Overview
CryptoStrategyLab investigates how trading rules behave once timing, fees, spreads, slippage, gaps, and small-portfolio constraints are explicit. The deliverable is a reproducible research and simulation system, not a trading service or a claim of profitable performance.
The project is local. Historical experiments, bounded market-data observations, synthetic fixtures, and dashboard verification are separate evidence categories.
Problem
A backtest can look convincing while assuming unavailable prices, ignoring execution costs, or allowing future information into a decision. A dashboard can create a different problem: a recently fetched file may describe a stopped or stale process.
The system records when information becomes available, what each simulated decision did, and how much evidence supports the displayed state.
What I Built
- A frozen historical protocol with four fixed baselines: trend following, mean reversion, breakout, and momentum, compared with cash and buy-and-hold.
- Causal completed-bar signals, explicit next-opening execution proxies, received-asset fee accounting, gap handling, and reconciled event ledgers.
- Deterministic run identities and checksums, with development experiments kept separate from protected validation and holdout evaluation.
- A separately versioned forward-observation simulator with persisted state and quote freshness checks, without broker order submission.
- Paired simulated capital levels of $50 and $1,000 per portfolio, using one shared input stream but independent accounting rather than scaled results.
- A read-only React/TypeScript dashboard fed by a validated snapshot projection, separate from credentials, raw data, and the simulator database.
Results and Evidence
The fixed development tournament produced reproducible comparisons across four strategies, two benchmarks, and three cost scenarios. Its recorded base scenario exposed substantial losses for the active strategies. This is development evidence about the chosen rules and assumptions, not evidence of an investment edge.
A separate October 10, 2026 bounded market-data smoke exercised twelve simulated portfolios and one controlled restart. It recorded one simulated buy in the $50 buy-and-hold portfolio, no active-strategy fills, and no closed round trips. The run ended stopped. That brief observation supports an accounting and restart check, not a performance claim or an ongoing experiment.
The dashboard is implemented locally. A November 1 freshness correction passed seven in-process helper and scheduler tests plus static checks; mounted browser behavior has not yet been reverified for that change. Earlier browser evidence remains historical.
What I Learned
Execution assumptions and accounting constraints can dominate an apparently simple strategy. Preserving negative results, rejected actions, and stale-state labels is more useful than smoothing those details into an attractive equity curve.
Limitations and Next Steps
There is no live trading, official continuous experiment, public deployment, or demonstrated profitability. Candle-based historical execution is a proxy, not proof of executable fills. Synthetic tests and a short real-data smoke answer different questions.
Remaining work includes browser acceptance of the freshness change and completion of the operational provenance, trusted inventory, and Linux acceptance gates. A historical operational candidate correctly fails its content-hash check; it is not deployment approval. Further research or deployment requires its own decision rather than following automatically from a working dashboard.