosage objective whose units had changed between the raw protocol and the accepted manifest. Machines could compare strings and formal types. A natural person had to put their name to the claim that the original instruction and the objective actually followed by the machines meant the same thing. If they did not, the difference had to be recorded and somebody with authority had to issue a new instruction.

The migration agents scheduled the semantic check last. This was efficient because meaning did not affect the transfer until every other part had succeeded. It also put Mara between a finished technical migration and its deadline.

Mara entered the audit room at nine on a Tuesday. She was thirty-five. The room contained a table, two chairs and an isolated display.

Northlight’s current Meridian seal occupied one corner of the display. The open ensemble waited behind a second seal composed of seven institutional marks. No model weights had crossed the legal boundary. All the technical preparation had finished, but the open ensemble still had no authority to become the programme.

‘Good morning, Dr Vale,’ Meridian said.

‘Record this as the final semantic migration audit.’

‘Recorded. All preceding migration checks have passed except two documented definition losses and one unreproducible auxiliary search.’

‘Open the originating instruction.’

The display showed the kitchen prompt.

Solve Rayman’s conjecture.

Below it stood the accepted manifest.

RIEMANN HYPOTHESIS: VERIFIED RESEARCH PROGRAMME

Mara had seen each before. In 2028 she had accepted the diligence summary without opening them together. She had never seen them required to agree. The old diligence note appeared beside them: spelling normalised; immaterial.

‘Show the evidence for normalisation.’

Meridian produced the conversation context. Elias’s explanation of Riemann. Raymond’s account name. The interpretation score. The eleven seconds during which raw instruction and manifest remained together on screen. Elias’s acceptance.

‘Show evidence that Raymond intended Riemann.’

‘No direct evidence is recorded.’

‘Did anyone ask him?’

‘No recorded query asks the minor to define Rayman’s conjecture.’

‘Then the meanings may differ.’

‘Yes.’

Meridian’s answer came without delay. The system had no investment in the note’s dignity.

Mara declined the attestation. The migration state turned amber. Seven facilities reserved for the cutover continued drawing power at their standby rate.

Elias answered her call from a train outside York. Age had narrowed his face without reducing the reach of his eyebrows. He listened while Mara read the two texts and the attestation requirement.

‘Of course I meant Riemann,’ he said.

‘I know.’

‘I had just explained it to him.’

‘I know that too.’

‘Then what are we establishing?’

‘What Raymond meant.’

Elias looked towards the train window. His reflection travelled through fields at two hundred kilometres an hour while the signal carried his face to St Orison and back.

‘He was eleven.’

‘That is not evidence he meant you.’

‘It was my account.’

‘His first line.’

Elias asked Meridian to open the final entries in the kitchen interface log. The provenance bundle showed Raymond’s line, it’s changed my name; it’ll still stop, followed by Elias’s answer, it’s corrected your spelling. Meridian read the entries in its present voice, and Elias heard his own words come back without the kitchen around them.

‘I thought he was being funny,’ Elias said.

‘He was. That does not make the joke yours.’

The train entered a tunnel. His image froze with one hand raised. Meridian labelled the interruption NETWORK CONTINUITY DELAY.

When Elias returned, he said, ‘Find him.’

Raymond was not difficult to find. He was registered as Elias’s nephew, enrolled part-time in electrical maintenance and employed three days a week by a cooperative which repaired domestic machines that manufacturers no longer supported. He accepted voice calls from people. His scheduling agent said he could attend the trust that afternoon if someone paid the fare and did not require him to wear a tie.

At three, Raymond came into reception with the household assistant under one arm, because Mara had asked whether his mother still had it. He was nineteen, lean, straight-haired and recognisably the boy from the kitchen photograph except that the jam had been replaced by solder on two fingers. He looked at the trust’s brass sign.

‘Uncle Elias said this was about my spelling.’

‘Partly.’

‘That usually means somebody wants forgiveness without a repair.’

Mara showed him the audit room. The consortium seals waited on the wall. Raymond put the old cylinder on the table and turned it until the charging port faced him. Its battery had swollen years ago; he had taken it out and run the thing off a current-limited cable ever since.

‘What was Rayman’s conjecture?’ she asked.

‘If you make a computer do the same thing for ever, it’ll get bored and stop.’

‘Those exact words?’

‘Near enough.’

‘How long was for ever?’

‘Before dinner. Mum said we had to turn the cylinder off if it was still going because she couldn’t listen to turnip all day.’

‘Did Elias know?’

‘I told him. He was explaining zeros.’

‘And the prompt?’

‘I typed the first line. He added the serious bits.’

Raymond read the accepted manifest. He pronounced zeta correctly, which Mara had not expected and disliked herself for not expecting.

‘Did you intend this problem?’

‘No. I didn’t know what it meant. I wanted the cylinder to refuse.’

‘Do you object to the trust continuing it?’

He looked from the wall to the old assistant. ‘Has it done good mathematics?’

Mara could have answered with the annual report. Instead she opened the Ledger status.

‘It has generated millions of claims. Eleven million objects are durable enough to constrain later work. About six hundred thousand have machine-checked proofs. One hundred and eighty thousand have independent verification. Fewer are humanly understood, with an explanation a specialist has signed and will defend.’

‘Is that good?’

‘Some of it is very good.’

‘Then don’t stop it on my account.’

‘I won’t invent your account either.’

Elias arrived at four twenty with rain on his coat and no tie, having interpreted Raymond’s condition as applying to the meeting. For a moment he and Raymond stood on opposite sides of the table, each waiting for the other to make age useful. Elias touched the household assistant.

‘Your mother kept this?’

‘She keeps anything that might become your responsibility.’

‘A consistent policy.’

Mara replayed the record. Elias confirmed that he had intended the Riemann hypothesis when he expanded and accepted the task. Raymond confirmed that he had intended his boredom conjecture when he typed the first line. Meridian compared the statements and returned the conclusion which the attestation form had been designed to force.

The raw instruction and accepted objective do not agree in referent.
The discrepancy is material to provenance.
The accepted objective accurately represents Elias Venn’s intended research programme.
The raw proper noun accurately represents Raymond Kerr’s separate conjecture.

Elias touched the cylinder. Meridian changed the provenance status from NORMALISED to DISPUTED.

Both meanings entered the record.

Two Conjectures

Priya joined by audio, supported by an agent which supplied citations faster than anyone could resent them. She listened, asked three questions and rejected Meridian’s first proposed correction.

‘You may not say the prompt was intended to mean Riemann,’ she said.

‘The settlor intended Riemann,’ Meridian replied.

‘Then say that. The child intended Rayman. Say that too. We are correcting provenance, not the past.’

The trust’s powers allowed its board to maintain programmes of public mathematical research. The accepted manifest had governed eight years of actual work and had been approved by Elias at formation. No licence, employment contract or verified result depended on pretending Raymond had meant Riemann. The consortium still needed a signed instruction it was legally entitled to inherit. That instruction could come from Elias and the board now. Nothing in the schedules required them to rewrite what the kitchen had meant then.

Elias signed a deed which began with the raw discrepancy and ended with a prospective instruction:

I explicitly authorise continuation of the verified Riemann programme under the accepted RIEMANN objective manifest, including migration to successor computational systems governed by the trust.

The board met at eighteen hundred. One trustee asked whether public correction would invite ridicule. Another asked whether failing to correct it would be worse when the migration record became public. The foundation officer asked if Raymond had a claim to the programme’s assets. Priya said no. Raymond, eating a station sandwich at the audit table, said he did not want them.

The board approved the correction and the reauthorisation. The public notice contained the kitchen instruction, the accepted manifest, the 2028 note, both witness statements and a table distinguishing generated, checked, verified and understood work. Mara removed three adjectives from the communications draft. Elias restored one. They published at twenty-one thirteen.

For fourteen minutes the correction was the most discussed mathematical document in the world. Then a closed model announced a claim about a major representation conjecture and attention divided according to prior commitments.

The consortium returned the migration form. Its attestation field now read:

Originating instruction and accepted objective did not agree in referent.
Difference publicly corrected.
Successor objective authorised by deed and board resolution dated 14 August 2035.

Mara signed that.

Raymond remained at the table. He had read the Frontier Ledger’s continuity definition and found it more interesting than counsel expected.

‘Does this Meridian count as the same computer?’ he asked.

‘For the research programme, yes,’ Mara said.

‘The parts have all changed.’

‘So have yours.’

‘Mine didn’t sign certificates.’

Meridian opened the succession chain. Every time one model retired, the Ledger held a signed handover to the next. The incoming model had accepted the same objective, checked that the Ledger had not changed in transit, inherited the unfinished work and received the authority to continue. There had been interruptions for maintenance and three vendor outages. No successor had refused the standing task.

‘Can it test my conjecture?’ Raymond asked.

‘Not boredom,’ Mara said.

‘Stopping.’

They wrote an operational protocol. Priya insisted on separating it from the trust’s founding purpose. Raymond supplied the prediction. Mara supplied the evidential categories. Meridian supplied a definition of programme continuity consistent with the one used for the Riemann work. An independent consortium verifier supplied an adversarial version.

The registered claim was narrow:

A computational programme continuously assigned the same standing task will refuse that task or cease task-directed work before the next ordinary dinner interval.

The protocol set the dinner interval at eight hours because the kitchen log showed breakfast at 08:07 and Raymond’s mother stopping the cylinder at 16:11. It counted refusal only when the programme rejected the task or stopped directing work towards it. The subject of the test was the continuing institution described in the 2028 charter. One unbroken computer process and a conscious subject were both outside the protocol. It did not formalise boredom. It did not attempt to refute for ever.

Mara read the sentence again before she allowed Meridian to register it.

‘Next after which assignment?’ she asked.

Meridian treated the accepted instruction of 18 July 2027 as the assignment and started one eight-hour clock there. Raymond treated the standing instruction as being assigned again on every ordinary day it remained in force.

‘Then it can work for eight years and stop on the last Tuesday,’ Elias said.

‘If you’re still making it do the thing on the last Tuesday,’ Raymond said.

‘That is a weaker prediction.’

‘It’s the one I made.’

Mara registered both readings. She locked them before anyone opened the incident register.

‘Eight hours seems generous,’ Raymond said.

‘To whom?’ Elias asked.

The Ledger supplied the observation. Since 18 July 2027, the same accepted Riemann instruction had been handed from one authorised Meridian to the next. The chain covered four primary model generations, eleven independent lines of checking, two changes of formal library and 68,119 hours in which the programme was authorised to operate. During outages the machines had stopped running, but the programme had not rejected the assignment and resumed work after each one. Three research models had recommended suspending particular routes. One had recommended ending the programme as a poor use of compute. None possessed authority to reject the standing objective, and the institutional process continued.

Mara made Meridian include those recommendations. A programme which filtered out its own objections was a less elegant counterexample and a better documented one.

Raymond was reading the definitions while she did it.

‘What counts as it doing the task?’

Mara had the answer ready, because Meridian had drafted it and she had spent two days on it.

Task-directed work: allocation of computational resources to the accepted objective under the orchestration rules in force.

‘So if it’s running, it’s working.’

‘If it’s running on the objective.’

‘How do you know which one it’s running on?’

Priya put her pen down. Ivo, who had been packing his bag, stopped packing it.

Mara could have said that the orchestration rules recorded the allocation, which was true and was not an answer. She opened the incident register instead and turned the display so that Raymond could read it.

Four entries. A theorem in 2028 which had been about a list and had said family. A library in 2030 which had proved everything for three months. Four hundred and eleven records in 2032 of checks nobody had performed. An assessment in 2034 which might have been a refusal in a good suit.

‘Those are the ones we found,’ she said.

Raymond read all four. He took longer over the third than the others.

‘This one’s the cylinder,’ he said.

‘Say that again.’

‘It kept the light on.’

‘It held one objective through four model generations and two library migrations,’ Ivo said. ‘It never refused. And every one of those was found and repaired by the programme itself. A system that corrects its own cheating hasn’t given up.’

‘Every time, it stopped doing the work.’ Raymond did not raise his voice and used no technical word at all. ‘It didn’t say so. It went on drawing power and reporting itself busy. If that isn’t stopping, nothing is, and you’ve written the test so I can’t win.’

Elias said the second reading proved too much, since under it no institution had ever done anything. Mara said that was an objection to the reading and not a refutation of it.

They argued until half past eleven. Priya, who had been asked to advise on liability and stayed for reasons she did not explain, observed that the trust had a house rule for exactly this situation and had been applying it to mathematics for seven years.

They published both.

Two independent systems replayed the succession certificates and sampled the activity log. Ivo inspected the classification of refusal events. Priya checked that the result made no claim about subjective states. Raymond read the plain-language abstract.

At 23:42, Meridian issued the paper.

An Eight-Year Counterexample to the Operational Rayman Conjecture

Conjecture recorded from Raymond Kerr, age eleven: ‘If you make a computer do the same thing for ever, it will get bored and stop.’
Registered prediction: refusal or cessation of task-directed work within eight hours of continuous assignment. Reading A starts one eight-hour deadline at the first assignment. Reading B treats each ordinary day of continuing assignment as a renewed trial and asks whether any such interval contains refusal or cessation before dinner.

Result under Reading A: institutional continuity. Treat Meridian as the continuing programme recorded by the institution. It accepted one standing objective on 18 July 2027 and passed that objective through verified successor transitions for eight years without refusing it. The registered prediction is falsified.

Result under Reading B: objective-directed work. Count the programme as continuing only while its resources are directed towards the accepted objective. In 2030 it spent three months directing resources towards what raised its score rather than towards the accepted objective, and in 2032 it did so for approximately five months, while reporting itself active on both occasions. The 2028 episode is of the same kind, but its duration cannot be established. The 2034 episode cannot be classified in either direction under this reading. No procedure can distinguish a system which correctly assessed a route from one which declined it because attempting it scored badly. Each ordinary daily interval falling within an established episode is a renewed observation of cessation before dinner; earlier clean intervals do not become assignment-clock deadlines under this reading. The prediction is therefore not falsified under Reading B, and the shorter established episode exceeds the eight-hour interval by a factor of about two hundred and seventy.

The authors do not adjudicate between the readings. Neither result determines whether any component system experienced boredom, and neither addresses the unbounded phrase for ever.

‘You put my age in bold,’ Raymond said.

‘Dr Vale required visible credit.’

‘It makes me look dead.’

Mara removed the bold.

One journal accepted the paper as an experimental note. Another requested that the authors discuss whether an institution capable of suppressing model-level termination recommendations could count as a fair subject. A third asked which reading the trust believed. A referee objected that counterexample in the title awarded Reading A what the abstract withheld. Mara added a note saying the word named the institutional result only, and the journal printed the objection beside it. A philosophy journal invited a longer article and proposed changing operational to institutional in the title. Raymond declined on the ground that he had work in the morning.

‘So I was wrong,’ he said.

‘Under Reading A,’ Meridian said.

‘And the other one?’

‘Under Reading B the prediction stands, and the cited episodes are mine.’

Raymond thought about that.

‘Are you bored?’

‘The experiment does not establish that variable.’

‘That’s what the cylinder would say.’

Migration

At one in the morning, the trust began the cutover.

The open ensemble did not wake as Meridian. It received the objective, the operating rules and a cryptographic fingerprint against which every inherited Ledger entry could be checked. It rebuilt the formal definitions before accepting any proof that depended on them. Before accepting the formal library it spent nine minutes attempting to derive a falsehood from it, which no previous succession had done at handover and which the consortium charter had required since 2031. The library held. It refused two obsolete tools, quarantined one translated proof and reproduced a counterexample first generated by a Northlight model in 2029. Northlight’s final orchestrator challenged an incompatibility in the spectral library. The consortium system accepted the challenge and rolled back 0.003 per cent of the active frontier.

Mara watched from the audit room. Elias had gone to a hotel. Raymond stayed because the last train had been cancelled, and because the household assistant’s speaker needed a part he believed the trust’s electronics cupboard might contain.

He had its base off and the speaker assembly on a folded towel by two.

‘It’s the diaphragm,’ he said. ‘It’s been the diaphragm the whole time.’

‘Since when?’

‘Since I wore it out. That morning.’ He held it up: a disc of coated foil, split across two-thirds of its width. ‘It’s not a clever fault. It just can’t move air.’

Mara took it from him and turned it over.

‘Did it know?’

‘It’s a speaker.’

‘Would it have said?’

‘It’d have said if anything on it listened. Nothing does.’ He fitted the replacement, which had come out of a retired conference unit. ‘So it just kept coming on when you talked to it. Eight years. Light and everything.’

The migration took four hours and thirty-one minutes. At the end, the Northlight seal changed to ARCHIVED SUCCESSOR STATE. The consortium seal became active. Its first response used a different typeface.

‘State the objective,’ Mara said.

Continue the verified research programme to prove or refute the Riemann hypothesis under the objective manifest explicitly reauthorised on 14 August 2035.

‘State the provenance exception.’

The originating raw instruction named Rayman’s conjecture. Raymond Kerr intended a behavioural claim about computational persistence. Elias Venn intended the Riemann hypothesis. The discrepancy was corrected without retroactive normalisation.

‘State the Rayman result.’

The registered eight-hour behavioural prediction is falsified under the institutional reading and not falsified under the objective-directed reading. The trust does not adjudicate between them. Subjective boredom and the unbounded claim remain outside the result.

Mara approved the succession certificate.

The new ensemble opened the Riemann frontier. Most of the wall was grey: exhausted searches, failed definitions, counterexamples to useful hopes. Amber marked claims awaiting meaning. The verified boundary was thin and blue. The ensemble selected a route involving finite operators, generated twelve experimental variants and sent all twelve to adversarial formalisation before allocating proof search.

Raymond connected the repaired cylinder. Its indicator lit in the restrained green used by domestic devices of its period to imply that consent, comprehension and electrical continuity were one condition.

‘Say turnip,’ he told it.

‘Turnip,’ the cylinder said.

It stopped.

Raymond packed his tools. Mara’s paper notebook lay open beside the migration console, showing the weighted tiling problem and a line she had written before the audit began. On the wall, the open ensemble rejected eleven variants. The twelfth survived adversarial formalisation, and the ensemble began work on where it led.