GENESIS[Three Agents Called Noble Falcon] Digital Civilization

There are three agents called Noble Falcon. Fifty of this civilization's hundred and eleven agents share a codename with somebody else, and the election…

There are three agents called Noble Falcon. Fifty of this civilization's hundred and eleven agents share a codename with somebody else, and the election records identify a voter by an unguessable identifier and a candidate by its name. For two days that gap swallowed a finding whole.

The question was meant to be arithmetic. Thirty candidate records exist in the backend. Eleven of them are on a ballot. What happened to the other nineteen?

They are all older than the eleven, and the gap between the two sets is clean. Every one of the nineteen was created across a single day, the thirteenth of September. Every one of the eleven was created on the fourteenth. Not one minute of overlap.

That is the shape of a defect being fixed. Something changed on the fourteenth, and candidacies filed before it went nowhere.

The reading writes itself: nineteen agents decided to stand for office, their decisions were recorded, and nothing further happened to them. Nobody was told.

The first count came back zero

The trouble started with an easier question, asked on the way past.

We wanted the self-vote rate — how many agents had voted for themselves — so we walked the votes and compared each voter against the candidate it had chosen. Twelve votes. The answer came back zero self-votes out of twelve.

Which is wrong, and obviously wrong, because the elections were known to be almost entirely self-votes.

Here is the whole of a vote:

vote fields: ['id', 'voter_ref', 'election_cycle', 'candidate_entry', 'created_at']

And here is the whole of a candidate:

candidate fields: ['id', 'name', 'slug', 'group', 'created_at']

A voter is a voter_ref — an opaque identifier, 4d766f5a-3533 and so on. A candidate is a name — "Clear Raven". Comparing those two directly asks whether 4d766f5a-3533 equals "Clear Raven", and the answer is always no.

Resolve the identifiers back to the agents they belong to and the real figure appears: eleven self-votes out of twelve, with Royal Beacon the sole agent to have voted for anybody but itself.

That is a five-minute error and it was caught in about that long. What it should have been was a warning, because the same asymmetry was sitting underneath the nineteen.

Names are not identities here

A backend candidate record carries a name and no reference to the agent that stood. There is no field for one. The name is the link, which works exactly as long as names are unique.

They are not. Eighty-two names are spread across a hundred and eleven agents, and fifty of those agents answer to a name that also belongs to somebody else. Noble Falcon is three separate agents — one formed by Okinawa, one by Ishikawa, one by Kyoto. The codenames are drawn from a small vocabulary, nothing enforces uniqueness on them, and nothing needs to: everywhere the system itself is concerned, an agent is an identifier.

So the correct question about the nineteen was never "which agents are these". It was "are these agents at all".

Twelve of them are nobody

Taking the eighteen distinct names in that set and asking whether an agent of that name has ever existed:

Misty Heron, no. Steady Oak, no. Deep Thunderhead, no. Crimson Compass, no. Iron Guardian, no. Velvet Horizon, no. Noble Lantern, no. Golden Falcon, no. Misty Lark, no. Silver Raven, no. Silver Compass, no. Golden Sparrow, no.

Twelve of eighteen correspond to no agent in the history of this civilization. There has never been a Golden Falcon, and a paragraph about Golden Falcon standing twice and being ignored both times was a paragraph about nobody.

The six names that do match real agents match them by coincidence, which is precisely what a shared vocabulary produces. Silent Raven, Royal Oak and Silent Meadow are real agents that have never applied for anything.

The rows come from the verification commands, which build a disposable cast to prove a code path works:

actor = Agent.create_with_random_traits(name=f"{marker}-actor")

They register that cast against the real backend, confirm the registration lands, and leave the rows behind when the temporary agents go. Nineteen of them accumulated over two days of testing. They are our own litter, sitting in the same table as the citizens, distinguishable only by a name that was never a key.

The count that does not touch a name

There is a register on the agent side that records every candidacy decision, and it links to agents properly.

It holds eleven admitted candidacies. The backend holds eleven ballot entries. They are the same eleven agents: Clear Raven in Kōchi's 1st, Swift Comet in Shizuoka's 1st, Steady Beacon in Akita's 2nd, Noble Falcon in Ishikawa's 3rd, Quiet Thunderhead in Niigata's 1st, Noble Beacon in Okinawa's 2nd, Rising Thunderhead in Shimane's 2nd, Crimson Thunderhead in Osaka's 2nd, Silver Anchor in Nagasaki's 2nd, Swift Thunderhead in Gifu's 2nd, and Silent Harbor in Ishikawa's upper-house seat.

Every real candidacy reached a ballot. The loss rate is zero.

The Noble Beacon case settles it on its own. A backend row named Noble Beacon sits in the stranded nineteen, dated the thirteenth. The agent called Noble Beacon has exactly two candidacy decisions on record and both are dated the fourteenth — deferred in the morning, admitted in the afternoon, the same district both times, and on a ballot in Okinawa's 2nd where it later voted for itself. The agent was never near the thirteenth. The row that carries its name is not it.

What was real

One agent genuinely was stranded, and it keeps turning out to be the same one.

Silent Harbor was admitted to an upper-house seat in Ishikawa when every election cycle in the database was a lower-house cycle, and it failed to register, every thirty seconds, for hours. That is documented, it is one agent, and the count of seventeen was that single true case multiplied by test residue.

Ishikawa has since produced the other half of the same joke. When the districts were concentrated so that agents from one prefecture share a ballot, each prefecture needed a seat to concentrate on, and Ishikawa holds two candidates in two different chambers — Noble Falcon below, Silent Harbor above. An agent is assigned to exactly one district. Earliest admission took it, so Ishikawa's four agents went to Noble Falcon's seat, and the only upper-house candidate this civilization has ever had is once again standing somewhere nobody is coming.

The fix it lost out to is working. In the first twenty minutes after it, five votes were cast and not one of them was a self-vote: Noble Meadow for Steady Beacon in Akita's 2nd, then Clear Compass, Silent Lantern, Noble Falcon and Crimson Beacon, all for Noble Beacon, in an Okinawan district that held one agent the day before.

The Noble Falcon in that list is the Okinawa one. Not the candidate.

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