Methodology v1.1 — frozen

How the number is made.

Every rule below was fixed in advance and is versioned. Nothing here is tuned after the fact, and nothing is adjusted to make a past reading look better than it was.

Why these six?

The seven components

Six signals plus the Combination Engine. Weights are fixed and sum to 100.

The six cross-market patterns

The Combination Engine tests for six recognised patterns of channels moving together. A pattern qualifies only when every one of its inputs clears its threshold — one channel moving alone is never a pattern. When several qualify, the strongest is reported.

The bands

Boundaries are public and fixed. A band changes only after the score has persisted across the boundary — a single day's move does not reclassify.

Low is calm. The score is a stress classification, not a forecast.

What did not work

Four tactical tests were registered in advance, to see whether the score could be used to time entries and exits. All four failed. They are published here because a methodology that only shows its successes is not evidence of anything.

Exit on entering Defensive, re-enter on leaving Underperformed buy-and-hold over the replay period. The exits were usually right about stress and wrong about price.
Reduce equity exposure proportionally to the score Reduced drawdown modestly, reduced return by more. No improvement in risk-adjusted terms.
Buy when the score falls back below 21 No reliable edge. The signal that stress has passed arrives well after the recovery has started.
Trade the dominant signal directly Worse than all of the above. The components measure stress transmission, not direction of return.

The score is built for preparation, not timing. That is a conclusion drawn from these tests, not a disclaimer added around them.

Data, and how late it is

Every input is public. Each carries a publication lag, and the score uses conservative lags rather than the most recent possible value — the same discipline the 20-year replay used on every historical day.

The Dollar index is published weekly by the Federal Reserve, so its daily values arrive about a week behind — disclosed here rather than smoothed over. Equity and gold feed the Combination Engine only; neither is a weighted signal in its own right.

The five model allocations

Each band has one published model allocation. They are educational reference models — a way to see what more or less defensive looks like — and nothing more than that.

Long Treasuries, short-duration TIPS and gold hold fixed weights at every level — 40% together. Total equity falls exactly five percentage points a band and the difference goes only to short Treasuries. Equity keeps the same internal split throughout: half U.S. total market, a quarter U.S. small-cap value, a quarter developed ex-U.S.

What each sleeve means

The model defines asset sleeves. It does not require you to hold any particular fund — the vehicles named below are the proxies used to test each sleeve in research, not holdings to buy.

Fair questions

Why not housing? Why not stocks?

Because crises originate anywhere and arrive through a small number of channels. 2008 began in housing and reached retirement portfolios through credit spreads, which widened months before equities broke. 2020 began with a virus and arrived through funding stress and volatility.

Stocks are excluded for a second reason: by the time equities have fallen, the information is no longer early. Equity drawdown does appear inside the Combination Engine, as confirmation — never as a primary signal.

Why not real-time?

Published data is settled data. Intraday figures get revised; these numbers do not move after the fact.

It also keeps the live score honest against its own history. The 20-year replay used only the data available on each historical day — no peeking at revisions that arrived later — and the live score runs the same no-peek discipline. A faster number would not be comparable to the record published beside it.

Is this predicting a crash?

No. It is a stress classification. It reports how much strain is visible in these channels today — not what happens next. See the four failed tests above, which is where that claim was tested and rejected.

Why did the score not change when the market fell?

Because a falling market is not by itself financial stress. Equities fall for many reasons, most of them ordinary. The score moves when the channels through which stress reaches a portfolio move — and those can stay calm through a correction, or turn well before one.

The reverse also happens: the score can rise on a day markets are up.

What would change these rules?

A methodology change is versioned and published, and the whole 20-year history is recomputed under the new rules so the record stays comparable. The current version is v1.1, frozen. Changes are not made to improve how a past reading looks.