AstrolabeMarket TerminalBETA

FUTURE WORK

Where the terminal goes after CP2

One codebase, four phases. Everything after the first phase is a plan, not a product: none of it is built, and each phase waits for a gate.

NOT BUILT

Nothing below the first phase is available today, and no date is promised.

Roadmap

Now · CP2

CP2 · the beta you see

IN PROGRESS

  • Historical replay, regimes, model explorer, robustness
  • Paper trading with risk limits and an audited journal
  • Rigour panels: purge disclosure, leakage checks, ablation view
  • Live LSE-derived prices on the private terminal
  • This public showcase site

After graduation

Product core

NOT BUILT

  • Bring your own data, strategies and models
  • Many instruments and more timeframes than M5
  • Monte Carlo runner on your own strategy
  • Guarded parameter optimiser with overfitting checks
  • Walk-forward retraining and a versioned model registry
  • Licensed market data in place of free sources
  • Multi-instrument screener

After product core

Enterprise

NOT BUILT

  • User accounts, single sign-on, roles and permissions
  • Server-side workspaces, alerts and settings
  • Full admin and operational audit log
  • Alerts by Telegram, email and mobile push
  • A mobile layout for the terminal
  • Broker-paper connection with partial fills
  • Portfolio optimiser and scenario risk

Licensing permitting

Execution

NOT BUILT

  • Supervised real-money execution through a regulated broker connection
  • Strategy-to-order automation with four-eyes approval
  • Order book and order-flow analytics
  • Advanced transaction-cost analysis (TCA)

The gates

Each later phase starts only when the gate before it is cleared. None is cleared yet.

Gates before each later phase
GateWhat must be trueStatus
IP confirmed, CP2 gradedUniversity IP and publishing rules confirmed; the capstone assessed.NOT YET CLEARED
Commercial data licencesPaid licences in place; free-tier data is not used commercially.NOT YET CLEARED
Broker partner and legal adviceA regulated broker partner and legal advice on licensing.NOT YET CLEARED

Research follow-ups

Ideas recorded in the project backlog. Each needs its own protocol and owner approval; none is scheduled.

  • A fresh out-of-sample test on bars after the historical fresh period, scoring the frozen models unchanged
  • Overfitting statistics at scale (probability of backtest overfitting; deflated Sharpe only where admissible)
  • A seed-noise study that measures run-to-run variation without replacing the frozen models
  • A ladder of simple forecasting baselines beside the TimesFM result
  • Component ablation: remove regime routing, multi-timeframe features or single timeframes one at a time
  • A one-command reproducibility container that rebuilds key tables from the frozen artefacts
  • Feature-interaction attributions for the frozen XGBoost models

The research limitations (for example the target-maturity purge, MR02) are listed on the Research page. Fixing them is research work, decided separately from this roadmap.