Research notes
Why we kill strategies.
Roughly five concepts get killed for every one that reaches the storefront. The kills are not failures; they are the working product of the pipeline. The same gates that kill weak ideas are the gates that you, the buyer, can audit on every shipped strategy via the robustness section.
Below are the categories that account for most of our graveyard. The actual graveyard (GRAVEYARD.md in the working repo) has the dated per-concept post-mortems.
Failed the lifetime gate
Concept tested clean but lifetime ROR / Sharpe / MAR fell below the auto-cull floor. Auto-killed by the pipeline; no manual approval needed.
Drawdown concentration in 2008
More than 75% of lifetime drawdown was concentrated in the 2008 GFC window. The pre-crash equity curve and the recovery curve flattered the average; the strategy did not survive crisis conditions intact.
Out-of-sample degradation
Strategy performed well in-sample (1994-2018) but Sharpe collapsed in the held-out 2019-2023 window. Almost always a curve-fit; almost always discovered only by running the OOS overlay.
Edge concentration in too few trades
Clipping the top 1% of trades killed the equity curve. The 'broad / moderate / narrow' Edge Concentration field on every product page is exactly this check.
Spike on the parameter grid
Heatmap showed a single-cell spike rather than a plateau. The production cell was lucky; the parameter neighborhood would have lost money.
Data-scope violation
Used a symbol or universe outside the approved Norgate data scope (e.g. a free-tier symbol that wouldn't reproduce). Killed even if numbers looked good.
Look-ahead bias detected
Signal computed using same-day close while exit fires same-day at close. Easy to miss; obvious in trades.csv once flagged.
Survivorship-leaky universe
Tested against current Nasdaq-100 instead of point-in-time membership. Delisted survivors-bias killed the credibility before any other gate.
Cohort correlation too high
Concept's returns correlated > 0.85 with an existing HoF holding. Adding it would have been a position increase, not a diversifier.
Already in the graveyard, just renamed
Before proposing a concept, we search GRAVEYARD.md. Re-running a previously-killed idea under a new name is the cheapest mistake in research; the index exists to prevent it.