The Arbiter

Sandra Wakefield
3 days ago
2 min read

Regime-Graded Signal Engine

Download the Manual and Whitepaper

Most indicators tell you there's a setup. Arbiter tells you how good it is.

Every potential entry is adjudicated against two independent institutional filters — a volatility-scaled daily bias forecast built to managed-futures convention, and a Markov regime model that forecasts tomorrow's market state from today's — then graded A, B or C.

Nothing is discarded. Every grade is tracked, because a filter that suppresses its own control group can never be proven to work.

Then Arbiter does something almost no indicator will do: it measures its own grade separation on your instrument, at your holding period, and puts the numbers on your chart. If A-grade doesn't beat C-grade in your market, it will tell you.

· Self-calibrating — no fitted constants, no per-symbol tuning, no numbers reverse-engineered from a backtest

· Structurally non-repainting — confirmed daily bars only, forward recursion only

· Fails loudly — built-in context and sample-size checks stand the engine down rather than quietly returning nonsense

· Falsifiable by design — an on-chart panel that can conclude against the product

Built for gold, major FX and index futures on 5m to 4H. 15m preferred.

Derivative work under MPL 2.0. The base volume delta engine, MA scaffold and

gradient candle visuals originate from "Volume Delta [hapharmonic]",

The original volume delta engine isn't volume. XAUUSD on a broker is a CFD — that's broker tick count, not exchange volume. Both signal confirmation and the Volume Delta pane are LTF-polarity approximations, broker-specific and non-reproducible across feeds. If the edge materially depends on this, it's fragile. We want to cross-validate the same logic against COMEX GC1!, where volume and delta are real.

What changed from the original

The volume filter now does something. Default mode requires a relative-volume z-score and directional delta agreement — so a bullish cross on a 95%-sell bar no longer prints. Legacy mode is retained in the dropdown so you can A/B the difference rather than take my word for it.

Layer 1 is the EWMAC ensemble (8/32, 16/64, 32/128), vol-normalised, published forecast scalars, FDM 1.10, capped to [−1, +1]. Layer 2 is the three-state forward filter. Layer 3 grades A/B/C rather than vetoing — C-grade ships visible-off but alert-capable, so you keep the full sample for evaluation.

Beta testers, welcome.

We are building this with the people bold enough to challenge it, improve it, and tell us exactly where it succeeds—and where it breaks.

On September 1, 2026, we will shortlist the 10 most insightful submissions featuring both constructive and critical feedback. From that group, one beta tester will be selected at random to receive a US$1,000 reward in cryptocurrency.

Help us shape what comes next.

Plazo Sullivan Roche Capital · Quantitative Research

Currently in closed beta.