Research · validation · limitations

What the diagnostic catches — and what it doesn't

The methodology has documented strengths and documented limitations. Both are load-bearing. This page exists so technical visitors can inspect the evidence directly rather than taking the dashboard's regime call on faith.

The core formula

Λ_F measures non-commutativity of factor covariance evolution:

Λ_F = ‖[ F_t , dF_t/dt ]‖_F / ( ‖F_t‖ · ‖dF_t/dt‖ + ε )

Where [A, B] = AB − BA is the matrix commutator. F_t is the rolling covariance matrix of factor returns (CMKT, CSMB, CMOM, CVOL) standardized over a 105-day window. The output is log-compressed, smoothed over 14 days, and 2-day-lagged for actionability.

R_t is a separate behavioral-cascade composite of cross-asset fear, crypto-specific reflexivity, macro stress, and risk-off proxies — VIX (0.40) + BTC volume |z-score| (0.30) + yield-curve stress (0.20) + DXY momentum (0.10). Hysteresis bands (enter ≥65 / exit ≤55) and 5-day EMA smoothing prevent whipsaw.

Crashes require Q4: high Λ_F (structurally unstable rotation matrix) AND high R_t (crowded positioning + macro stress). Q3 alone is rotation. Q2 alone is crowded-but-stable. Q1 is the trend-following baseline.

Retrospective validation ledger

Tested in-sample on the 4-asset crypto cross-section (BTC, ETH, LTC, XRP) over 2015-01-01 → 2021-12-31. Detection threshold P75 on Λ_F percentile within a 30-day lookback. Lead time = days between first threshold cross and the event.

EventClassMax Λ_F %LeadResult
Dec 2018 Fed Panic — Powell capitulation; ~ −50% BTC drawdown setup Institutional rotation P97.2 17 days DETECT
Nov 2021 BTC Top — ~$69K peak before 2022 bear Institutional rotation P98.4 30 days DETECT
Oct 2025 BTC Top — pre-2026 correction (in-window for prior-POC) Institutional rotation P79.0 4 days DETECT
May 2021 mid-cycle dip — BTC retraced ~50% from April high without regime break Negative control (no rotation) P38.9 CORRECTLY SILENT
Jun 2022 Luna aftermath — post-Terra crypto cleanup, no institutional rotation Negative control (no rotation) P63.9 CORRECTLY SILENT

Validation outcome: 3/3 DETECT on documented institutional-rotation events + 2/2 correct IGNORE on negative-control mid-cycle dips. This is the "5/5" figure on the dashboard. But it is a small-N retrospective fit on events that were partly known to the methodology authors — it is sanity-check evidence, not proof. The prospective walk-forward record starting 2026-06-05 is the binding evidence stream.

Known limitations

The diagnostic is materially weaker on pure exogenous shocks. Several documented examples below where Λ_F either did not fire ahead of the event or fired too late to be actionable.

Across vonlambda's expanded 47-event ledger, Method C (a sister variant using direct asset returns rather than derived factors) reports ~78.7% detection at ~26.85% false-positive rate. The factor-model variant on this dashboard is more selective (catches fewer events, fewer false positives). They're different operating points on the precision/recall curve, exposed in parallel on the live dashboard for cross-validation.

What is on the live dashboard vs. what is here

Pre-registration discipline

The parameter freeze is 2026-05-31. The out-of-sample run executes once on 2026-06-01. From 2026-06-05 the prospective walk-forward record begins accumulating. Calibration changes after the freeze date void the pre-registration claim — this is enforced by tagged release. The HMM canonicalization audit panel on the dashboard exposes the per-refit overlap-agreement record in real time (51 refits over the 2022-2026 OOS window so far; 0 fallback firings).

Patent + IP

The diagnostic methodology is the subject of US provisional patent application 63/903,809 (R. J. Mathews, filed October 22, 2025). Full backtest specification is published as a pre-registered research document.

Source & reproducibility

A reference implementation of the core Lambda-F factor-model + Method C variant lives at github.com/vonlambda/lambda-f-dashboard. The QuantSurf dashboard runs the SPEC v1.7 factor-model methodology and exposes Method C in parallel for cross-validation.

← Back to dashboard