Behavioral guardrail — brake yourself, not the market strong evidence · about you
Aimed at the one lever the evidence says matters most for outcomes — your behaviour, not the market (the evidence is about investor behaviour in general; this journal itself is untested). It audits your decisions, holds your pre-commitments, and adds a pause exactly where panic happens. Walled off from every market reading by design.
Volatility-sizing dial — a down-only throttle risk, not alpha
Size down when forecast volatility exceeds the comfort band you type — it caps at 100% (never levers up); you set the band and pull the lever. Shown with the honest drawdown↔turnover trade-off, no Sharpe.
Crash-context badge — the regime you're already in coincident, not a forecast
A pure annotation: turbulence, absorption (variance share of the top modes, vs its own history), shock clustering and skew read together as the current tail regime. Wired to nothing — it moves no dial and predicts nothing.
Diversification reality check fair-weather caveat
How many bets you really have and how few names carry your risk — with the honest warning that average correlations understate how they collapse in a crisis.
Contrarian sentiment thermometer context only · gutted
Panic / neutral / euphoria from VIX, valuation and turbulence extremes — shown with the forward-return distribution from here as proof it's a thermometer, not a clock. No trade, no buy/sell, ever.
Scenario stress-test — synthetic crises to stress-test a size you already chose FHS · not a forecast
The transparent cousin of a generative model: thousands of synthetic next-month paths, each resampling real vol-standardized moves re-scaled to today's volatility. A fuller range of bad months than the few real crises — to stress-test your sizing, never to predict.
Drift & rebalance — a state, not an alert band, not a trigger
Where a year of relative performance has pushed your mix off the target weights you set in the Crisis-replay card — shown as a passive state. Never a red action; rebalance on a rule, not a signal.
Stock-bond ballast read describes the past quarter
Whether bonds are currently diversifying your equities or moving with them (like 2022) — present-tense context for your ballast, with no persistence claim and no allocation steer.
Directed spillover network — who leads whom
Lagged cross-correlation asymmetry (LCCA): an arrow i→j means i's past returns forecast j's future more than the reverse, over the analysis window. Permutation-gated: the 12 strongest asymmetries are drawn, solid only if they beat a circular-shift null, dashed if they don't. Red = net transmitter, blue = net receiver. Lead-lag is predictive association, not causation. Global note: the international ETFs trade US hours, so any "lead" the US shows over them is partly mechanical — they reprice overnight foreign news during the US session — not pure information flow.
net transmitter (leads)
net receiver (lags)
Asymmetry matrix & net transmitter / receiver
Left: directed asymmetry A(i→j)=D(i→j)−D(j→i), anti-symmetric (red i leads j, blue j leads i, near-white = pure co-movement, no leader). Right: net directional spillover per asset (right/red transmits, left/blue receives).
Cascade simulation — shock propagation descriptive drill
Inject a shock at one market and propagate it along the directed spillover network — the way a failure cascades through a water system. Directed reachability over lead-lag links at or above the strength you set (all estimated links, not only the permutation-gated arrows). A what-if on the current estimated network, not a forecast.
Network structure — hubs, choke points & amplification graph-theoretic
Borrowing from evolutionary graph theory and metabolic pathways: where the spillover network concentrates (preferential-attachment hubs), which market is its structural bottleneck (choke point), and whether its topology amplifies or damps a shock.
Systemic coupling — spectral radius ρ̂ & absorption ratio
How tightly the whole basket is moving as one thing. ρ̂=λ_max/N is the dominant eigenvalue of the rolling cross-asset correlation matrix; the absorption ratio is the variance share in the top modes (Kritzman-Li). High = little local shock-absorption capacity. This is the closest honest analog of the water system's fixed topology — except here the topology is estimated and rewires in a crisis.
Factor baskets — what actually moves together hierarchical · empirical
Your nominal nine asset classes are highly redundant. This clusters the basket by its real correlation structure into the handful of independent factors that actually drive it — the "actual baskets," not the labels.
Game-theoretic lens — beauty contest & coordination risk framing
The strategic structure behind the numbers: why direction cannot be forecast (higher-order beliefs), and when the system is vulnerable to a coordinated run (global games).
Resting bid vs ask size — live book microstructure crypto · seconds · descriptive
A live microstructure readout — not a rung on a prediction ladder: how much resting size currently sits on the bid vs the ask of the Binance.US book (top 20 levels), refreshing every few seconds. Descriptive flow, not a forecast: any move it anticipates is smaller than the spread and decays in seconds, so it is not directional alpha and not tradable through a browser. A skewed book is liquidity resting there now — not a buy or sell call. Crypto only (the only free live order book); live mode only.
Financial turbulence — system-level surprise (Mahalanobis)
How statistically unusual today's whole cross-asset move is vs its recent covariance (Kritzman-Li). A multivariate extension of the per-asset surprise / variational-free-energy lens.
0.20
Structural geography — market-state manifold (diffusion map)
The nonlinear generalization of the coupling panel: every dot is a 10-day cross-asset state of the macro names, placed by a diffusion map so similar regimes sit together. Calm states cluster; crises (2020/2022) form a stressed arm. Where today sits — and which past state it most resembles — is a coincident read on the regime, never a forecast. It is a nonlinear generalization of PCA/spectral structure.
Anticipation — where stress starts, & does the transmitter actually lead? predictive test
The companion paper's actual engine made live: a VAR + Diebold–Yılmaz directed volatility-spillover decomposition picks the net transmitter (who exports stress) vs the loudest (most-volatile) market, and runs an out-of-sample test of whether the transmitter truly leads. Maps where anticipation has structural leverage — not a price forecast.
Alpha Lab — walk-forward stress→trade backtest research · net-of-cost
A backtest, not sold alpha. It tests, honestly, whether the stress signal monetizes: net-of-cost PnL and an SPY buy&hold benchmark lead; gross is buried support. Direction is a chosen stance — the data gives you stress, not direction. The likely honest result is the edge bleeds out after friction.
Crisis replay — your mix through past stress windows historical · not predictive
Set the weights you choose and replay them, daily-rebalanced, through a real past stress window. Pure historical replay of data already in the console — no model, no forecast, no recommendation. It shows how a mix actually held up, and how its protection shrank as assets coupled.
Premia allocator + stress-test — your passive-income plan, made concrete education · not advice
Set how much you put in each risk premium and see the honest tradeoff: the income you'd passively collect, an illustrative total return, and the drawdown your mix actually took in one real stress window, the COVID crash (Feb–Apr 2020), next to its deepest drawdown anywhere in the embedded history (with dates) — the COVID window is not necessarily the worst. Income is not what costs you drawdown here: the highest-yield sleeves (cash, Treasuries, IG credit) are the defensive ones. Across the presets, a larger equity share raises the illustrative total return and deepens the drawdown, while income falls. No forecast, no recommendation.
Forward-volatility lab — predict what's predictable purged CV
The constructive flip side of the Alpha Lab: a ridge model on the risk-state features, evaluated with purged + embargoed cross-validation, predicting next-5- and 20-day realized volatility (learnable) versus direction (flat at both horizons). Shows where the data actually carries information.
Forecast reliability — the meta-learning layer regime-conditional
A layer above the forecasts: it does not predict — it says when each forecast can be trusted. Volatility is learnable, but more so in some regimes than others; direction is learnable in none. It describes forecast reliability only — it is not an input to exposure, and it never calls direction.
Overnight vs intraday — the one free directional crumb risk premium · show, not trade
Asked "what free data predicts short-term direction?", the honest answer is: almost nothing — except this. Splitting each day into its overnight (close→open) and intraday (open→close) halves reveals a real, documented, index-level directional regularity. It is a liquidity-provision risk premium, not tradable alpha — shown as the closest free data gets, and why it still is not money.
Long-horizon valuation thermometer years · descriptive
What the canon — Cochrane's discount-rate revolution and Shiller's CAPE — identifies as the one genuinely directional free-data signal, and it lives only at the multi-year horizon, with wide bands and a documented out-of-sample-failure record. Where we sit versus ~145 years of CAPE history (1881 on), from Shiller's dataset as embedded (through Jun 2026).
Factor companion — harvested premia (1–12 month) cross-asset · descriptive
The cross-asset / time-series premia the canon (Ilmanen, AQR) identifies as genuinely moving relative returns at the 1–12 month horizon — trend, carry, low-risk — read off prices the console already holds. Harvested statically, never timed.
Cross-sectional factor demo — what "doing what quants do" returns honesty exhibit · no edge
The genuinely accessible version of the pros' playbook: switch the target to cross-sectional (rank 85 stocks), manufacture breadth, combine three documented free-data factors (12−1 momentum · 1-month reversal · low-vol) in a no-fit composite, market-neutral quintile long-short, monthly. Frozen result of a real backtest — and a demonstration of why it returns nothing.
⚠ Survivorship: universe = 85 current (2026-picked) large-caps — failed/delisted names excluded. This inflates momentum/reversal and may flip low-vol's sign; the true effect is ≤ what is shown. Picking liquid mega-caps separately deflates the real premia (which live in small/illiquid names). The number is squeezed from both sides.
Is the signal there?
Composite rank-IC (forward 1mo, out-of-sample)+0.021
95% CI · p-value[−0.021, +0.062] · p=0.33 — indistinguishable from zero
t-stat vs the field's discovery hurdlet=1.0 (need t>3.0 — Harvey-Liu-Zhu)
Per-signal ICmom +0.029 · reversal +0.045 · low-vol −0.041 (suspect)
Does it pay? — cost ladder (gross +4.6%/yr · Sharpe 0.26)
− 5 bps/side+3.2%/yr · Sharpe 0.18
− 10 bps/side+1.8%/yr · Sharpe 0.10
− 20 bps/side−1.1%/yr · Sharpe −0.06
− 30 bps/side−3.9%/yr · Sharpe −0.22
Break-even cost~16 bps/side — below honest all-in (20−30 bps spread+impact + short borrow ~50 bps/yr)
Max drawdown (net)−37.6%
Does it persist?
IC first half (2016−21) → second half (2021−26)+0.042 → −0.00
…though the decay is itself not significant (Δt≈1.0) — you can't separate "real edge that faded" from noise sloshing around zero.
Does breadth rescue it? — the one lever that could
Same model, 85 cherry-picked mega-caps → 589 names (√breadth 9→24)IC +0.021 → −0.003 · gross Sharpe +0.26 → −0.06
Broadening the universe didn't tighten a positive edge — it collapsed it (break-even fell to −3 bps/side: a loss even trading free). The +0.02 was selection in the liquid mega-caps; breadth amplifies real signal, so here it amplified ~nothing and exposed the apparent edge as noise. This is Grinold's law run in reverse: √N × (a true zero) = zero.
No investable edge — and that is the finding. This is the honest, no-fit, free-data version of the only accessible levers (change-the-target + breadth), and it correctly returns ~zero: a full-sample IC indistinguishable from zero, decayed to nothing, with honest costs (turnover + short borrow) eating the entire +4.6% gross. What it proves is epistemic — the accessible levers are honesty instruments that cleanly reject a fake edge, not machines that manufacture a real one. Pros make the same tiny IC pay with what free data can't buy: a delisting-inclusive universe, microcap access (premia ~19% vs ~9%), a securities-lending desk, and execution that cuts turnover below the cost cliff. Demo only — never a signal, never a recommendation.
Forward macro — recession & financial conditions months horizon
Where the best free forward-content data actually carries signal: the yield curve, credit spreads, and financial-conditions index — leading recession and stress at a months horizon (not days). The constructive complement to the vol lab's finding that these overfit a short-horizon model.
In plain English — what the latest data says
Auto-written from the latest readings (live feeds if reachable, otherwise the embedded snapshot — the heading says which date), for a non-specialist. Updates with the data (and with the cursor in Backtest).
Perspectives — documented frameworks as lenses interpretive, not predictive
Live numbers read through ten publicly documented investor/economist frameworks. Each is really a stance — contrarian↔trend, value↔momentum, risk-on↔risk-off, micro↔macro, instability, fat-tails, factor-balance, sizing-discipline — wrapped in a memorable name, chosen so the set spans what this console actually measures. Generated from the current data; interpretive lenses "in the spirit of" each thinker, not their words, endorsements, predictions, or advice. The aggregate view synthesises them.
Structural backdrop — distribution & real economy
The wealth-distribution and real-economy data Gary Stevenson's thesis actually rests on — now on the board (it was the honest gap in his lens). Slow-moving (quarterly/annual, lagged) and US, so it is structural context, not a trading signal.
value = latest / as-of cursorright column = ~10-year change (pp for rates & shares, % otherwise)
Fragility thermometer sort: basket ▾
0–100 = how unusually fast each asset's fragility indicators have been rising (their 60-day trend) versus its own history, on three channels (persistence · amplitude · tail). A rate of change, not a level: a steadily volatile asset reads ~50. High ≠ crash; it can persist for years. Three mini-bars = the sub-scores.
calmelevatedleast resilient vs own norm
mini-bars: persist amplitude tail
Indicator detail —
The seven critical-slowing-down-analogue indicators (three channels) behind the score, each with its honest market failure mode. They are not independent votes (persistence channels especially overlap).
Surprise lens (active-inference analogue)
One-step negative log-likelihood of a simple causal AR(1)+EWMA-vol model — an analogue of variational free energy, not the full Active-Inference/EFE engine. Rising = today's moves are more surprising to a simple model.
Shock clustering — Hawkes self-excitation
Do large moves beget more large moves? Fits a self-exciting point process to ≥2.5σ return shocks; the branching ratio n is how self-reinforcing clustering is (→1 = critical). It is the temporal twin of ρ(Φ)→1.
Scenario sandbox (drill)
Inject a synthetic structure to see how the directed engine responds. Clearly a drill — never live data.
Data health
Embedded real-history snapshot is the floor; live endpoints refresh the tail when reachable. A single feed failure never blanks the console. Cross-asset panels only use dates every market in the panel shares: all 51 for the directed network (so in Backtest it is empty before crypto starts on 2023-09-21), and the 48 non-crypto names for coupling, turbulence and the manifold (back to 2016). A series that stops early holds back every panel that includes it, so the as-of date is the oldest series end shown below. Macro (FRED) is published on a daily schedule and checked anonymously from this site; original observation dates and the embedded fallback stay visible. CAPE (Shiller) remains a separately dated embedded snapshot.
Embedded macro fallback; observation periods are shown separately from source retrieval.