The Lucid Decision System
This is the running falsifiability core of myKungFu — a decision loop, not a concept. Every decision is recorded so that its reasoning can later be proven right or wrong.
Falsifiability is a hard constraint: the system refuses to record claims that no outcome could ever contradict. It implements the cognitive models defined at thelucidmind.ai — referenced here, not redefined.
Decision
A choice is recorded with its interpretations (each carrying falsification criteria) and assumptions (each carrying a confidence score).
Outcome
What actually happened is recorded against the decision — the real-world result the interpretations and assumptions are tested against.
DQS
The Decision Quality Score scores the decision after its outcome is known, feeding the learning engine and Capability Pressure.
All epistemic content lives in one epistemic_objects table with a type discriminator and a content JSONB field — new kinds of record need no schema change.
Every interpretation needs falsification criteria. Every assumption needs a confidence score that will be tested.
An interpretation that no outcome could contradict is not knowledge — it is decoration. The Decision System will not add a field whose value can never be wrong.
This is what makes a decision learnable: when the outcome arrives, each interpretation and assumption is checked, and the result moves confidence up or down rather than leaving it asserted.
Assumption history
Every assumption’s confidence is tracked over time and tested against outcomes — calibration, not assertion.
Perspective scoring
The interpretive stances that fed a decision are scored by how well they held up against the outcome.
Capability Pressure
Validated/failed deltas adjust capability confidence under spec-defined thresholds — the signal that drives learning.
The DQS formula and Capability Pressure thresholds are spec-defined; this page references them rather than restating the math.
The full decision → outcome → DQS loop runs today via the lucid CLI and the FastAPI backend (Phases 0–2 complete: schema foundation, decision loop, learning engine).