Memory & Decision Ledger

Your company stops forgetting.

Companies pay for the same lesson annually: the churned segment rediscovered, the failed channel re-tried, the pricing mistake with new fonts. Institutional memory lives in whoever hasn't quit yet.

Mechanism

Four memories.

Semantic

What's true — the state.

Episodic

What happened — runs, events.

Decision

What was chosen and why — the ledger.

Procedural

What works here — plays learned from verified outcomes.

Anatomy

A DecisionLedgerEntry, sealed.

state snapshot: Aster, week 2 — pre-launch state
options: Services-led · Design-partner program · Paid campaign
prediction: 8–14 qualified conversations (model v1.4)
decision + rationale + decider: Focused design-partner program — founder + Cortex, jointly
missions: 3 missions compiled
outcome: 11 qualified conversations
variance: within predicted range
learning: confidence in v1.4's mid-budget band increases

Worked example

Aster's D-104 closes.

Predicted 8–14, actual 11; the agency signal writes a new procedural memory (“agencies over-index on willingness to pay”) and the next recommendation changes.

Control

Corrections supersede, never overwrite.

Predictions preserved verbatim · memory respects sensitivity and tenant walls · corrections supersede, never overwrite.

UnderstandDecideDeliverVerifyLearn

Outcomes get sealed here, and the next recommendation gets sharper.

Hard question

Does my data train your model?

Your workspace’s state and outcomes improve YOUR company’s recommendations by default. Contribution to shared model improvement is explicit, consented, and de-identified — and the exact policy lives in the Trust Center, not in a marketing page’s fine print.

Explore Cortex →Why this compounds →