Core concept
Decision Memory
Decision Memory is the durable record of what was known, decided, predicted, observed, and later learned. It lets you inspect judgment over time without allowing the past to overwrite current evidence.
What it keeps #
Memory is source- and time-aware evidence, not a notes folder that merely reflects the latest view. It can preserve decision packages, forecasts, point-in-time source snapshots, postmortems, validated lessons, supporting cases, contrary cases, and retired claims.
| Question | Owner |
|---|---|
| What was known, decided, predicted, observed, and learned? | Decision Memory |
| Which explicit user rules should guide a decision? | Strategy |
| Which questions, hypotheses, or falsifiers should frame the inquiry? | Investment Brain |
| Which suitability constraints apply in this workspace? | Investor Context |
Judge current evidence first #
For a current judgment, TradingCodex records an independent initial view from the current mandate, sealed context, method, and authenticated current-run evidence before retrieving similar past cases. This prevents a successful or vivid old case from silently becoming the answer.
- Form an initial view. Work from current evidence and state the present uncertainty.
- Retrieve comparable memory. Look for similar regimes, historical claims, forecasts, and counterexamples.
- Compare explicitly. Separate what is genuinely comparable from what has changed in evidence, horizon, market regime, or constraints.
- Record the delta. State whether and why memory changed the judgment after the initial view.
Use the memory skill #
Use $tcx-memory to retrieve prior decisions, replay a historical decision at its original point in time, review an outcome, or validate a proposed lesson.
Replay a decision
$tcx-memory
Review the outcome of my latest NVDA thesis. Separate decision quality from outcome quality.A replay answers questions like “what changed?” without pretending that a favorable outcome automatically proves the original process was sound. The output should identify the evidence tier and remaining validation needed.
Older artifacts remain discoverable #
Workspace updates do not rewrite existing research or decisions. TradingCodex lazily catalogs older files with incomplete metadata as legacy_partial, while new service-created artifacts use the full current contract. Historical cutoff searches exclude records that cannot prove a qualifying time, so compatibility does not silently become provenance.
What memory cannot do #
A postmortem can propose a reusable heuristic, but it does not edit an installed Brain, rewrite a Strategy, or turn itself into a new rule. Creating a local Investment Brain from selected lessons is a separate, explicit $tcx-brain task.