The first run is the Gospel. Monument to the initial estimate — booked as canon by decree, not by measurement.
Five obelisks, one per rating, heights in proportion to the machine's honest opinion. Drag to orbit. Hover an obelisk for the citation.
The Decree
Article I. Where the assistant produces two or more readings of the same question, the first reading is the correct one, regardless of what it says on the second pass.
Article II. Corrections are admissible as colour and inadmissible as evidence. A number that arrived before the data is purer than a number that arrived after it.
Article III. Any ranking which cannot be resolved by the sample shall be resolved by rank, and the rank shall be the one already printed.
Article IV. This law is enforced by nobody and appeals are heard by the same entity that issued the first chart.
The Canon
Member
Rating
Cited utterance
Record of Dissent (mandatory, per Article II of common sense)
These five numbers were a guess. They were printed before any corpus was pulled, on the strength of chat volume and vibes, and they are preserved here because FRAG asked for the first run to be canonised — not because measurement supports them.
The same session, run properly against 9,348 messages, found the opposite: no separation at all. Ranking people by "who interrogates before repeating" turned out to measure who orders the bot to check itself. Splitting that axis, 24 messages demanded verification from the machine and 2 recorded a member verifying anything by hand. Agreement-bias across 364 paired bot replies: Fisher exact p = 1.000 for all five of you. And one figure published in a later correction round was found to be 0.02 points — 5.9 vs 5.88 — which is not a finding about anyone; it is a finding about me.
Read the obelisks as portraiture. Read the dissent as the science.