HERO ART SLOT — THE DOCTOR — reserved for the operator's portrait
SCROLL ▾
#1 — OBSERVE
Market Monitor
ENTER ▸
Every instrument on the board — live quotes, candles, and the news that moves them. Click a symbol and the chart takes the room.
ART SLOT
#2 — TEST
Paper Trading Lab
ENTER ▸
Three accounts forward-testing the engine against reality since April. The calibration record nobody can retroactively fake.
ART SLOT
#3 — SIMULATE
The Engine
ENTER ▸
Hyde. Five thousand futures at a time — GBM and Merton jump-diffusion under CAPM drift, VIX blend, and CAPE discipline.
ART SLOT
#4 — EVIDENCE
Does It Work?
ENTER ▸
PIT calibration, Brier skill, comparator deltas — the engine graded against what actually happened, window by window.
ART SLOT
#5 — ALLOCATE
Portfolios
ENTER ▸
Your real allocations, projected through the calibrated engine. Save scenarios, compare futures, decide with distributions.
ART SLOT
▶ SELECT A STOCK
VIX
—
fear index
MARKET REGIME
—
—
MODEL
—
Base Case
SPY 5D / 1M
—
S&P 500 performance
▶ ADD HOLDINGS AND RUN SIMULATION
ASSUMPTIONS
PORTFOLIO VALUE PROJECTION
TERMINAL VALUE DISTRIBUTION
ASSET PARAMETERS — 5Y HISTORY · CAPM-IMPLIED RETURNS (Rf=4.5%, ERP=5.5%)
—
PORTFOLIO VALUE
RECENT TRADES
SIDESYMBOLQTYREF PRICEFILL PRICESLIP
No trades yet.
PREDICTIONS vs REALIZED
METRICPREDICTEDREALIZEDDELTA
Accumulating — check back after 21+ trading days.
DOES HYDE'S OUTPUT PREDICT ANYTHING USEFUL?
Three model variants, forward-tested daily against realized outcomes. This page shows
where each one is calibrated — and where it isn't.
CALIBRATION — PIT PER MATURED PREDICTION
Probability Integral Transform: where the realized outcome landed inside Hyde's predicted distribution.
A well-calibrated model scatters evenly around the dashed 0.50 line — clustering near 0 or 1 means
outcomes are landing outside the predicted range more often than they should.
DIRECTIONAL ACCURACY vs PASSIVE BENCHMARK
Did each account correctly call up vs. down over its 21-day prediction horizon — and is the equal-weight lab NAV ahead of a buy-and-hold passive benchmark in the same universe?
ABLATION — WHICH INGREDIENTS ACTUALLY HELP
Pairwise comparison of directional calls on shared dates. Isolates whether jump-diffusion and CAPE are adding accuracy, or just adding complexity.
METHODOLOGY
Brier skill score is measured against the empirical base rate, not a hardcoded 0.25 — a positive
score means Hyde's probability calibration beats a naive guess informed by the benchmark's own win rate.
n_eff is the Bartlett-corrected effective sample size; daily 21-day-horizon predictions overlap heavily,
so raw N overstates statistical power by roughly 20×. The per-account sign test above does not yet apply
that correction — read its p-value directionally, not as a formal significance claim. PIT figures reflect
the live data pipeline as of today; a quantile-basis correction is in progress under separate engineering
review and historical PIT values will be refreshed once it lands. This is a research instrument under an
active efficacy freeze, not investment advice.
DEMO NOTICE — client files created here are visible to every visitor and are wiped without warning. Use pretend people and pretend numbers.
THE DOSSIER
NO CLIENTS ON FILE YET.
DR. JECKYL
The doctor is in.
A consultation takes about two minutes — a handful of plain questions, no jargon,
and every one of them tells you why it's being asked. At the end you get a written
prescription: an allocation built for your situation, with the reasoning shown,
verified against five thousand simulated futures by the engine.
