A dossier that recommends a paper-trading trial for Model M is worthless to a second researcher if the weight vector that produced the claimed Sharpe cannot be regenerated. This note decides whether M's research package is rerunnable by someone who did not write it, and forces a concrete run record — code revision, data snapshot id, environment pin, seed and command — without reteaching virtual environments or notebook mechanics already covered in reproducibility.
Learning objectives. By the end of this note you should be able to:
- decide pass or fail for a dossier's replay claim from its run record alone;
- populate a minimal run record for M's weight-generation job;
- state why a fixed seed does not freeze vendor data;
- separate a reproducible claim from a reproducible environment.
In this lesson
- The decision: is the dossier rerunnable?
- Artifact: the run record for M
- Limitation: a seed does not freeze vendor data
- Replay protocol before validation is booked
- What reproducibility is not
- Desk checkpoint: refuse unverifiable claims
Before you start
The full lesson, with worked examples and exercise solutions, is for members.
Read the full lessonNew to SelfMadeQuant.com? See membership options or browse the Professional Quant Practice syllabus.