Market Regime Lab
A daily research workspace that preserves sources, versions, and data gaps from raw observations through metrics, snapshots, and experiments.
The problem
After reviewing a day of market indicators, I wanted to be able to return to the data behind a judgment: where did the number come from, what was available then, and would different parameters change the result? Market Regime Lab keeps raw observations, derived metrics, and hypotheses separate so that a historical interpretation can be reproduced and gaps remain visible.
My contribution
I specified Raw / Canonical / Feature separation, freshness rules, pluggable metrics, and admission rules for manual evidence. I reviewed the roadmap and UI and participated in Ubuntu VPS deployment and acceptance. AI assisted implementation, migrations, and testing; I drove staged reviews and defect feedback.
Engineering results
- Release A delivered PostgreSQL storage, idempotent refresh jobs, immutable historical snapshots, and a three-language dashboard. I confirmed acceptance on 27 September 2026.
- Release B added versioned strategies, experiments with frozen datasets, candidate evaluation, and relationship hypotheses with counterexamples.
- Its isolated validation covered 26 mechanisms, 354 Python tests including 85 real PostgreSQL tests, two browser engines across three languages, non-empty migrations, and backup recovery.
- The 9 October record confirms VPS acceptance activity and merging into core. It does not fill in unreported individual checks as passed.
Design choices
Changing a research parameter should not silently change the active version. Experiments therefore keep their own parameters, and candidates require review before promotion. When data is insufficient, partial or not_evaluable shows which judgments cannot yet be made. Manual evidence also stays separate from automatically collected facts.
Current limits
The completed work concerns the data and research process. Saving, replaying, and comparing a judgment does not establish profitability, sufficient historical coverage, or promotion of a real candidate. Release C remains ongoing. Source and research data stay in a private repository; this page describes the project.