From market knowledge to tested, running systems.
NOVAQ Quant Center is an independent ecosystem for quantitative futures research. It connects a 145,000-note knowledge base, 16 years of annotated market data, statistical research, AI agents and an execution platform in one working system, built in under six months.
What NOVAQ has built
Six connected layers
Knowledge
Turns books, courses, papers and videos into attributed, searchable claims
About 145,500 atomic notes from 384 author profiles, with 239,000 searchable text passages
Data
Provides clean, contract-accurate market history and live feeds
NQ and ES minute data from 2010, plus session analytics across more than 400 columns
Research
Tests ideas with realistic costs and strict controls
About 700 catalogued studies and tests, and 161 trading legs
Platform
Runs the whole chain as supervised services
9 service groups, 42 web pages, about 72,000 lines of platform code
Trading
Executes systematic strategies under layered risk control
Strategy book in simulated execution, with native strategies in the trading platform
AI
Analyses markets, runs trading experiments and audits the system
Live market briefs, a shadow expert, an AI trader in simulation and an eight-role audit swarm
Why the connection matters
Most trading projects have one of these parts. NOVAQ's value comes from the links between them.
- Knowledge feeds research. A claim in a book can become a pre-registered test. In one round, 15 setups from 7 authors and groups went straight from the knowledge base into testing.
- Research reaches execution. A strategy keeps the same logic from backtest to live engine, so results can be compared trade by trade.
- AI works from verified inputs. Analysis agents read fields that have been recomputed from raw data, and an AI expert reads knowledge distilled from the vault under a tracked fingerprint.
- The system checks itself. Independent AI auditors review research and running code, so a result can be traced from claim to fill.
Selected results
Research
A diversified rule-based portfolio (historical test, 2020 to Oct 2026).
53 strategy books, equal weight, after costs: Sharpe about 2.15, return to drawdown about 23, and all 7 calendar years positive (2026 partial, to 1 October). A historical result of selected strategies, not a track record. See the Evidence page
About 12 effective exposures.
Two separate methods, in June and October, both found roughly 12 distinct portfolio exposures inside a much larger book. This measures diversification and is not a count of proven edges.
Robust under a point-in-time review (Oct 2026).
A review of the inputs of 58 rule-based books left 44 unchanged and reduced total result by about 8%.
The size of the day can be forecast (walk-forward, 2020 to Oct 2026).
A pre-open model beat a rolling base rate on session size in every year tested, with an AUC near 0.7 on NQ and ES, where 0.5 is chance.
Forward-to-backtest check (Oct 2026, mostly simulation).
A technical check on 2,272 paired trades found no robust gap from the backtest in any strategy family. It tests tracking, not profit or full agreement: the interval is wide, and more than half of forward trades could not be paired.
Technology
Exact reproduction.
87,260 backtest trades rebuilt on pinned data with zero value mismatches.
Clean market data.
Zero mixed-contract bars among about 11 million live minute bars, in an audit of 341,992 files.
Faster internal cycle, same decisions.
The platform's median internal cycle fell from 16.5 to 2.77 seconds outside the heaviest work windows, with each step checked for identical outputs. This is system processing time, not order execution at the broker.
A running migration.
A live engine moved to a new architecture with 52 of 52 strategy providers, the modules that generate signals, passing fidelity checks.
Scale
- Atomic knowledge notes
- about145,500
- Concepts in the knowledge graph
- about2,200
- Author profiles
- 384
- Searchable passages
- about239,000
- Annotated session history
- NQ and ES, 2010–2026
- Trading legs (3 Nov 2026)
- 161
- Catalogued studies and tests (3 Nov 2026)
- about700
- Tested variants and configurations
- over3 million, including reruns
- Pre-registered studies
- 106
- Automated test functions
- about4,900
- Numbered findings from the audit swarm
- about100
Development in brief
- May 2026Knowledge base and quantitative laboratory founded
- June 2026First strategy book, live data bridge and risk layer
- July 2026Migration to NovaQ, a federation of services
- August 2026First AI expert reading distilled vault knowledge, and a clean continuous data corpus
- September 2026Eight-role audit swarm and native platform strategies
- October 2026Platform speed-up, full data audit and a 42,000-atom expansion of the knowledge base
Collaboration
Part of NOVAQ's research is carried out jointly with Vašek (BTE). Vašek brings studies on 16 years of NQ and ES data with pre-registered tests, an out-of-sample period and correction for multiple testing. NOVAQ reproduces them independently on its own data and checks whether a result can become a tradable strategy after costs. Joint studies are published in a generalised form.