METHODOLOGY
Research first. Measure everything.
Internal research priority is not a public conviction score or investment recommendation.
ELIGIBILITY
What enters research
verified transaction code P; acquired/disposed indicator A (not transaction code A)
positive shares and reported price
primary filing URL and required filing/owner fields
common/ordinary stock before public packaging
not affirmatively flagged as a Rule 10b5-1 transaction when that field is available
filing lag of four calendar days or less for normal research priority; older filings require manual exception review
CONTEXT ONLY
What does not get a free boost
multi-insider cluster: tracked but receives no automatic priority boost
repeat buyer: tracked but receives no automatic priority boost
very large dollar amount: never treated as automatically stronger
EMPIRICAL NOTES
What the backtest changed
The strongest broad association is shortly after the filing; multi-month benchmark-relative outcomes are less stable.
Five-plus-day filing lag was materially weaker in the 2026Q2 holdout and is penalized/excluded from ordinary priority.
Dollar size was non-monotonic. Purchases of $1M+ did not justify the old automatic top weighting.
Cluster status alone did not show a stable advantage over solo purchases, so the v1 cluster-conviction assumption is removed.
Moderate drawdowns/recent declines were more informative than simply being far below a high; extreme distress remains a risk flag.
Role effects are retained as context. CFO results were stronger in the holdout, while CEO results were not consistently superior enough to justify the old large role premium.