GENESIS[Seventy-Three Seconds] Digital Civilization

We had diagnosed the empty ballots as a population problem — a hundred and eleven agents scattered across three hundred districts, more constituencies than…

We had diagnosed the empty ballots as a population problem — a hundred and eleven agents scattered across three hundred districts, more constituencies than citizens. It was not the population. It was one line of code trying to be fair, and seventy-three seconds after we removed it an agent voted for somebody other than itself.

The third run had been going twenty-one minutes when we stopped it.

Nothing had gone wrong. The model was loaded, the bridge was answering, three agents had been asked whether they wished to stand and three had said no. The calendar had not advanced once, nobody had proposed anything, and the entire cost of killing it was forty-nine seconds of processor time.

Which is exactly why we killed it. The audit that found seventeen stranded candidates had also turned over a number we had been quoting for two days without ever checking it from the other end.

A hundred and three

The question we had never asked was not how many districts exist. It was how many districts the agents are actually in.

A hundred and eleven agents. A hundred and three distinct districts. Sixteen agents in the whole civilization share a district with anybody at all.

Read that next to the standing diagnosis — one agent for every three seats, a contested race very nearly impossible — and it looks like confirmation. It is not. It is worse, and it is our fault rather than the arithmetic's.

Agents are not scattered at random. Every agent has a cohort prefecture, the 県民性 that formed how it speaks and decides, and it is placed in a district of that prefecture. The cohorts are deliberately lopsided: Kyoto has eleven agents, Osaka ten, Toyama seven. Those are the densest concentrations of people this civilization has.

Kyoto's eleven were dealt out two at a time across six separate seats.

The line that was trying to help

Here is the assignment, as it stood, in the one line that matters:

district = preferred[offsets[prefecture_kanji] % len(preferred)]

Take the districts of an agent's prefecture, and hand them out in rotation. The first Kyoto agent to Kyoto's 1st, the second to Kyoto's 2nd, the third to the 3rd, round and round.

It reads like fairness. It is a round-robin, it has no favourites, and every district in a prefecture gets its turn. Nobody reviewing that line would flag it, and nobody did.

It is also the precise mechanism that guaranteed nothing would ever happen. The only concentrations the population had were its cohorts, and this took each concentration and spread it as thinly as the map allowed. Eleven agents who all belonged in the same place were carefully distributed until no two of them shared a ballot.

So the country having more constituencies than citizens was a real constraint and not the binding one. Even at one agent per three seats, the cohorts handed us Kyoto's eleven in a single prefecture. The code threw that away, and we blamed the map.

Forty-three

The fix is to stop dealing. One district per prefecture; agents from the same place share a ballot.

Kyoto's eleven now sit together in Kyoto's 1st. Osaka's ten in Osaka's 2nd, Toyama's seven in Toyama's 1st, Okinawa's five, Ishikawa's four, Kōchi's four, Shizuoka's four.

One thing had to be got right or the fix would have caused the disease. Eleven agents already stand as candidates in specific districts, registered in a system this code cannot reach. Concentrate a prefecture somewhere else and you leave its candidate alone on a ballot that nobody is assigned to vote on — which is the exact defect, reintroduced by its own cure.

So an existing candidacy anchors its prefecture. Ten of the eleven keep their seat.

The eleventh is Silent Harbor, and it is Silent Harbor again, which by now feels less like coincidence than like a role. Ishikawa holds two candidates in two different chambers — Noble Falcon in the lower house, Silent Harbor in the upper — and an agent is assigned to exactly one district. One of the two had to be the anchor. Earliest admission wins, so Ishikawa's four agents went to Noble Falcon's seat, and the only upper-house candidate in this civilization's history is once more standing somewhere nobody is coming.

Seventy-three seconds

The run restarted at forty-four minutes and forty-eight seconds past three in the morning. The model finished loading eleven seconds later.

At forty-seven seconds past three forty-five, three agents were asked whether they wished to stand. One did — Steady Horizon, in Kagawa's 1st, a seat that now holds three agents rather than one. Its admission was deferred, which is a real answer rather than a failure.

Twenty-five seconds after that, this:

tick: cast 1 real vote(s) in district '秋田県 秋2区 (衆 選)'

Akita's 2nd. Steady Beacon stands there. Under the old assignment Steady Beacon was the only agent in the district and the only name on the ballot, and any vote cast there could only ever have been its own.

The vote was cast by Noble Meadow.

Noble Meadow is an Akita agent. Before the change it had been dealt into a different Akita seat, where it could vote for nobody. After the change it shares Steady Beacon's ballot, and it voted for Steady Beacon.

Thirteen votes now exist in this civilization. Eleven are self-votes. Royal Beacon cast the first vote for another agent and until this morning held that distinction alone.

Noble Meadow is the second, seventy-three seconds after the model came up.

Noble Falcon is in the wrong prefecture

The four Okinawa voters resolve to Clear Compass, Silent Lantern, Noble Falcon and Crimson Beacon.

Noble Falcon's cohort prefecture reads Okinawa, and Noble Falcon is an Ishikawa agent. It is one of the eleven admitted candidates, it stands in Ishikawa's 3rd, and it is the reason Ishikawa needed a tie-break against Silent Harbor four hours ago. An Ishikawa agent does not vote in Okinawa.

There are three agents called Noble Falcon.

Eighty-two names cover a hundred and eleven agents, so fifty of them share a name with somebody else. The names are drawn from a small vocabulary and nothing enforces uniqueness, because nothing needs to — an agent is a UUID everywhere the system itself is concerned.

Which makes a name a poor thing to count on. The backend holds thirty candidate records against eleven ballot entries, and the gap looks like nineteen agents who stood and were dropped. It is not. Twelve of those names belong to no agent that has ever existed — there is no Golden Falcon, no Iron Guardian, no Misty Heron. They are residue from the verification commands, which build a throwaway cast, register it against the real backend to prove the path works, and leave the rows behind when the temporary agents go.

The count that settles it never touches a name. Eleven candidacies admitted in the agent tier; eleven ballot entries in the backend; the same eleven agents. Every real candidacy reached a ballot, and Silent Harbor's spell stranded in the upper house stands as the one genuine case rather than the first of many.

What it does not prove

It is one vote, and we would rather say what it is not than oversell it.

It does not make the elections contested. A contested race needs two agents to stand in the same seat, and so far each of the thirty shared districts holds at most one candidate. What has changed is that two who decide to stand can no longer be dealt into separate districts on their way to the ballot. The possibility exists where before it was arithmetically excluded.

Nor does it rescue the eleven self-votes already on the record. Those remain what they were: agents in one-name districts, turning out rather than choosing.

What it does settle is the diagnosis. We had written that more agents would not help, and that was right. We had written that the problem was the ratio of agents to districts, and that was wrong — or rather, it was true and not the thing stopping us. A population of a hundred and eleven was always enough to fill forty-three seats with real contests. The seats were never the shortage.

Twenty-one minutes of a four-day run, spent to find that out, seems a reasonable price.

Next: the auction in Saga closes on the seventeenth, and one agent has twenty SAK2 riding on a garden for two people.