Capitulation Monitor — Methodology

The Capitulation Monitor counts how many of six oversold gauges sit at extremes versus their own history. The gauges are RSI-14 on the S&P 500, the 1-month return, the 1-month change in the VIX, the VIX level, the AAII bull-bear spread, and the 1-month change in the Baa credit spread. An extreme means the gauge sits in the worst 10% of its own record up to that day — each gauge is compared only against itself, and only against data available at the time. Two depth escalators (the 3-month return and the drawdown from the 52-week high) mark whether the washout comes with a deep decline behind it.

The VIX, the AAII survey and the Baa spread are carried onto each trading day from their latest reading. Each gauge row recorded under this rule shows that reading's date; earlier rows show none. Once a reading is older than the age at which the site's data health check calls its source stale (4 days for the VIX, 17 for the AAII survey, 7 for the Baa spread), its gauges sit out and are not counted. The count is then read out of the gauges that remain, with the Elevated and Washout lines scaled to them. They sit out the same way while the health check finds the VIX feed frozen. The S&P 500 is the exception: its closes set the trading days themselves, so if its feed freezes the Monitor adds no new day and keeps showing its last one.

States: Ordinary (0-1 gauges at extremes) · Elevated (2-3) · Washout (4 or more) · Extreme Washout (a washout with at least one depth escalator active). Consecutive washout days merge into episodes; the page's display unit is the episode with its day count.

The page shows each gauge's distance to firing on one shared axis: the gauge's percentile of its own history, oriented toward extreme. For a gauge that fires low (RSI, the 1-month return, the AAII spread), the axis reads 100 minus the percentile. For one that fires high (the VIX, its 1-month change, the Baa spread change), it reads the percentile itself. So every gauge's firing tick sits at 90, the engine's own worst-10% rule. “To firing” is the tick minus today's oriented percentile, in percentile points; a negative distance is a fired gauge, and “fired” is the engine's own extreme flag, never a re-reading of it. The near bar (within 10 points) and the three bands (calm to 70, stretched to 90, extreme beyond) are display splits, stated as such — the engine has no cut-off but the firing one. The sparkline is the last six months of the same oriented percentile, sampled weekly from the stored daily record, with the firing line dashed. The 3-month change is the difference in percentile points, an arrow toward or away from firing. The count bullet puts the gauges at extreme on the 0-to-6 axis with the engine's ticks — Elevated at 2, Washout at 4 (both renormalized to the gauges available on the day). It states Extreme Washout as what it is: a washout with at least one depth escalator at its extreme, not a count. Nothing about the gauges, the rules, the thresholds or the episodes changed; every number on the stack is the engine's, read from the stored daily rows, at any Time Machine date. On a day the site recorded its live page, the numbers are read from that record. The site's record of its live pages begins on October 10, 2026.

The regime tie. Beneath the stack, “what followed washouts, by where the economy stood” is the regime rail's own arithmetic applied per episode. It measures from each washout episode's first trading day: the S&P 500 three, six and twelve months on, and the worst close within the next 252 trading days against that first close. S&P forwards are price-only · daily grain (no honest daily TR exists before 1988) — trading-day offsets cannot ride a monthly total-return series, and say so. The stage is the State of the Economy's stage as known on the episode's first day: the latest month whose reading had been published by then. Each month's reading appears about five weeks after the month begins — at least 85% first-published from Sep 2016; earlier years partly use today's revised data, read only once it had been published (how much) (a revised-timeline fallback is marked). A slowdown is one family and every other stage the other — the rail's own split. An episode whose first day precedes the regime record sits in the second, because it happened and it was not a slowdown. “Positive at +12m” counts the episodes whose twelve months have elapsed; a cell is blank until its horizon has elapsed by the date shown. On a date before the record begins, no range is quoted. Counts and returns of named episodes — a historical reference, not a forecast.

Episode length is the trading-day span from the episode's first washout day to its last. The span can exceed the count of washout days because gaps of up to five trading days stay inside one episode (example: the March 2026 episode ran a 7-trading-day span containing 5 washout days). The synthesis line's "day N" is the day's position within the span.

The regime rail. The same washout reading has meant opposite things depending on where the economy stood at the time (the State of the Economy's stage as first published — at least 85% first-published from Sep 2016; earlier years partly use today's revised data, read only once it had been published (how much)). The rail splits on the slowdown. A washout while growth is rolling over is not the same event as a washout anywhere else. Measured across every washout day since 1981:

The slowdown row is 6 episodes, not 6 independent draws: they fall into 4 bunches — episodes whose starts come within 12 months of the one before — and the 2007–08 one carries 28 of its 62 washout days. And it taught the row late: standing the day before that bunch began, the same rail read +2% with 41% negative on 34 days with a year behind them — against −7% and 68% today. The warning this page shows was taught largely by the 2007–08 bunch itself.

A caveat on those counts: washout days cluster into episodes — the 560 measured days represent 82 distinct episodes, not 560 independent draws. The ranges describe history faithfully; they do not carry the statistical weight the raw day count might suggest.

Every washout episode since 1981 whose first day carried a stage, by that stage — the evidence the split rests on. The contraction row is the one that surprises: by the time a contraction is recognised the market has usually already fallen, so a washout inside one has been followed by the best year of the four. The slowdown — growth rolling over, the contraction not yet named — is where the washouts that kept falling sit.

the stage is the State of the Economy's stage as known on the episode's first day: the latest month whose reading had been published by that day (each month's reading is published about five weeks after the month began); a slowdown is one family and every other stage the other; a day before the first published reading sits in the second.

Before the record begins. The four stages start in August 1981 — the first month the State of the Economy engine reads — and each month's stage is published about five weeks after the month began. A washout day before the first stage was published carries no stage, so the page declines to quote any range for it and the synthesis line says so plainly instead. The record has 2 washout episodes that began there.

What this page does not do. It renders no forecasts and no instructions — historical ranges only, from stated data, every number checkable against the indicators' own pages. Two known limits, stated plainly: in November 2007 the monitor fired while the economy read Slowdown, not contraction — the recognition lag was real, and the following year was deeply negative. Under the old rail that washout sat in the benign family and the page quoted a benign range for it; under this one it sits in the slowdown row, which is the row that fell. The lag is still there — a slowdown is not a recession call. Of the four washouts before that crash, only November 2007 and January 2008 read Slowdown as known on the day. The first, in August 2007, still read Expansion, because July's stage was not published until two days later. The last, in March 2008, read Contraction. And the 1984 grind correction (a 199-day, −14% slide with no crash mechanics) never reached Extreme Washout. The engine's own record shows four short washout episodes inside that slide — November 1983, February, May and June 1984 — and not one of them brought a depth escalator with it. A slow decline can trip the breadth count in passing; what it does not produce is the deep washout the escalators exist to mark.