ABOUT
A capstone project that grew into a terminal, built to show its evidence.
- CP2 CAPSTONE · BSC (HONS) COMPUTER SCIENCE
- SUNWAY UNIVERSITY, MALAYSIA
- NON-COMMERCIAL BETA
The project
Astrolabe Market Terminal began as the CP2 capstone of a BSc (Hons) Computer Science degree at Sunway University. The research asks whether recognising market regimes improves short-horizon direction models for gold and a Nasdaq-100 ETF; the terminal is how that research is inspected, replayed and stress-tested.
It is built by Yong Yung Jian (Ashton). After the capstone the plan is to keep developing the same codebase; the Future work page shows what that would take.
Principles
Every number says where it came from
Panels carry a source, an as-of time and an evidence label: reported, source-inspected or independently tested.
No look-ahead
Replay hides every bar after the cursor; the regime decoding caveat is measured and published (MR-01).
Sealed data stays sealed
Holdout periods are never read by the drift and robustness tools; frozen models are read, never retrained.
Negative results are published as found
The TimesFM test did not improve on persistence, and the site says so. The MR02 purge limitation is listed, not hidden.
Paper only
There is no route to a real order. The paper order system fills on stored bars and keeps a hash-chained journal.
How it was built
The work runs in four threads, each in its own lane, with every change reviewed before it is merged.
| Thread | Area | What it covers |
|---|---|---|
| A | Research | Methods, frozen protocols, the research runs and the report. |
| B | Engineering | The terminal: front end, back end, the paper order system, the live data services and this site. |
| C | Design | Interface design and the fictional showcase prototype, kept separate from real data. |
| D | Tooling | Continuous integration, checks that block secrets and large files, and the build pipeline. |
Use of AI tools
Much of the code and documentation was drafted with AI coding assistants (OpenAI Codex and Anthropic Claude) working on tasks the owner scoped. Each change goes through independent reviews and automated checks, and the owner approves every merge that touches research, data or decisions. Research methods, targets and interpretations are the owner's, not the assistants'.
Contact
Contact links (GitHub and LinkedIn) are added by the owner before launch. There is no contact form on this site, so no personal data is collected.
Acknowledgements
Thanks to the project supervisor at Sunway University for guidance, and to London Strategic Edge for permission to show derived prices in this non-commercial showcase. Open data comes from BLS, BEA, the Federal Reserve, FRED, the World Bank and the IMF, credited on the References page.
Legal: Disclaimer · Terms · Privacy