Weekly pulse · May 17 – May 24, 2026.
Auto-published from anonymous structural telemetry. Every Council run, every script eval, every analysis the system performs adds one row regardless of opt-in — no scenario text, no client data, no outputs. The reliability graph at /reliability-graph shows this data live; this page is the weekly digest.
Deliberations this week.
Tier distribution.
Three user-selectable tiers — Quick (~15s), Standard (~45s), FINRA Battle-Tested (up to 3 min). The split shows what depth of review visitors are choosing.
- Quick: 0 runs
- Standard: 11 runs
- FINRA Battle-Tested: 0 runs
- No tier specified (legacy callers): 12
Average wall-clock duration: standard 44.1s ·
Engine-path activity.
Cross-examination fire rate: 0.0% over 23 eligible runs this week. The chairman flags requires_cross_examination = true only when the dissent introduces a factor not already addressed in the synthesis. A low fire rate means the chairman's first-pass synthesis is already calibrated; a high rate means the second round is doing real work.
FINRA-examiner gap rate: 0.0% over 0 examiner runsthis week. The examiner reviews each chairman synthesis for Marketing Rule 206(4)-1 / IAA §206 / FINRA Rule 2111 / Reg BI / SEC 2026 Exam Priorities failure modes. A 0% gap rate combined with the examiner-validation harness's 5/5 catch rate on synthetic violations indicates the chairman is producing artifacts that don't trip the examiner — the verification stamp is the buyer-facing feature, not a quality gate that catches misbehavior.
Sources.
Where the deliberations came from. Each source distinguishes real-user activity from script-eval / admin-eval activity so the graph reads honestly.
other: 12 runsapi_stream: 11 runs
What this means.
This page is a transparency artifact, not a marketing instrument. The numbers are what the system actually did this week. The reliability graph at /reliability-graph visualizes the same data live. The /methodology page describes how the engine works architecturally. The /privacypage describes what telemetry is captured and what it isn't.
The page regenerates on every visit; the data is current to within the last polling window. There is no editorial step.
Historical archive: each ISO week is a stable URL. Last week (2026-W20) →
Public benchmark dataset: /benchmark — every head-to-head eval comparison row queried live from the production database, with anonymized JSON download. The data behind every claim on this page.