The one line that can move the stage
The sensor sums both daily repo operations, excludes small value exercises, and drops any future-dated operation as an error. Lapse clock: 35 quiet operation days of 20. State DORMANT.
The rule this page runs
The page moves to step 2 — The leak — the day any one of the eleven lines goes past its fire level. It stays there even if two or more of the six core plumbing lines break, because the pipes breaking is not the same event as the plumber arriving. It moves to step 3 — The plumber — only when the Fed's own operations data shows a real facility in use. Being close never counts, and a pile-up of broken lines never stands in for the facility.
Today 0 of 11 are past their fire level, and 2 are close. Close counts for nothing.
The eleven lines
Each row is one of eleven. A missing feed keeps its row and prints no reading this run — the count above does not change.
What the state words mean
- calm — well inside its normal range.
- moving — drifting toward the fire level, still inside normal.
- almost at fire — within a quarter of the fire level. This still counts for nothing.
- past fire — the line has broken. One broken line moves the page to step 2.
The colour of a bar never carries the state; the word does.
What these numbers are not
- Not a score. There is no total, no grade, and no ranking of danger.
- Not a forecast. A line past its fire level says something has broken, not that anything will be rescued.
- Not a trade signal. Nothing here is advice.
Method: how this is measured, and what would break it
Exactly what the code does, in order. The Standing Repo Facility sensor is read first: if it reports an observed operation ($5bn on one operation day, or $1bn on two consecutive operation days, both daily operations summed, small value exercises excluded) the step is 3. Otherwise, if two or more of the six core plumbing lines are past their fire level, the count is computed as step 3 and immediately demoted back to step 2 — the code's own comment reads “two or more corridors broken; no facility observed yet”. Any single broken line gives step 2 directly. Step 4 is never assigned automatically from these lines; a dark facility feed prints as blind and cannot promote the page.
The hazard. h = 1 − exp(−exp(a0 + a1·A)), a grouped-time
cloglog fitted on a panel of monthly readings across the seven lines with recoverable history,
where A is the number of broken lines with hysteresis (defined below). Today
A = 0.00 and the monthly hazard is 1.81% per month. Cumulative probability is
1 − ∏(1 − h), with the tape evolving month to month inside every
draw.
The receipt. Summed over the fitted panel the model produces 13.001 events against 13.0 that occurred — a gap of 0.01%. The build refuses to print a number unless that check passes: it catches a fire-level edit, a doubled rate, or a target quietly swapped.
Hysteresis. A line arms at 1.0 (past its fire level) and disarms only after three months below 0.90, so a single noisy print cannot flip the count.
What is refused here. Partial credit for a line merely close to firing — tested, likelihood ratio 1.68 on one degree of freedom, and 0 of 25 near-firing months contained an event. A 15-month 2007 lag — off support and mis-dated; TAF came 12 Dec 2007. And a January 2019 “pivot with no line broken”: high-yield OAS printed 5.44 on 3 Jan 2019, above this page's own fire level, so that story is not an example of anything.
The stress index. The full index lives at stress.myagentworkers.com. Today it reads 4.8 (watch), pressure 0.36. That pressure figure is published for reading the tape and is not the hazard's input — the hazard reads A, the count of broken lines.
The older calibration set (kept for audit, not used as page content): Sep 2019 repo spike (SOFR spiked to 10% on 17 Sep → NY Fed repo operations, same week, ~1 day); Mar 2023 SVB (SIVB failed 10 Mar → BTFP announced 12 Mar, ~3 days); Mar 2020 covid (credit spreads blew out from 24 Feb → CPFF/PDCF/QE 17-23 Mar, ~4 weeks); Aug 2011 downgrade (US downgraded 5 Aug → Operation Twist 21 Sep, ~6 weeks). These four are the retired support set; the six dated episodes on Past cycles are what the history tab prints.
The lag pool, break to first facility, in months: {0.03, 0.16, 0.26, 1, 1.08, 1.55}.
What would break this model: a new facility announced with no line having broken for a sustained period, or a stress episode where every line stays past its fire level for more than a year with nothing announced. Both are logged, dated and scored — not argued about later.
Model r6-2026-11-21, refit 2026-11-21, 20,000 draws, seed 20261121.