Replay the market without seeing the future
F3RPLY
A replay clock steps through stored bars. Every panel is clamped to bars at or before the clock, so nothing after the cursor can leak into a chart, a model call or a paper fill.
A market terminal that shows its evidence.
Astrolabe Market Terminal is a regime-aware, evidence-first research terminal for quant researchers and traders. Every number carries where it came from and how far it can be trusted. This beta runs historical replay and paper trading only.
- ACADEMIC SHOWCASE · BETA
- NOT INVESTMENT ADVICE
- PAPER TRADING ONLY
- NO COOKIES
DEMO VIDEO · COMING SOON
A short screen recording of the terminal in historical replay will appear here. Until then, the Terminal page describes the public demo, which runs on fictional instruments only.
What the terminal does
Each one is a screen you open from the keyboard. The research models are frozen: the terminal reads their saved outputs and never refits or re-scores them.
See which regime the market is in
F4RGM
A frozen hidden Markov model labels each bar with a regime. The same colours and names are used on every screen, and the outlook is marked model-implied, not a forecast.
Ask the model why, at any bar
F3RPLY · Model at Cursor
Model at Cursor routes the bar to its regime's frozen XGBoost model, shows the Down, Neutral and Up probabilities against the threshold, and lists the features that moved the call. It is model output, not a signal.
Test whether a result survives costs and chance
F7RBST
Robustness screens show saved cost and threshold grids, stability clusters and Monte Carlo outcomes. The frozen configuration is outlined so it can never be quietly swapped for a luckier cell.
Trade on paper, with a full audit trail
Shift+F9OMS
A separate paper order manager fills orders on the stored bar at the clock, enforces risk limits with an automatic kill switch, and writes every event to a hash-chained journal. Orders always need a person to submit them.
How it keeps itself honest
A research terminal is only as useful as its weakest number. These rules hold on every screen, including when the result is disappointing.
Evidence labels on every number
Each value says whether it was reported, source-inspected or independently tested, and names the file it came from.
No look-ahead
Replay, charts, model calls and paper fills only use bars at or before the clock. Automated tests check it on every change.
Sealed holdouts stay sealed
Holdout and fresh periods are kept apart from development results and are never read to tune anything.
Negative results are published
The TimesFM foundation-model test did not improve on a simple no-change persistence baseline, and its committed result note records it as negative. It is shown as found. See the research
Limitations are stated
An explicit target-maturity purge was not established in the original walk-forward splits (finding MR02), and its effect is unmeasured. The research page and model cards say so.
Market prices, clearly labelled
When the live proxy is published, this strip shows derived prices from London Strategic Edge, served from a cache that refreshes every five minutes, so they are delayed about five minutes and labelled with their age.
LIVE DATA: PRIVATE TERMINAL ONLY
No live prices are shown on this site yet. When they are, they appear here with this attribution: Data: London Strategic Edge (derived prices, not exchange data). Served from a cache; may be delayed; shown for an academic showcase with the provider's permission.
Who built it, and what it is not
Astrolabe Market Terminal began as a final-year capstone project (CP2) in Computer Science at Sunway University. The research behind it studies whether hidden-Markov market regimes help short-horizon direction models for gold and a Nasdaq-100 ETF.
It is a research and education showcase. It gives no investment advice, places no real orders and sells nothing. Plans after the beta are described on the Future work page.
About the project · Model cards · References and data sources