The memory layer is your dialogue with the model
Markets change. What predicted outcomes in Q1 may be noise by Q3. The memory layer lets you correct the model's understanding, add your own knowledge, and flag trades as meaningful or anomalous. Everything you do here directly affects how the model weights future decisions. Approved memos feed into weight updates. Annotations change training weights. Knowledge entries inform scoring context.
Collecting data — model will adapt after 50+ trades.
Free-form knowledge entries about how specific market types behave. Each entry has a confidence score that decays over time without reinforcement.
Examples: "Political markets show elevated velocity noise 30 days before elections" · "Crypto markets: divergence with BTC implied probability is reliable when BTC vol is low"
Every trade can be annotated to tell the model how to treat it during weight updates.