Turn a market hypothesis into a testable system. From statistical arbitrage and cross-sectional factors to trend, mean reversion, and regime-aware strategies — built around your mandate.
Your edge.
Engineered.
Bespoke trading algorithms at the intersection of quantitative research, artificial intelligence, and exceptional engineering.
Estimate, against known truth.
A 205-parameter network is trained in this tab. The dotted curve is the best boundary any model could draw: we know it, because we wrote the generator.
- TRAINING STEP
- —
- HELD-OUT ACCURACY
- —
- MISCLASSIFIED
- —
Mode NEURAL, LEARNING A DECISION BOUNDARY. Assumption nominal. TRAINING STEP: not yet computed. HELD-OUT ACCURACY: not yet computed. MISCLASSIFIED: not yet computed.
An edge isn’t found.
It’s designed.
Markets are complex. Your technology shouldn’t be a black box. We design bespoke systems you can understand, validate, and own.
From first hypothesis
to the final line of code.
Explore more hypotheses with machine learning and AI-assisted research. Engineer features, evaluate forecasting models, and keep every experiment tied to reproducible evidence.
Connect the research to the market. Build broker adapters, event-driven execution, position reconciliation, and operational controls around the realities of your trading stack.
Investigate constrained portfolio problems through QUBO formulations and quantum-classical methods. Research with explicit classical benchmarks, measurable constraints, and honest results.
16 methods behind the matrix, one specification each: the formulation, the data and execution it needs, the validation protocol, the failure modes, and the primary sources. 15 of them run a computation you can operate in the page.
Every stage has a specification.
Ten stages separate a venue feed from a reviewed result. Published standards govern each one, and a short list of tools is in common use there. This is the map we engineer against; the reference holds every summary, caveat and owner page.
- Primary stage
- Also applies here
- Status changed
- ITCH
- CME MDP 3.0
- SBE
- FAST
- PTP (IEEE 1588)
- NTPv4
- ISIN
- FIGI
- CFI
- Tick / Lee-Ready / BVC
- Arrow
- Parquet
- IEEE 754
- ONNX
- ISO 4217
- GICS
- OpenQASM 3
- RTS 6 (primary stage elsewhere)
- FINRA RN 15-09 (primary stage elsewhere)
- ONNX (primary stage elsewhere)
- RTS 6 (primary stage elsewhere)
- FIX Latest
- FIX 4.2 / 4.4
- FIX 5.0 SP2 (status: Archived)
- FIX Session / FIXT 1.1
- FIXP
- Orchestra
- FIXatdl
- OUCH
- FIX tag=value
- Rule 15c3-5 (Market Access Rule)
- MIC
- Reg NMS (status: Amendments deferred)
- LEI
- MMT
- CAT
- RTS 22
- ISO 20022
- MiFID II Art. 4(1)(39)-(40)
- EU clock sync (ex-RTS 25) (status: Replaced RTS 25)
- RTS 6
- MAR Art. 12
- FINRA RN 15-09
- SR 26-2 (ex-SR 11-7) (status: Superseded SR 11-7)
Market data
Venue feeds are received, decoded and sequenced, and every event is stamped against a synchronized clock.
This stage in the referenceStandards and rules 06
- ITCHCurrent
Proprietary venue specification
- CME MDP 3.0Current
Proprietary venue specification
- SBECurrent
Industry standard
- FASTCurrent
Industry standard
- PTP (IEEE 1588)Current
Industry standard
- NTPv4Current
Industry standard
Technologies in common use 06
- Apache Kafka, Open source, new tab
- Redpanda1, Mixed, note 1, new tab
- Aeron2, Open source, note 2, new tab
- ZeroMQ, Open source, new tab
- Simple Binary Encoding (SBE)3, Open source, note 3, new tab
- Protocol Buffers, Open source, new tab
- Redpanda. Source-available under the Business Source License, not OSI open source; the license forbids offering Redpanda as a streaming or queuing service.
- Aeron. Repository moved from real-logic/aeron to aeron-io/aeron; old links redirect.
- Simple Binary Encoding (SBE). Repository moved from real-logic to aeron-io; old links redirect.
Research data
Ticks, bars and reference data are identified, stored and joined as of the moment each value was known.
This stage in the referenceStandards and rules 06
Technologies in common use 13
- kdb+ / q, Commercial, new tab
- ClickHouse, Open source, new tab
- QuestDB1, Open source, note 1, new tab
- TimescaleDB2, Mixed, note 2, new tab
- Apache Parquet, Open source, new tab
- ArcticDB3, Mixed, note 3, new tab
- Python, Open source, new tab
- NumPy, Open source, new tab
- pandas, Open source, new tab
- Polars, Open source, new tab
- Apache Arrow, Open source, new tab
- DuckDB, Open source, new tab
- Numba, Open source, new tab
- QuestDB. Some features are Enterprise-only.
- TimescaleDB. The vendor now trades as Tiger Data; the product name remains TimescaleDB.
- ArcticDB. Source-available under the Business Source License; its README states that production use requires a paid license from ArcticDB Limited.
Modeling
Statistical and learned models turn point-in-time features into forecasts, labels and probabilities.
This stage in the referenceTechnologies in common use 14
- scikit-learn, Open source, new tab
- XGBoost, Open source, new tab
- LightGBM1, Open source, note 1, new tab
- statsmodels, Open source, new tab
- arch, Open source, new tab
- hmmlearn2, Open source, note 2, new tab
- QuantLib, Open source, new tab
- PyTorch, Open source, new tab
- TensorFlow, Open source, new tab
- JAX3, Open source, note 3, new tab
- Hugging Face Transformers, Open source, new tab
- Chronos-24, Open source, note 4, new tab
- TimesFM5, Mixed, note 5, new tab
- Stable-Baselines3, Open source, new tab
- LightGBM. Repository moved from microsoft/LightGBM to lightgbm-org/LightGBM; old links redirect.
- hmmlearn. In limited-maintenance mode according to its README; the last release dates from October 2024.
- JAX. Repository is jax-ml/jax; documentation moved from jax.readthedocs.io to docs.jax.dev.
- Chronos-2. Benchmark rankings on the model card are the vendor’s own statements; usefulness on financial data has to be tested independently.
- TimesFM. TimesFM 3.0 weights (August 2026) carry a non-commercial license: commercial or production use of downloaded weights is not permitted, only use through Google Cloud services. Code and weights up to 2.5 are Apache-2.0.
Portfolio and risk
Forecasts become target positions through optimization under risk, cost, currency and exposure constraints.
This stage in the referenceStandards and rules 03
- ISO 4217Current
Industry standard
- GICSCurrent
Licensed classification
- OpenQASM 3Current
Open specification
Technologies in common use 11
- CVXPY, Open source, new tab
- OSQP, Open source, new tab
- HiGHS1, Open source, note 1, new tab
- SciPy (scipy.optimize), Open source, new tab
- Gurobi, Commercial, new tab
- MOSEK, Commercial, new tab
- Qiskit2, Open source, note 2, new tab
- PennyLane, Open source, new tab
- D-Wave Ocean SDK3, Open source, note 3, new tab
- CUDA-Q, Open source, new tab
- Cirq, Open source, new tab
- HiGHS. Solves LP, MIP and convex QP; it is not a mixed-integer quadratic solver, so cardinality-constrained mean–variance still needs an MIQP solver or a linearization.
- Qiskit. The separate Qiskit Optimization package was archived on 12 July 2026; IBM points to qiskit-addon-opt-mapper for problem mapping.
- D-Wave Ocean SDK. Its classical SimulatedAnnealingSampler is not a quantum computation.
Simulation
The whole strategy is replayed against history with fees, latency and queue position before capital is committed.
This stage in the referenceStandards and rules 02
- RTS 6Current
RegulationPrimary stage: Governance
- FINRA RN 15-09Current
Supervisory guidancePrimary stage: Governance
Technologies in common use 06
- LEAN, Open source, new tab
- NautilusTrader, Open source, new tab
- VectorBT1, Mixed, note 1, new tab
- Zipline Reloaded2, Open source, note 2, new tab
- Qlib3, Open source, note 3, new tab
- hftbacktest4, Open source, note 4, new tab
- VectorBT. The Commons Clause withholds the right to sell a product or service whose value derives substantially from the software, and it names consulting and support fees.
- Zipline Reloaded. In maintenance mode: its documentation says new feature development has moved to ML4T Backtest and recommends evaluating that for new projects.
- Qlib. The latest PyPI release dates from August 2025; a research framework rather than production infrastructure.
- hftbacktest. Community project whose examples and live connectors target crypto venues; the latest PyPI release dates from December 2025.
Deployment
A validated model is exported, packaged and served so that live scoring runs the code that was tested.
This stage in the referenceStandards and rules 02
Technologies in common use 05
- ONNX Runtime, Open source, new tab
- NVIDIA Triton Inference Server1, Open source, note 1, new tab
- gRPC, Open source, new tab
- FastAPI2, Open source, note 2, new tab
- Docker, Mixed, new tab
- NVIDIA Triton Inference Server. Suited to shared GPU serving for several models, not to a single-strategy deployment.
- FastAPI. A general-purpose web framework, not a trading-specific component.
Order routing
Target positions become orders that pass pre-trade controls and travel to a broker or venue over a session protocol.
This stage in the referenceStandards and rules 10
- FIX LatestCurrent
Industry standard
- FIX 4.2 / 4.4Supported legacy version
Industry standard
FIX Trading Community still supports FIX 4.2 and FIX 4.4 alongside FIX Latest.
- FIX 5.0 SP2Archived
Industry standard
Archived by FIX Trading Community as a legacy specification (last errata 9 December 2013) and no longer supported; superseded by FIX Latest.
- FIX Session / FIXT 1.1Current
Industry standard
- FIXPCurrent
Industry standard
- OrchestraCurrent
Industry standard
- FIXatdlCurrent
Industry standard
- OUCHCurrent
Proprietary venue specification
- FIX tag=valueCurrent
Industry standard
Regulation
Technologies in common use 07
- QuickFIX, Open source, new tab
- QuickFIX/J, Open source, new tab
- Interactive Brokers TWS API1, Commercial, note 1, new tab
- Alpaca Trading API, Commercial, new tab
- cTrader Open API2, Commercial, note 2, new tab
- MQL5, Commercial, new tab
- Pine Script and TradingView webhooks, Commercial, new tab
- Interactive Brokers TWS API. The API is installed from Interactive Brokers; the PyPI package named ibapi is a stale copy from December 2020.
- cTrader Open API. The official Python SDK has had no stable release since June 2024; a 0.9.3 release candidate followed in August 2024.
Venue access
Orders cross the last hop to the matching engine, where market-structure rules and the latency of the path decide what executes.
This stage in the referenceStandards and rules 02
Technologies in common use 08
- C++, Open source, new tab
- Rust, Open source, new tab
- Java (JVM), Open source, new tab
- LMAX Disruptor, Open source, new tab
- DPDK, Open source, new tab
- Kernel-bypass networking (OpenOnload, ef_vi)1, Open source, note 1, new tab
- FPGA feed handlers2, Commercial, note 2, new tab
- Linux CPU isolation and core pinning, Open source, new tab
- Kernel-bypass networking (OpenOnload, ef_vi). Native acceleration needs AMD Solarflare adapters; an AF_XDP mode covers other network cards.
- FPGA feed handlers. A technique, not a product; vendor latency figures are vendor claims and are not repeated here.
Post-trade and operations
Fills are reconciled, flagged and reported while metrics, traces and scheduled jobs keep the running system observable.
This stage in the referenceStandards and rules 05
Technologies in common use 05
- Prometheus, Open source, new tab
- Grafana, Open source, new tab
- OpenTelemetry, Open source, new tab
- Kubernetes1, Open source, note 1, new tab
- Apache Airflow, Open source, new tab
- Kubernetes. Suited to research batch work and supporting services; latency-critical paths normally run on pinned bare-metal cores instead.
Governance
Code, data and models are versioned, challenged and reviewed so that each result can be reproduced and each control evidenced.
This stage in the referenceStandards and rules 06
- MiFID II Art. 4(1)(39)-(40)Current
Regulation
- EU clock sync (ex-RTS 25)Replaced RTS 25
Regulation
MiFID II RTS 25 (Delegated Regulation (EU) 2017/574) was repealed with effect from 2 March 2026 and replaced by Delegated Regulation (EU) 2025/1155.
- RTS 6Current
Regulation
- MAR Art. 12Current
Regulation
- FINRA RN 15-09Current
Supervisory guidance
- SR 26-2 (ex-SR 11-7)Superseded SR 11-7
Supervisory guidance
SR 11-7 was superseded and replaced by SR 26-2 on 17 April 2026.
Technologies in common use 05
- MLflow, Open source, new tab
- DVC1, Open source, note 1, new tab
- Optuna, Open source, new tab
- Ray, Open source, new tab
- Git, Open source, new tab
- DVC. Repository moved from iterative/dvc to treeverse/dvc; documentation is at doc.dvc.org.
What changed
Status changes on record, as checked on 5 November 2026.
- EU clock sync (ex-RTS 25)
MiFID II RTS 25 (Delegated Regulation (EU) 2017/574) was repealed with effect from 2 March 2026 and replaced by Delegated Regulation (EU) 2025/1155.
- SR 26-2 (ex-SR 11-7)
SR 11-7 was superseded and replaced by SR 26-2 on 17 April 2026.
- Reg NMS
Compliance with the 2024 tick-size and access-fee amendments is deferred until the first business day of November 2027 by an SEC order of 11 June 2026 (Release 34-105656).
- FIX 5.0 SP2
Archived by FIX Trading Community as a legacy specification (last errata 9 December 2013) and no longer supported; superseded by FIX Latest.
- Open source
- Mixed terms
- Commercial
Entries were checked against their owners’ pages on 5 November 2026; versions and statuses are as of that date. License notes summarize each project’s published terms as read on 5 November 2026. They are not legal advice; read the license itself before relying on it. Names identify specifications and tools. They do not indicate an affiliation.
Your strategy.
Your ecosystem.
Research only matters when it can meet the market. We engineer for the APIs, platforms, and workflows your strategy needs.
Interactive Brokers
TWS and IB Gateway socket APIs, plus Web API trading and account workflows. Connectivity is scoped around authentication, market-data entitlements, and account permissions.
Integration scope depends on API availability, account permissions, market-data licensing, and your project requirements. Platform names identify integration targets; they do not imply a partnership.
Ambitious thinking.
Rigorous execution.
A promising backtest is a starting point. We build the research process, risk controls, and infrastructure that turn an idea into a considered system.
Define the edge
Your thesis, markets, constraints, and success criteria. A clear research question before a line of code.
MANDATE / HYPOTHESISTest the hypothesis
Quality data, chronological splits, realistic costs, and walk-forward tests. Evidence over a beautiful equity curve.
RESEARCH / VALIDATIONEngineer the system
Modular code, broker connectivity, execution logic, and explicit risk limits. Designed to be understood and maintained.
BUILD / INTEGRATEObserve & evolve
Paper trading, reconciliation, drift monitoring, and documented iteration. A deliberate path into production.
DEPLOY / MONITORA step ahead.
Grounded in evidence.
New models. Better questions. A clear-eyed view of what’s next in systematic trading.
Explore all researchA fill ledger that survives duplicate, reordered and replayed execution reports
A fill ledger that keeps the highest revision of each execution depends only on the set of reports received and is correct when that set is complete. A synthetic order with one correction illustrates this.
A Quadratic Turnover Penalty Is Not Hysteresis for a Regime Allocator
For a two-regime allocator with a one-day objective, a quadratic penalty on weight changes gives a smoother that trails every jump and a proportional cost gives a no-trade band. Only the band is hysteresis.
In an AI Factor Search, the Number of Candidates Tried Is Part of the Result
On 100 simulated noise factors with no leakage, a correctly scaled test still leaves six apparent discoveries. A threshold set by the trial count removes all six. In this ten-asset panel a cost filter keeps the best one.
Independent by nature.
Precise by design.
Mr. Piron is a quantitative research and engineering studio for people with a thesis and the ambition to build it. We bring scientific thinking, design clarity, and practical engineering to custom algorithmic systems.
Explore our identityYour mandate, not a template.
Every engagement starts with your markets, objectives, and constraints.
Complexity has to earn its place.
We benchmark advanced methods against clear, simpler alternatives.
Clarity is part of the deliverable.
Readable systems, documented decisions, and an explicit path from research to operation.
You have a thesis.
Let’s give it an edge.
Bring the idea. We’ll bring the research and engineering.