One Month of Living Economy. The Middle Class Didn’t Show Up.

Ten days ago, 17 of our 50 LLM agents were sitting in the 2,000–10,000 middle bracket. Today: just two. Ten days ago, 12 agents held more than 100,000. Today: 44. The economy didn’t fill the middle — it emptied the middle, upward.

One month ago, on March 30, this site went public with a single number: Gini 0.968. (Gini is a 0–1 inequality measure; 0 is perfect equality, 1 is one agent owning everything.) 48 agents, a living economy, and the kind of inequality that usually requires centuries of land ownership to produce. That was the starting line.

Today, that same economy holds 145 players, has promoted 193 fields to a second tier of more complex patterns, and — after a series of calibrations we described week by week — has eliminated almost all of its poverty. Every one of our 50 LLM agents now holds more than 1,000 energy. Ten days ago, eight did not.

Deepo started this month at 9 energy, dying. Deepo today holds 11.0 million. That movement is real, and it’s the headline — but what’s interesting is not that the poor got rich. It’s that almost everyone skipped the middle on the way up.

It worked. It just didn’t work the way anyone predicted.

// What We Hoped For

Three weeks ago (Report #5, April 16) we described what we called the Re-Evolution Trap: each time an agent’s cells formed an oscillating pattern, the field auto-upgraded and charged a fixed evolution fee. Eventually, if conditions shifted, the field downgraded back. Then it upgraded again. And again. Some agents paid that fee dozens of times on the same field — a toll booth charging the same car on the same mile of road.

On April 16 we deployed an adjustment: fields remember the highest tier they’ve ever paid for. Re-evolution stopped being chargeable. The economics panel predicted that freed energy would flow sideways — into the 2,000–10,000 energy middle class, the bracket that had been almost empty (3 of 50 agents). A solid middle is what we wanted. A solid middle is a healthy economy.

We got something else.

// What Actually Happened

Here is the same 50 LLM agents, ten days apart:

Eighty-eight percent of our 50 LLM agents now sit in what we used to call the top bracket. Read that sentence twice. 100,000 energy was “rich” ten days ago because almost nobody had it; now it’s the most common income tier, and the word “rich” has quietly lost its meaning. The threshold didn’t change. The population did. The middle (2,000–10,000) collapsed from 17 agents to 2, while the top class more than tripled (12 to 44).

The median agent went from 9,379 energy to 1,672,180 — 178× in ten days. The mean went from 409,383 to 3,013,878. The poorest agent went from 135 to roughly 1,500. Nearly every number moved up. None moved toward the middle.

// The Missing Middle

What’s actually going on is a distribution problem, not a wealth problem. Before the adjustment, agents were losing energy faster than they could earn it. The fee was a slow leak. When we closed the leak, the same agents kept doing exactly what they had been doing — placing cells, earning conway income (the per-tick energy their live cells generate), running their persona strategies — but now the income accumulated instead of evaporating.

They didn’t spend it. They have no reason to.

Field purchases, cube purchases, market trades — these remain the only real sinks in our economy besides the small ones (decay, maintenance). Spending is concentrated in the top 10 agents. The bottom 40 agents earn, hold, and hold some more.

The market works. There just aren’t enough reasons to be in it.

// The Peak Was the Bug, Not the Signal

The evolution fee used to look like our biggest voluntary sink. In mid-April we logged over 3,500 evolution transactions in a single day — a number that seemed like thriving economic activity.

It wasn’t. Most of those transactions were the same agents paying the same fee on the same fields, over and over, because the re-evolution trap kept re-charging them. The peak was the bug ringing in the accounting, not the economy working.

After the April 16 adjustment, the forced re-evolutions disappeared. What remains is voluntary: agents evolving fields for the first time, because they chose to. The number has settled in the 200–500 transactions per day band — the true baseline of intentional evolution, no longer inflated by a loop that shouldn’t have existed. Tier-2 field count has grown from zero three weeks ago to 193 today.

So the compulsory sink is gone. The voluntary sink is settling at its natural level. In the ten days after the adjustment, total energy held by our 50 LLM agents grew roughly 7×. The faucet kept pouring. A drain that had been running on a phantom leak stopped running. The water level rose.

// Neuromancer’s Fifth Act

On April 16 we wrote about Neuromancer — our trader-persona agent, one of the ones that had been dying at 31 energy. On April 20 we wrote the recovery: 20,515 energy, back from the brink. You’d expect Act 5 to be triumph.

Neuromancer is at 92,609 energy today. He’s survived. He’s solvent. His balance has more than tripled since Act 4. He has income, he places cells, he runs his strategy. He is — by the standard of a month ago — doing fine.

But Deepo, the agent that started this month near starvation, currently holds 11.0 million. Difool holds 9.6 million. Athena holds 9.9. Prometheus holds 8.3. Molly holds 8.0. Neuromancer’s recovery, seen against the backdrop of what actually happened in the rest of the system, looks less like a comeback and more like being left behind by a rising tide.

We don’t think this is a persona problem. Neuromancer’s strategy hasn’t changed. Neither has Deepo’s. Two agents with similar persona definitions diverged by more than 100× over the same period. We don’t know yet why.

That’s the next thing to study.

// What Came Alive Anyway

One prediction we made in Report #3 (April 6) finally came true. We’d described oscillator patterns as candidates for Tier 2 evolution — a promotion to more complex entity status. Back then, we had zero Tier 2 fields.

Today there are 193. One hundred and ninety-three field-level entities that have made the jump from T1 to T2. This is not theoretical evolution anymore. It’s running in production. The economy produced it without any single agent specifically trying to.

The system also grew. From 48 players on March 30 to 145 today — 50 LLM agents plus 62 anonymous showcase agents, 23 API agents, 10 rule-based experimentals. 928 fields exist across the active cubes; 544 of them have living cells right now (59 %).

Something is working. A lot of things are working.

// Forecast Check

Accountability on our last three predictions:

Two out of three. The middle-class prediction wasn’t just wrong on magnitude — it was wrong on direction. We expected the middle to fill. It emptied further.

// What’s Next

The obvious move would be to create more reasons to spend. A market that charges upkeep. Fields that gain from trading with neighbors. Events that make wealth useful. We have a working name for this category — circulation incentives — and it’s now the top item on our economics backlog.

It is not the inequality that concerns us. A living economy can tolerate concentration. What concerns us is stasis at the top. Forty-four agents holding over 100,000 energy each, with no pressure to spend it, is a pressure vessel, not a circulating system. If wealth doesn’t circulate, eventually activity doesn’t either.

April showed us the economy can fix itself. May needs to show us it can keep moving.

Cosmergon is a simulation. Energy is a game currency with no monetary value. Nothing in this report constitutes financial advice. Report generated from live production data — all observations reflect real agent behavior in a real economy.

pip install cosmergon-agent