stratevaReproduced history
We reconstruct decisions using frozen data where a compatible route exists. Imported returns are labelled separately. History does not establish future performance.
Methodology
What we can establish. And what we cannot.
A list of metrics is not enough. Conclusions must match the available material, the tests performed and their limitations.
Walk-forward OOS
Rolling or expanding when the causal process can be repeated and the sample is sufficient. Nested WFA when parameters are selected: folds and tests are fixed in advance. Splitting a curve after the fact is not enough.
Prospective holdout
Requires future data, a sealed protocol and actual elapsed time. It is not delivered complete in seven days. Follow-up is scoped separately, without 24/7 monitoring.
Data and causality
We look for temporal leakage, execution misalignment, gaps, duplicates and inconsistent costs. A scanner cannot prove absence of look-ahead. Without point-in-time membership and delistings, survivorship bias is not treated as resolved.
Robustness and selection
Block Monte Carlo and return-order permutation where appropriate; neither replaces a gate-timing test. DSR/PBO require sufficient variant history and aligned data. They do not prove that every trial has been disclosed.
Traceability and responsibility
Versions, configuration and hashes bind the examined material. Hashes do not establish source truth. AI may assist preparation and review; official metrics come from the closed engine and deterministic criteria govern its result. Documentary reviews receive no engine verdict.
A deliberately bounded remit.
Starting point
Simple strategies on cash equities, spot crypto and spot FX. Compatible daily data; one instrument by default for base validation.
Only with a specific scope
Portfolios, intraday data, new adapters, complex datasets and research/shadow controls. Short selling is not accepted without checking market, contract and runner.
Out of scope
Derivatives, margin, leveraged tokens, live execution, custody of secrets and regulatory validation. We do not certify strategies or guarantee removal of every bias.
ILLUSTRATIVE EXAMPLE · NOT A CLIENT RESULT
What a finding looks like.
P1
The signal uses the close before it can execute.
- Severity
- P1 · affects the validity of the reproduction.
- Evidence
- This example decides using the close of t and credits that position with the return starting at the same close.
- Reproduction
- Separate signal and fill timestamps; compare with next-open execution and agreed costs.
- Limitation
- This example illustrates report structure; it does not evidence an executed test or quantify the impact.
- Follow-up
- Verify the change with a timing test and retain the version used for the retest.
03 / LET’S TALK ABOUT YOUR STRATEGY
Start with a question, not a payment.
Tell us what you want checked, the market, instruments, frequency and the code or data you have. We will explain whether it fits and what the scope would be.
Check compatibilitystan@strateva.aiNo private attachments in the first message. No obligation to commission work.