Overview

A thesis is a claim about where the world goes next. Pick one and see what it actually commits you to: the regimes it makes likely, how long today keeps mattering, and how soon it says the thing it is really betting on arrives.

Your thesis
Falsifier
Breadth broadens while realised vol stays at the floor for a full quarter — crowding unwound without a repricing.
A thesis without one of these is a mood. This is the line that retires the claim.

The state of the world, under this thesis

monthly · fitted 2026-09-30

Node area is the long-run share of time spent in that regime. Flow density on each edge is its transition probability — the picture is the matrix, not an illustration of it.

Now
Expansion
Breadth positive, credit calm, dips bought.
Steps to Shock
6.0
Expected first passage from the current regime, in months.
Memory half-life
1.1 mo
How long until today's regime stops informing the forecast.
Entropy rate
1.90 bits
Per step. Lower means the thesis claims a more determined world.
Sampled
0
Realised steps in this walk, against the stationary target.

What your thesis moved

against the no-view base rate
Expansion
26.2%-20.7
Late cycle
19.6%—
Shock
23.3%+15.5
Deleveraging
19.2%+3.3
Repair
11.8%+2.0
generator Dirichlet posterior, 119 transitionstilt multiplicative, then renormalisedresolution monthly

Where these numbers come from

fitted
Observations
120
monthly, 2016–2026
Universe
507
names in the panel
Transitions
119
across 25 cells
Panel says now
Late cycle
as of 2026-09-30
How much of each row is data, not prior
Expansion
88%61 obs
Late cycle
71%20 obs
Shock
53%9 obs
Deleveraging
69%18 obs
Repair
58%11 obs
Dirichlet posterior, not a plain fit. 119 transitions across 25 cells is thin: maximum likelihood would hand you probabilities of exactly 0.00 and 1.00, which are the sample mistaken for the world — and on screen those would look more authoritative than the honest prior they replaced. Each row starts at a hand-specified prior worth 8 pseudo-observations and moves toward the data exactly as fast as there is data. 5 of 25 cells have no observations at all and remain pure prior.