Correction Dashboard

Baa−10y is the price of corporate default risk. A widening of ≥ +50bp in three months means credit is repricing corporate balance sheets faster than equities — the seller is fundamental, not mechanical. Strongest depth predictor in the backtest: correlation with final event depth ρ = +0.65 (p < 0.001); widening accompanied 11 of 12 major episodes since 1974.

Precedent: GFC 2007–09: +406bp → −56.8%. Dot-com 2000–02: +168bp → −49.1%. 1980 credit controls: +156bp → −17.1%. 1973–74: +148bp → −48.2%. COVID 2020: +136bp → −33.9%. Counterexample: the 1987 crash widened only +36bp in a −33.5% event — purely mechanical cascades can bypass credit entirely (that is what RV_ACCEL is for).

How to read: TRIGGERED + router ESCALATE = Tier-2 entry regardless of drawdown, and pushes the depth range toward its upper half. Δ3m ≤ 0 (narrowing) is required to exit Tier 2.

Level, not flow. A Baa−10y spread above 250bp marks the funding-crisis regime where refinancing is impaired and selling becomes forced — the dose-response into capitulation-class outcomes. Distinct from CREDIT_IMPULSE: the impulse catches the repricing in motion; the level confirms the regime has arrived.

Precedent: Only the GFC produced a sustained crisis-level regime (max widening +406bp) — the deepest event in the catalog at −56.8%. 1973–74, 1980, and 2020 all widened > 130bp without holding the crisis level and stopped at lesser (still severe) depths.

How to read: One leg of the only non-price Tier-3 entry: Sahm FIRED + credit crisis + policy CONSTRAINED.

The Fed-put switch. Below 4% CPI the central bank can ease into weakness; above it, easing risks un-anchoring inflation, so drawdowns run deeper before help arrives. Backtest: median event depth 13.9% with CPI above 4% vs 9.4% below (ρ = +0.30, p = 0.03).

Precedent: The two great inflation-constrained bears: 1973–74 (CPI accelerating through 12%, −48.2%) and 1980–82 (CPI 12.6%, −27.1%). 2022 (CPI 7.5% at the peak): −25.4% with no Fed put until inflation broke. Contrast 1987, 1998, 2020 — policy free, violent declines but fast recoveries.

How to read: CONSTRAINED is an overlay, never a trigger: it forces the depth-engine range to its upper half and is one leg of the non-price Tier-3 entry. CPI is marked provisional pending revisions.

A ≥ +80bp rise in the 10-year over three months compresses valuations and forces de-risking — but standalone it preceded a correction in only 10 of 18 instances (56%), barely better than a coin flip. The backtest therefore demoted it to a conditioning variable: state-relevant only in combination with a top-tercile CAPE or a credit trigger. Keeping a weak signal demoted is a design feature, not a gap.

Precedent: Rate-led and contained: 1994 bond massacre (−8.9%), 2013 taper tantrum (−5.8%), 2023 10y-to-5% (−10.3%). Rate shock into rich valuations: 2022, compounding to −25.4%.

How to read: ARMED is a flag, never a trigger. The card note states whether the combined condition (CAPE top tercile or credit TRIGGERED) currently makes it state-relevant.

The Sahm rule — 3-month average unemployment ≥ 0.50pp above its 12-month low — is the most reliable real-time recession marker, and recessions are what turn corrections into bears: Sahm triggered in 56% of ≥15% events vs 24% of shallower ones. But raw Sahm stays elevated through early recoveries and produced the Aug-2024 false positive, so FIRED requires two confirmations: claims 4-wk MA +15% YoY AND negative revision breadth.

Precedent: Triggered in 18 of 54 catalog events, including every capitulation-class recessionary bear (1973–74, 1980–82, 2000–02, GFC, COVID). The false-positive record the confirmation legs exist to filter: the 1991–92 cluster of five shallow pullbacks with Sahm still elevated after the recession, and Aug-2024 (Sahm 0.53, no recession, −8.5%).

How to read: ARMED with incomplete confirmation is capped by design — missing breadth data cannot fire the gate. FIRED is one leg of the non-price Tier-3 entry.

The behavioral core of the system: after the first −5% close, does the market snap back or roll over? A quick regain is a shakeout; a failed bounce plus a lower low reveals sustained institutional distribution. In the 1990–2022 daily cohort (n = 33), escalated events extended to ≥10% 58% of the time vs 7% for clean recoveries — an 8× separation, the strongest binary discriminator in the study.

Precedent: ESCALATE cohort: 1990 Gulf War, 1998 LTCM, 2000–02, 2007–09, 2011 debt ceiling, 2015–16 China/EM, 2018 Q4, 2022. CLEAN cohort: 1997 Asia, 2013 taper tantrum, 2019 trade-war dips, 2021 Evergrande. Router-moot: 1990 January break, 2018 Volmageddon, 2020 COVID — crashes that reached −10% before the 30-td window could resolve; the speed was the answer.

How to read: OPEN starts at the −5% cross. ESCALATE requires BOTH legs: no peak regain within 30 trading days AND a close below the initial-cross low within 60. ESCALATE + credit impulse TRIGGERED = Tier-2 entry regardless of drawdown.

When 10-day realized volatility runs ≥ 1.75× the 60-day baseline, the mechanical seller class engages: vol-targeting funds, risk parity, CTAs, and dealer gamma hedging all de-risk on the same signal, so selling begets selling regardless of fundamentals. Directionally validated in the backtest (small n) — held as activation context, not a driver.

Precedent: The mechanical archetypes: 1987 portfolio insurance (−33.5% with almost no credit signal), 2010 flash crash, 2018 Volmageddon (−10.2% in nine sessions), 2020 COVID (fastest ≥30% decline in the catalog), 2024 yen-carry unwind.

How to read: TRIGGERED marks mechanical-class engagement — expect gap risk and overshoot. Feeds activation context and depth-engine cascade scoring; not a transition input.

VIX is endogenous — it spikes with the drawdown, not ahead of it. It appeared in only 19% of Tier-1 events but 77% of Tier-2 and 100% of Tier-3: an excellent grader of cascade severity and a worthless predictor. The engine encodes this structurally — the transition function’s input type does not contain the VIX field, and a regression test asserts it can never alter a transition.

Precedent: Every ≥20% event in the catalog ran VIX ≥ 30 sustained; most sub-10% events never did.

How to read: TRIGGERED grades the severity of the current tier, nothing else. If a tier ever changed because of VIX, that would be a bug — report it.

Starting valuation is dry tinder: it does not time the spark, it multiplies how often sparks catch. With Shiller CAPE in its top trailing-30-year tercile, corrections start ~3× more frequently — but once an event begins, CAPE has zero power to predict its depth. The trailing-30y window keeps the tercile honest across regimes (an absolute CAPE bar would read “always expensive” after 1995).

Precedent: 2000 dot-com and 2022 both launched from top-tercile readings into ≥25% events; the 2010–2016 correction cluster started from mid/low terciles and stayed shallower on average.

How to read: ARMED = elevated event-start frequency and a red flag in depth cascade scoring — nothing more. Never a trigger.

10y−3m inversion means the market prices policy as restrictive enough to break growth — historically preceding recessions with a long and famously variable lead measured in quarters, not days. That makes it a regime flag: it tells you which kind of cycle you are in, never when an event starts.

Precedent: Inverted ahead of the 2000–02 and 2007–09 bears. The 2022–24 inversion was the longest on record while the associated equity event (2022, −25.4%) arrived before the traditional recession lag resolved — the standing case study in why this flag is never timing.

How to read: ARMED raises L0 vulnerability context. Participates in no transition and no depth stage.