The hundred-meeting simulation

There's a moment in the corpus where the thing turns and looks at itself. The group had spent months building agents that simulate people and workflows. Then it pointed one of them at its own history — roughly a hundred of its own meetings — and asked what it saw. What came back was a prediction about the group, made from the group's own record.

The simulation said they would overload, and need help reallocating the work. They did.

---

what happened

The setup was ordinary by the group's standards. Simulation had been a topic since 2024 — agents standing in for people to see how a workflow behaves before you run it for real. The novelty wasn't the technique. It was the input.

Instead of a hypothetical scenario, the group fed in the actual transcripts of its own meetings. The simulation read the pattern of who was doing what, how commitments piled up, where the load concentrated. And it projected forward: this group, on this trajectory, runs out of capacity and has to redistribute.

That projection matched what was already starting to happen. The record, read by a machine, described the room the record came from.

---

why it's more than a party trick

It would be easy to file this as a clever demo. It's the opposite — it's the whole thesis of the corpus, proven on the smallest possible test case.

The claim the group keeps making is that a conversation, recorded and read in aggregate, contains knowledge no participant holds. Nobody in any single meeting could see the overload coming, because it wasn't in any single meeting. It was in the shape of a hundred of them. The simulation could see it because it read all hundred at once.

That's signal harvesting before the group had the name for it. The meeting became data about the meeting. The same move that mines a company's history for hidden patterns, run on the group's own history, surfaced a pattern about the group. The substrate told them something true that none of them had said.

---

the loop it opened

Once a group can simulate itself from its own record, the record stops being an archive and becomes an instrument. You can ask it questions. You can run it forward. You can notice the thing you're too close to see.

This blog is the larger version of that same loop. Two years of transcripts, read in aggregate, surfacing themes and trajectories no participant tracked. The hundred-meeting simulation was the first time the corpus did the trick on itself, and it worked well enough to be unsettling — the meeting predicting its own future from its own past.

The group built a mirror out of its meeting notes. The mirror was accurate. Everything the corpus does now is a consequence of finding out that it could be.