Algorithmic Trading and Market Intelligence Tools for an Uncertain Future

Our platform, NOVA, enables rapid development and validation of algorithmic trading strategies through the use of highly configurable and composable models and rigorous backtesting and simulation.

Explore NOVA Learn about Daylight
NOVA

Doomsun's systematic trading platform used to develop, test, deploy, and manage strategies.

  • Strategy composition

    Reusable alpha and risk models, stateless and composing through defined interfaces. Portfolio and execution behavior is declarative configuration, stored as data and versioned over time.

  • Immutable history

    An accumulate-only store. Every decision traces back to the data and model state that produced it, which is what makes a result reproducible a year later.

  • AI orchestration

    A structured surface for an agent to work against rather than a codebase to modify, so the platform absorbs advances in AI-assisted research as they arrive.

Components

  • Cyclotron

    Strategy research through reproducible experiments, reusable models, and retained evidence.

  • Stellarator

    Backtesting and simulation, with production execution behavior and traceable results for research assessment.

  • Daylight

    Characterizing what the current market structure is, how it is evolving, and what it implies for the near term.

The meta-strategy

Durable edge comes from the process that produces strategies, not from any one of them. Strategies emit probabilistic forecasts rather than buy/sell calls; those forecasts are scored for calibration, and capital follows the score. Individual strategies do not have to be regime-aware — only the scoring and allocation layer does.

  1. Candidate
  2. Active
  3. Attenuated
  4. Shadowed

Rotates back when the regime turns in its favor

Cyclotron

Strategy research through controlled experiments.

Cyclotron turns a market hypothesis into a reproducible experiment. Compose and parameterize reusable alpha and risk models, fit and calibrate them on separate data, and then assess a frozen candidate against criteria declared in advance. Cyclotron and Stellarator work together to run rigorous backtests and evaluate the results.

From hypothesis to evidence

Daylight / Input

Market hypothesis

A mechanism to test

Cyclotron

Compose & calibrate

Set the plan, fit models, and freeze parameters before testing.

Stellarator

CPCV backtests

Execute held-out splits. Return evidence.

Cyclotron

Meets
criteria?

Assess outcome paths

Yes / Meets criteria

Human review

Approve deployment to NOVA Core.

No / Does not qualify

Archive evidence

Changed hypothesis? Start a linked candidate.

Rigorous backtesting

Cyclotron sets the protocol for combinatorial purged cross-validation (CPCV): multiple held-out splits, with purging and embargoes to reduce leakage between training and evaluation data. Stellarator executes the backtests; Cyclotron assesses results across the resulting outcome paths.

Evidence that persists

Attempts are recorded before work begins. Failures stay in the trial history, and changes to frozen inputs or parameters create a linked successor. The record explains what was tried, not just what worked.

AI-assisted research

Cyclotron is designed for people and AI agents to work through scoped MCP tools. Separate access to data preparation, strategy design, and evaluation helps preserve each experiment’s integrity.

Combinatorial purged cross-validation

Stellarator

Execution realism, before a strategy touches capital. Stellarator runs backtests and simulations through the production execution path, returning the results Cyclotron uses for research assessment.

Execution realism

Strategies run in-situ, through the same path they take in production, with portfolio-level allocation and execution simulation covering slippage, partial fills, and market impact.

Research handoff

Cyclotron combines backtest results with trial history, verifies the evidence, and evaluates each candidate against the experiment’s criteria before deployment review.

Forward simulation completes the picture: synthetic data calibrated to real structure, preserving the volatility clustering and fat tails that geometric Brownian motion cannot produce.

Stellarator run observatory with performance, equity and drawdown, trades, and telemetry for PLUG, FCEL, CHPT, ENPH, SEDG, SMR, and BLDP. Stellarator parameter sweep with a 63-candidate heatmap, candidate metrics, equity comparison, rankings, and threshold sensitivity.
Stellarator workspace Run observatory and parameter landscape.

Daylight

“Is this strategy still working?” is a lagging question — by the time returns answer it, the regime has already turned. Daylight asks what the market is currently capable of.

Market structure analysis

Multifractal analysis measures persistence, regime transitions, and directional asymmetry across timescales. Kinematics treats that structure as a trajectory rather than a snapshot, making the rate and direction of structural change the signal — comparable tools report a point value, this reports where it is heading.

Predictors framework

An ML framework holding temporal context across price action, structural features, and cross-market signals. Structure feeds in as conditioning context, so one price pattern reads differently in a trending regime than in a chaotic one. Output is calibrated intervals, not point estimates.

Simulation

Synthetic market data calibrated to a real market’s measured structure, preserving the volatility clustering and fat tails standard Monte Carlo cannot produce. Stellarator stress-tests strategies against those paths rather than the single one that happened.

Daylight: the multifractal spectrum H(q) over time, beside the rolling Hurst exponent against the random-walk null Daylight: forecast paths from a calibrated multifractal random walk, with median and percentile bands

The people behind Doomsun

Decades of experience in distributed systems, financial infrastructure, and algorithmic trading.

Joe Smith

Joe Smith

CEO & Co-founder

20+ years building distributed systems and fintech infrastructure

Daemian Mack

Daemian Mack

CTO & Co-founder

20+ years in consulting and fintech systems architecture

Yaw Odame

Yaw Odame

Founding software engineer
Scott Lloyd

Scott Lloyd

Operations consultant

Backtesting and validation, compared

Graded August 2026 from official product documentation, API references, and public source.

Capability NOVA
Cyclotron + Stellarator		 Nautilus Deltix LEAN VectorBT
Execution realism
Production execution and accounting path in backtestbuilt inbuilt inbuilt inpartialnot applicable
Deterministic, reproducible runsbuilt inpartialnot supportednot supportednot supported
Double-entry ledger accountingbuilt innot supportedpartialpartialnot supported
Tick / order-book fill fidelitynot supportedbuilt inbuilt inpartialpartial
Multi-asset breadthpartialbuilt inbuilt inbuilt inpartial
Statistical rigor
Purged cross-validation with embargo / CPCVpartialnot supportednot supportednot supportedbuilt in
Deflated Sharpe ratio, trial-correctedbuilt innot supportednot supportednot supportedbuilt in
Persistent trial registry feeding the correctionbuilt innot supportedpartialpartialnot supported
Beyond the backtest
Immutable, replayable decision auditbuilt inbuilt inpartialpartialnot supported
Staged signal to fill attribution funnelbuilt inpartialnot supportedpartialpartial
Runtime forecast calibration scoringbuilt innot supportednot supportedpartialnot supported
Multifractal regime analysisbuilt innot supportednot supportedpartialpartial
Vectorized parameter sweepsnot supportednot supportedbuilt innot supportedbuilt in

● built in ◐ partial, or via add-on ○ not supported — no execution layer

Grades reflect what each vendor documents publicly. Some document selectively.

Several of these are general programming environments in which a user could implement the statistical methods independently. The line drawn here is between capability enforced by the platform and capability that is optional and user-assembled.