Skip to main content

Frequently asked questions

What is MTS1B?​

A 29-repo open-source institutional-grade quantitative trading stack under Apache 2.0. Covers the full lifecycle: data ingest → research → portfolio construction → execution → operations → reporting.

Why 29 repos instead of one?​

A monolith pays for everything every time. A hard-split lets contributors clone the 3 repos they care about and ignore the 26 they don't. We documented the design rationale in detail.

Where do I start?​

Is MTS1B production-ready?​

The current state is pre-alpha. Foundation (the types library) is real code, tested. Other repos have code ported from the source monorepo but with bridge-stage import paths — full internal consistency is still a work-in-progress. See STATUS.md for the latest.

Don't run live money through this stack until you've validated end-to-end against your own funds.

What's the license?​

Apache 2.0. Use it for anything — commercial, internal, research. No copyleft. Explicit patent grant.

Can I sell a product built on MTS1B?​

Yes. Apache 2.0 permits commercial use without obligation to open-source your additions. Attribution required (keep the LICENSE files).

How does MTS1B compare to backtrader / zipline / vectorbt?​

MTS1Bbacktraderziplinevectorbt
LicenseApache 2.0GPL-3.0Apache 2.0Apache 2.0
Backtest engineGPU-accelerated (mts1b-GPUbacktester)CPU, event-drivenCPU, event-drivenCPU, vectorized
Live tradingYes (via mts1b-oms + mts1b-brokers)YesQuantopian only (dead)No (separate vectorbt PRO)
Multi-assetEquities + options + fx + futures + crypto + sports + PMMostly equitiesEquities onlyEquities + crypto
Multi-fund treasuryYesNoNoNo
Risk gates7-gate pipelineNoneBasicNone
Plugin SDKYes (mts1b-pluginsdk)LimitedLimitedLimited
Docs siteYes (this one)OKOK (frozen)OK
MaintenanceActiveMostly abandonedAbandoned 2020Active (PRO is paid)

We owe debt to all three. They handled real-world cases we draw from.

I'm a current backtrader user. How do I migrate?​

See Migration from backtrader.

What Python version?​

3.11+. We use modern type hints (X | Y instead of Union[X, Y]) and match statements. 3.13 is supported.

Why Python? Why not Rust / Go / C++?​

Quant primitives are dominated by NumPy/SciPy/pandas + CUDA kernels (cupy). The hot path is C-bound. Python is the right glue language for accessibility + ecosystem.

For latency-critical OMS state machines, we have a clean Protocol boundary — a Rust implementation could plug in via pluginsdk. Not yet contributed; PRs welcome.

Do I need GPUs?​

No. mts1b-GPUbacktester has a CPU fallback (numpy + scipy). GPU gives 10-100x speedup for large universes but isn't required.

Why pydantic v2?​

  • Validation: every cross-repo message is shape-checked
  • Performance: pydantic v2 is rust-native, fast
  • Schema export: model_json_schema() + OpenAPI integration
  • Frozen-by-default: immutable wire models

What's NATS for? Why not Kafka / RabbitMQ / Redis Streams?​

NATS hits a sweet spot for MTS1B's scale:

  • Sub-millisecond pub/sub for live data
  • JetStream gives durable streams + ack semantics (Kafka-style)
  • Single Go binary; runs on a Proxmox LXC with 256 MB
  • TLS + JWT auth out of the box
  • Subject hierarchy (mts.v1.oms.>) is excellent for filtering

Kafka is heavier than we need. Redis Streams aren't durable enough. RabbitMQ is operationally finicky.

Where are secrets stored?​

HashiCorp Vault. Every service reads from secret/mts1b/<service> via mts1b_platform.config.load_config. The mts1b-deploy tool can bootstrap a Vault for you with mts1b-deploy vault bootstrap.

Local development can use a .env file — but never in production.

Where do logs go?​

OpenTelemetry → Loki / Grafana Cloud (configurable). Structured JSON via mts1b_platform.logging.get_logger. Secrets are auto-redacted by the redact filter.

Where does metrics + tracing go?​

OpenTelemetry → Prometheus / Tempo / Jaeger (configurable). Every primitive in mts1b-platform emits counters + spans labeled by service + operation.

How do I add a new factor?​

Tutorial 3 walks through the full lifecycle. Short version:

from mts1b_quantkit.factors import register

@register("f_my_factor")
def f_my_factor(panel, /, h=21):
    return zscore_cross_sectional(panel.close[-1] / panel.close[-h] - 1)

The name must start with f_. Then it's discoverable via mts1b_quantkit.factors.get("f_my_factor").

How do I add a new broker?​

Tutorial 5. Implement BrokerProtocol, register via entry points, ship as a plugin via pluginsdk.

How do I add a new market-data provider?​

Same pattern as brokers but with MarketDataProtocol. See foundation.protocols.

How does the dedupe contract work?​

Across the source monorepo we found 17 patterns duplicated 3-6 times each (HRP, Black-Litterman, Sharpe, Telegram dispatch, etc). Each is consolidated into exactly one canonical implementation. CI runs an AST scan to verify no duplicates outside the canonical home.

Full table: Concept — Dedupe contract.

Can I use just one repo?​

Yes. Each repo is independent. You can:

  • Use just mts1b-quantkit for HRP / walk-forward in your own pipeline
  • Use just mts1b-brokers to talk to IBKR / Coinbase with a unified API
  • Use just mts1b-foundation for the typed wire format

How do I run the test suite?​

# Per repo
cd mts1b-foundation
pip install -e ".[dev]"
pytest -v

# Cross-repo integration tests (require running NATS + Postgres)
cd /apps/MTS1B-oss
docker compose -f extraction/tests/docker-compose.yml up -d
pytest extraction/tests/

What's the release cadence?​

  • v1 (foundational 12 repos): months 1-3
  • v2 (engines 11 repos): months 4-7
  • v3 (community 6 repos): months 8-12
  • LTS v1.0: month 18

See Release waves for the full schedule.

How do I report a security issue?​

DO NOT use GitHub issues. Email [email protected] or use the GitHub Security Advisory interface on the affected repo.

How do I contribute?​

PRs welcome. Read CONTRIBUTING.md in any repo. Sign off your commits with git commit -s.

For non-trivial changes, open a Discussion first.

I have a bug. What do I do?​

  1. Search existing issues
  2. If new: open an issue on the relevant repo, include repro steps + Python version + OS
  3. For urgent issues: tag in Discord #help

I have a question. Where do I ask?​

Will MTS1B always be free?​

The code is Apache 2.0 in perpetuity. We may eventually offer hosted services (managed deploy, managed alerting, managed bot infra) but the core code stays free.

Who runs MTS1B?​

A small group of contributors. Maintainer cohort visible on each repo's CODEOWNERS. The project is independent (no corporate parent).

Why is the documentation so good?​

We wrote it as the implementation contract for the code itself. Per the docs-first workflow, the API surface lives in docs BEFORE it lives in code. Hard to ship a function whose docs you can't write.

What's a "session" in MTS1B?​

A trading session — a market-open-to-close window. MTS1B distinguishes:

  • RTH (regular trading hours) — 9:30-16:00 ET for U.S. equities
  • Pre — 4:00-9:30 ET
  • Post — 16:00-20:00 ET
  • IBEOS — 20:00-3:50 ET (IBKR overnight for select symbols)
  • 24h — for crypto, FX (24×5)

See platform.calendars and MarketCalendar.

What's a "fund" in MTS1B?​

A NAV-attributed sub-account. Every order belongs to exactly one fund. Funds have:

  • A target NAV
  • A broker + broker account id
  • A list of strategies that can write to it
  • A risk envelope
  • Tax-lot accounting

mts1b-treasury allocates capital between funds.

What's a "strategy" in MTS1B?​

A configured-and-named factor application. momentum_v3 = factor f_momentum_12_1 + params {h_long: 252, h_skip: 21} + universe us-large-cap + sizing kelly_voltarget_12 + ... registered via mts1b-research.

Strategies are first-class objects in the registry; you can list, enable, disable, shadow.

What's a "factor" in MTS1B?​

A pure function that takes a UniversePanel and returns a (T, A) ranking. Convention: name starts with f_. Registered via @register("f_name"). See Concept — Factor system.

What's a "signal" in MTS1B?​

The output of a strategy at one rebalance — target weights per symbol + metadata (IC, t-stat, etc). Published to mts.v1.research.signals.published.

What's the difference between Order.idempotency_key and Order.order_id?​

  • order_id: UUID assigned by mts1b-oms when the order is created. Globally unique.
  • idempotency_key: caller-provided dedupe key. Multiple submissions with the same key within the dedupe window resolve to the same order_id.

The caller (strategy) generates idempotency_key. The OMS generates order_id.

How do I trace an order through the system?​

mts mts1b-platform tail --subject "mts.v1.>" --filter "order_id=ord-abc"

Will show every NATS event mentioning that order across every service. Plus check the audit chain:

mts mts1b-operations audit show --subject-id ord-abc

How do I roll back a release?​

mts1b-deploy rollback --to <commit_sha>

The deploy tool maintains a history of installed configs. Rollback restores the prior state.

Glossary of terms?​

Glossary page.