Read Me First — Research Library

The Research Library is the evidence archive for experiments you choose to retain. Its purpose is to preserve the complete research record—parameters, benchmark, provenance, and results—rather than only screenshots of favorable outcomes.

What this lab is designed to do

The Research Library is the evidence archive for experiments you choose to retain. Its purpose is to preserve the complete research record—parameters, benchmark, provenance, and results—rather than only screenshots of favorable outcomes.

Key controls and inputs

• Archived records. Each record should correspond to a completed research run or a deliberately saved protocol/result.
• Strategy and parameters. Verify that the exact delta, DTE, regime, volatility assumptions, seed, and other controls are stored.
• Benchmark / control. A result without its comparison benchmark can be misleading.
• Simulated paths / sample size. This helps judge the precision and scale of the experiment.
• Archive status and timestamps. Useful for distinguishing current work from superseded or exploratory runs.
• Export library JSON. Exporting the structured record is useful for external analysis, replication, and manuscript/article work.

What the outputs mean

• Research record. Read the hypothesis and setup before the performance numbers.
• Provenance. Engine version, seed, run metadata, and strategy parameters establish what was actually executed.
• Performance distribution. Prefer the complete metric set over a single best statistic.
• Negative and null results. These are valuable evidence and reduce selective-reporting bias.
• Comparability. Two records should only be compared directly if their benchmarks, sample sizes, and assumptions are compatible.

A good first experiment

1. Open one saved study and verify its strategy, benchmark, seed, regime, path count, and date.
2. Read the hypothesis or protocol before looking at the outcome.
3. Identify whether the record is exploratory, validation/holdout, or a final retained experiment.
4. Compare it with a nearby experiment where only one parameter changed.
5. Export the library JSON periodically so the research archive exists outside the browser/session.
6. Keep informative failures and neutral tests; do not curate the library into a highlight reel.

How to interpret the result

Do not judge the strategy from one path, one seed, or one favorable market environment. Read return, drawdown, exposure, trade frequency, and benchmark-relative performance together. A result is more credible when it persists across reasonable parameter changes and when the comparison benchmark has similar economic exposure.

Important assumptions and limitations

• The library is only as trustworthy as the provenance saved with each run.
• Changing engine versions can make old and new studies non-comparable unless version differences are documented.
• Missing or selectively deleted experiments can create survivorship and publication bias.
• Exported records should be backed up if they are being used as evidence for articles, books, or public claims.
• A saved simulation is still model-based evidence; its archive status does not convert it into observed market performance.

How this complements backtesting

Use the Library to connect simulation with backtesting over time. Store both types of evidence, label them clearly, and avoid blending them. A simulator result and a historical backtest answer related but different questions; the archive should preserve that distinction.