PERMANENT RESEARCH ARCHIVE

Failed versions.
Nothing deleted.

Retiring an experiment does not erase it. This archive keeps discontinued PAPER systems visible, records why they were stopped, and links each failure to what changed next.

ARCHIVE POLICYPERMANENTEVIDENCEPAPERCHERRY-PICKINGNO
What “failed” means hereA version appears here because its experiment was formally discontinued—not simply because it had losing trades. Retirement reasons are recorded separately from performance.
EXPERIMENT GRAVEYARD

Retired systems and what they taught us.

V1–V3RETIRED

Orderflow / Imbalance Research

Did not demonstrate a durable PAPER edge.

WHY RETIRED
Short-horizon orderflow and imbalance logic produced losses and unstable behavior. More filtering risked fitting the strategy to its mistakes.
WHAT WE LEARNED
Plausible microstructure logic is not evidence. Net forward results matter more than an attractive market narrative.
WHAT FOLLOWED
V4 control branch
Open preserved research →
V5.1RETIRED

Adaptive ML Shadow

Mixed PAPER results across experimental policies.

WHY RETIRED
Reward design and policy selection did not adapt reliably after losses. Conflicting LONG/SHORT exposure exposed weaknesses in the learning architecture.
WHAT WE LEARNED
A learning system needs explicit reward accounting, exposure constraints and a frozen comparison baseline.
WHAT FOLLOWED
V6 / FreqAI
Open preserved research →
THE ARCHIVE RULE

A replacement does not rewrite history.

Current systems remain in the live experiments dashboard. Once a branch is formally retired, its record moves here and stays accessible.

Current experimentsFull version history →