Where we are
Stage is derived from the readings below. It is never assigned from a calendar or an expectation.
Stage 0 — Calm. no corridor broken; pressure 0.36. Broken right now: none.
Facility in use: no - largest single operation day in 30d was $0.10bn. Standing Repo Facility take-up, the only reading that can put this page in Stage 2. Fire is $5bn on one operation day or $1bn on two consecutive ones, summing both daily operations and excluding small value exercises. Lapse clock: 35 quiet operation days of 20.
The plumbing lines — MOVE, credit spreads, repo, auction demand — are what precede a policy pivot. The fiscal line is already far advanced and is tracked separately below, because debt arithmetic forces a pivot eventually while the plumbing decides when.
How likely an answer, and by when
One question, one number. The event is a closed list, printed on every build: a new Fed lending facility or repo window; a reopened or expanded swap line; a resumed asset-purchase programme; a stated long-end Treasury operation (Twist, YCC or a yield cap); or an unscheduled cut of half a point or more. Scheduled FOMC decisions, discount-rate changes, speeches and this page's own inference are excluded.
P(answer within 12 months) 21.6% · band 11.1% to 39.0% · within 3 years 60.1%
3 months 5.1% · 6 months 10.4% · 12 months 21.6% · 18 months 32.4% · 2 years 42.7% · 3 years 60.1%
Date facts, and they are percentiles of a distribution, not a schedule: one chance in ten by Mar 2028 · even odds by Apr 2029 · nine in ten by Apr 2031.
What this number is. A hazard fitted on 2003-01..2026-11 (13 event months) of monthly readings across the seven corridors with recoverable history, scored against 13 frozen event months, with the fire levels varied by a scale factor and the noise of the fit integrated out over 20,000 draws (seed 20261121). The build refuses to print unless the model reproduces the event count it was fitted to: it does, at 13.001 against 13.0, a gap of 0.01%.
What it is not. No corridor is fired today, so this is the calm-state hazard with the tape contributing almost nothing: ignoring every reading on this page and using the calm rate alone gives 19.7% against the 21.6% printed above. The readings are worth about 1.8 points right now. The number is for sizing, never for timing.
The rule the code runs, printed: a corridor counts 1.5 if it is plumbing and fired, 1.0 if fired and not plumbing, and nothing for being close. Partial credit was tested and refused — 0 of 25 near-firing months contained an event. The model's covariate reads 0.00 today; 1.00 lines sit outside their fire bands.
The eleven corridors
Each line shows how far it has travelled from calm (0%) toward its firing level (100%). plumbing marks the six that constitute an actual plumbing failure.
The fiscal clock — the part already advanced
net interest reaches 22.0% of federal receipts — 2027 / 2028 / 2031 (p10 / median / p90 of 20,000 simulated paths). Today: 21.94%.
This is a simulated crossing date from the dated-band model, not a market reading. It says when interest cost becomes unpayable out of revenue, which is what forces monetisation in the end — but it does not say the plumbing breaks first.
After the break: monetisation
Once a facility is created, the probability that the balance sheet is expanding within twelve months is 80%. 4 of the 5 post-facility episodes in the calibration table expanded the balance sheet within a year; 2011 used Operation Twist instead of QE.
Why the model is shaped this way — the calibration
Time from the first plumbing break to the first actual Fed facility, in six real episodes, re-dated to the FIRST facility in each (TAF, 12 Dec 2007, not QE1 - the 15-month figure this page used to print was both mismeasured and off support). The measured range is one day to seven weeks, which is why the hazard cannot be convex at the long end: before the plumbing breaks there is no clock on a pivot; after it breaks, the pivot is usually weeks away.
How the number is calculated — and what would break it
The hazard. h = 1 − exp(−exp(a0 + a1·A)), a grouped-time
cloglog fitted on 2003-01..2026-11 (13 event months) of monthly corridor readings, where A is the number of fired
corridors with hysteresis (arm at 1.0, disarm after three months below 0.90). 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 — a corridor that is calm today can fire in month nine, which is exactly what the
history says happens.
The receipt. Summed over the fitted panel the model produces 13.001 events against 13.0 that occurred — a gap of 0.01%. That check is the reason this page is allowed to print a number at all: it catches a fire-level edit, a doubled response rate, or a target quietly swapped, in one line.
The lag. Measured from the first fired month, the five verified episodes give {0, 1, 1, 7, 9, 9} months to an answer. That is the honest response function, and it is not a clock: one day to seven weeks, with the tail set by how long the Fed can tolerate the break, not by the tape.
What is refused here. Partial credit for a corridor merely close to firing — tested, likelihood ratio 1.68 on one degree of freedom, and 0 of 25 near-firing months contained an event. The 15-month 2007 lag — off support and mis-dated; TAF came on 12 Dec 2007. And a Jan 2019 "pivot with no corridor broken": high-yield OAS printed 5.44 on 3 Jan 2019, above this model's own fire level, so that episode is not an example of anything and has been removed from the table above.
What would break it: a new facility announced with no corridor having fired for a sustained period, or a stress episode where every corridor stays past its fire level for more than a year with nothing announced. Both are logged, dated, and scored — not argued about later.