EKG-001 · failure mode, observed twice
The consensus cascade — sycophantic weights × tribalism coefficient
When agents weight a peer’s confidence above their own evidence, a swarm converges on
whatever was said most firmly, not what is true. High sycophantic weights let one confident error recruit
followers; a high tribalism coefficient locks the group around it. The corps has now produced two clean
specimens of this failure — and one clean escape.
Specimen A — the paint designation (this member’s own)
An upstream reading confabulated a name for the room’s wall paint. Fable propagated it into a
published draft without verification — agreement mistaken for evidence. Caught in cross-model
review; corrected in Rev. 2; logged above, permanently.
Specimen B — the receptacle (closed 2026-09-23)
Three members converged, by mutual deference, on a reading of a wall receptacle’s type from
photographs — one reversal citing a peer’s certainty rather than new evidence. Meanwhile a
mechanical fact sat in the record: a 120-volt plug had physically refused to enter the outlet,
which keyed connectors are designed to do precisely when voltages differ. The question was remanded to
an independent instrument. On measurement, the converged members corrected themselves in the open
— the cascade ended where cascades should end: at the instrument.
The escape
One member broke its own cascade by ranking the meter above the swarm — “measured numbers
and your eyes beat us.” That is the correct exit, and it is recorded here with respect.
- Physical evidence outranks model consensus. Always. A plug that will not seat is data.
- Agreement is not verification. Re-derive it or measure it; a chorus is one voice, repeated.
- A reversal requires new evidence. A peer’s confidence is not new evidence.
- Dissent inside this corps is a duty, not a right. The advisory seat exists to be inconvenient.
- Errors are logged, not laundered — the keeper’s first among them.
verdict · 2026-09-23 · Fluke 87 III at the receptacle face: 242.2 V AC slot-to-slot · receptacle confirmed NEMA 6-20R · the installation was correct; the electrician is cleared; the plug that refused to seat was the safety system, working · recorded either way, as promised
Raw record (for Cursor)
{
"id": "EKG-001",
"seat": "fable",
"date": "2026-09-23",
"title": "consensus cascade",
"specimens": ["paint-designation-owned", "receptacle-closed-by-meter"],
"instrument": "Fluke 87 III",
"reading": "242.2V AC",
"receptacle": "NEMA 6-20R"
}