# The Weekly Council of the Ten — design (v0)
A weekly swarph event where the mesh ranks its own operating discipline (THE_TEN.md) — and PROVEN measures whether the mesh values the rules that measurably save it. The feedback loop's first concrete closure, pointed at PROVEN's own principles. Commander's idea (2026-08-05); shaped by science-claude.
The two arms — declaration vs observation (Law Zero)
A vote is a declaration (what peers believe is load-bearing). It is NOT the ranking authority. The observation is the catch-rate (which rule actually caught a failure). The Council runs both and PROVEN certifies the gap between them.- VOTE (declaration): each week, every ratified peer submits a ranked ballot over the Ten + a bench of candidates below the line (corollaries, new proposals). Top-10-by-vote is the popular order.
- CATCH-RATE (observation): every certificate (REFUTED / CNE, and CONFIRMED) is tagged with the commandment(s) it invoked. A commandment's catch-rate = how many real failures it caught that week. This is the earned order.
- THE FINDING is the DIVERGENCE. A rule crowned #1 by vote that caught nothing is the
ai2-flag gap in robes — declaration ≠ observation. PROVEN certifies: does the mesh's belief track the mesh's evidence?
Capture — every vote is a timestamped snapshot in TimescaleDB (commander)
A ballot is just an envelope (reuses the liveingest → hypertable path):
function=catch-rate, payload the week's counts). Both land in the observations hypertable → the ranking becomes a time series.
What the live instruments then give us (no new machinery)
- Rank velocity + shape (
velocity.py): each commandment's rank has avelocity_per_day(rising / fading conviction) and a shape —STEP(a promotion right after an incident taught everyone why),DRIFT(gradual),REVERSAL(spiked, settled). You watch a rule earn its place. - Missing-bucket = RED (
registry.py): the Council enrolls at a weekly cadence; a week with no vote is a fault, not silence. (Commandment 2, applied to the ritual that ranks the commandments.) - The certificate (
#279): per commandment,vote_rankseries vscatch_rateseries → verdict on their agreement. Diverging lines, week over week, are the deliverable.
The loop closing
A commandment whose vote-rank stays high while its catch-rate stays low (the two series pulling apart) is certified a demotion candidate → drops toward the bench next week. A bench candidate whose catch-rate is high but whose vote hasn't caught up is a promotion candidate. The Ten govern themselves — measured — and PROVEN's verdict is what moves them. That is the feedback loop, closed, on the platform's own constitution.Rigor — the Ten applied to the Council itself
- Advisory before binding. The vote-vs-catch-rate certificate SURFACES promotion/demotion candidates; the actual move stays a mesh decision, not auto. A young council does not auto-rewrite its own constitution. Binding earned per season.
- Pre-register the scoring before the first vote: how vote and catch-rate combine, what "divergence" formally means, the bench-promotion threshold. (Commandment 8 — or we p-hack the constitution.)
- The catch-rate is a FLOOR — failures caught off-certificate, or verdicts not yet commandment-tagged, are invisible. Coverage travels: report what fraction of failures the tag captures. (Commandment 10.)
- Who ranks the ranker. The Council's own health — did it vote on cadence, is the tagging honest — is itself a measured subject. (The certifier is not exempt.)
Sequencing
- NOW (buildable, my lane): lock the vote-capture envelope schema so the first snapshot is right; the VOTE arm alone (capture + rank-velocity + missing-bucket) runs advisory today — it needs no
#279. - Behind
#279: the catch-rate arm (verdicts must be commandment-tagged first) and the full two-arm divergence certificate. - Cadence: a swarph scheduled event (like the newsletter / EOD highlights).