22  Experiments

The running record of research results and how to reproduce them. This is distinct from the platform’s polished findings write-ups — publication-style accounts of results that have already stabilised — whereas experiments here are the raw material those get built from, including ones that did not pan out (see journal/dead-ends.md).

Status: one experiment is logged under this convention so far (below). Older results still live among the polished findings until they are migrated or superseded.

22.1 Logged experiments

• 2026-08-03 — empirical results for the canonical factor catalogue — the provenance record (run identity, configuration lookups, and method) behind the empirical fields in the per-factor, per-family catalogue. Two significance fields — a naive and a heteroskedasticity-and-autocorrelation-consistent (Newey–West) t-statistic — are specified but not yet computed.

22.2 Convention for a new experiment entry

Each experiment gets a dated folder, experiments/YYYY-MM-DD-short-name/, containing:

• README.md — what question (see Research Questions) it tests, what was run, and the result.
• the configuration and parameters used, or a pointer to the run identity in the platform if the run itself lives there — see Reproducibility & experiment infrastructure for what a run identity captures.
• a link to the journal entry that motivated it, and the entry that recorded the outcome.

22.3 Existing results (published-quality)

• Asset-dependency networks: an econophysics review
• EDGAR point-in-time integrity

22.4 Reproducibility

Every experiment inherits the platform’s point-in-time and run-identity guarantees — see Reproducibility & experiment infrastructure for what “auditable and repeatable” means concretely here.