[ APPROACH ]
How the work is structured.
The strategies, the regime detection layer, the validation pipeline, and the boundaries.
01 / DISCIPLINE
What we do.
We build and operate a portfolio of automated trading strategies across US index, equity, energy, and rates futures, with the supporting research infrastructure to validate, evolve, regime-condition, and continuously scan for new opportunities across instruments. The strategies are systematic. Every entry, exit, and management decision is rule-based and reproducible from code. The infrastructure is mostly what we spend our time on: a research engine with second-resolution exit modeling, a custom panel-builder for cross-asset and macro data, a regime detection layer that drives position sizing on top of unchanged strategy logic, and a sim-to-live reconciliation harness that keeps the research stack honest against the live execution platform.
The work sits at the intersection of three disciplines: quantitative research, software engineering, and operational risk management. Most retail algo trading collapses all three into "I built a backtest." We treat them as separate, load-bearing systems.
02 / DIFFERENTIATION
Where the discipline lives.
Portfolio, not a single optimization.
We run multiple intraday strategies that use structurally different edge mechanisms. Fade setups on US-session opening ranges, breakout setups anchored to the pre-cash-open print, session-anchored strategies on the European session. Internal tracking has shown low daily P&L correlation across the live period. The portfolio is the unit of decision, not the individual trade. Single point of failure on any one strategy or instrument is treated as a portfolio risk to be engineered out, not a constraint to live with.
→View the algorithm fleetCross-asset regime detection.
Most retail traders never address regime conditionality at all. The ones who do typically rely on equity-volatility regime classifiers (VIX states, realized vol bins). Our regime layer is built on cross-asset correlation matrices, yield curve features, and 5-day cross-asset returns. A macro-relationship-aware design, not a price-action design. It produces daily archetype labels, a continuous stress score that is orthogonal to the archetype, and a severity verdict that drives the operational disposition of each strategy. The detection layer's outputs are used as a position-sizing overlay, not an entry filter. A deliberate design choice that preserves the strategy's edge attributability and avoids the curve-fitting feedback loop that entry-gating produces.
→Read the regime overlay memoA1
Healthy Growth
Risk-on tape, positive correlations, low cross-asset stress
A2
Late-Cycle Compression
Tightening rates, compressed vol, building correlation drift
A3
Liquidity-Debasement
Loose policy, weak dollar, hard-asset bid
A4
Flight-to-Quality
Stress regime, bond bid, equity sell, correlations break
Versioned strategy libraries.
Rather than chasing a single all-weather parameter set, each strategy maintains a library of tunings indexed by regime. The regime classifier selects which version is live. The retuning infrastructure is fast enough that a new regime-specific tuning can be calibrated, validated, and shelved for deployment within a single research session. Like a race car that gets retuned for each track rather than driven on universal settings.
Continuous opportunity scanner.
We built a multi-instrument pattern scanner that audits candidate intraday edges across US index, equity, energy, and rates futures. It produces ranked candidate setups with statistical significance scores, walk-forward stability checks, and explicit out-of-sample replication tests. The point is not to find a winner on every pass. The point is to maintain a continuous research pipeline that surfaces structurally different edges as candidates for new strategy development, so the portfolio keeps adding diversifying components rather than stagnating on its existing strategies.
Negative findings are results.
Multiple instrument scans have produced candidate edges that did not survive out-of-sample testing. Documenting and shelving them, rather than continuing to torture the data, is the methodology point. The convergent failure patterns across instruments became their own findings about scanner methodology and produced a tighter validation protocol used in subsequent work.
Sim-to-live reconciliation.
The research engine is treated as a digital twin of the live execution platform. Every strategy version is bar-by-bar reconciled between the two engines under a strict drift tolerance before any size scaling is permitted. Engineering bugs in fill modeling, bar labeling, multi-leg execution, and within-bar event ordering have been caught and documented inline in the strategy files. Prior backtest numbers carry explicit "do not cite" tags when superseded by corrected versions.
03 / VALIDATION
How the validation actually works.
- ▸Locked holdout windows defined before parameter sweeping. Results that don't replicate on the locked holdout are discarded, not retuned.
- ▸Walk-forward validation across multi-year windows that span deliberately different regimes, not just the favorable training window.
- ▸Sample-size floors enforced per cell. Anything thin gets flagged as low-confidence and excluded from sizing decisions.
- ▸Multi-policy within-bar audit. Setups whose results vary materially under conservative, geometric, and optimistic fill assumptions are discarded.
- ▸Neighborhood robustness check. Chosen parameters must outperform baseline across most of the surrounding parameter grid, not be lone peaks.
- ▸Smoothness metrics (drawdown, consecutive loss streaks, monthly variance, exit-mix stability) prioritized over raw return, because smoothness travels across regimes better than P&L does.
- ▸Every published P&L number ships with its trade log. No floating numbers.
04 / TOOLING
The research stack.
Python (NumPy, pandas) for the research engine, simulation, regime classifier, and opportunity scanner. C# (NinjaScript) for live execution.
NinjaTrader 8.
Custom multi-resolution pipeline (1-second tick chunks, 1-minute, 15-minute) spanning multiple years across US and European session instruments, including index, equity, energy, and rates futures. Continuous-contract stitching with rollover handling, exchange-native symbol formats (ICE Impact, CFE Pitch), and macro data integration (FRED yield fetches).
Cross-asset feature panel updated daily across multiple asset classes. Outputs consumed by strategy disposition logic.
Multi-instrument pattern audit harness with locked-holdout discipline, multi-policy within-bar resolution, and significance scoring. Produces ranked candidate setups for downstream strategy development.
Vectorized, second-resolution, validated against the live platform's replay engine under a drift tolerance.
05 / DIFFICULTY
Why this is harder than it looks.
Execution versus simulation drift is structural.
A simulation that ignores fill geometry, within-bar event ordering, and platform order-state semantics will silently invent P&L. Tracking the drift, defining a tolerance, and refusing to ship past it is the discipline.
Bar-level event reconstruction matters.
Within a single bar, breakeven, take-profit, and stop-loss can all trigger. The sequence decides whether a trade is a winner or a loser. One-minute resolution hides that ordering. Getting it right requires second-level data and an explicit ordering convention you commit to and audit.
Regime-dependent edges are hard to classify.
A strategy that prints in one window and bleeds in another might be tuned to a real regime change or tuned to noise. Distinguishing these requires structural diagnostics: exit-reason mix shifts, cell-level consistency across the directional axis, cross-asset correlation signatures. Not just P&L curves.
Anti-overfitting discipline costs returns.
Refusing to retune after a bad month, refusing to add ad-hoc gates that fix one historical event, refusing to scale up before regime-shift evidence accumulates. Every one of these decisions feels like leaving money on the table. They are the entire reason serious systematic trading is hard.
06 / BOUNDARIES
What we're not.
- ▸Not a signal service, alert vendor, or call room.
- ▸Not selling courses, mentorships, or "systems."
- ▸Not running curve-fit backtest theater. Out-of-sample regime testing, sample-size floors, sim-to-live parity, and portfolio-level validation are non-negotiable.
- ▸Not a vendor-backtester user for parameter decisions. The platform's built-in optimizer is disqualified for intraday timeframes because it overstates results through fill-model artifacts.
- ▸Not discretionary. Every position is rule-based, logged, and reproducible from code.
- ▸Not a single-strategy operator. The portfolio is a stack of independently-validated edges with documented correlation and regime properties.
07 / BACKGROUND
Background.
Operated by Brandon Showalter, whose background combines technology infrastructure, high-accountability sales execution, and full-time systematic futures research. The work brings a software-engineering posture to a discipline that retail traders typically approach as a charts-and-intuition exercise. Treating the validation layer as load-bearing infrastructure, rather than optional tooling bolted on once a strategy seems to be working, is the part of the work most consistently missing from the broader retail systematic community.