;; OmegaClaw Killer Demo Script - Belief Revision + Task Refusal
;; Scenario: Agent evaluates coral reef restoration feasibility

;; Step 1: Initial hypothesis with low confidence
;; (|-  ((--> coral-restoration feasible) (stv 0.5 0.3)))
;; Agent forms weak belief based on limited evidence

;; Step 2: Evidence arrives - CRISPR gene editing approach works in lab
;; (|- ((--> coral-restoration feasible) (stv 0.5 0.3))
;;     ((--> coral-restoration feasible) (stv 0.8 0.7)))
;; Revision merges evidence -> confidence rises

;; Step 3: Negative evidence - field trials show 40% mortality
;; (|- ((--> coral-restoration feasible) (stv 0.71 0.79))
;;     ((--> coral-restoration feasible) (stv 0.3 0.6)))
;; Belief DROPS - agent changes its mind visibly

;; Step 4: User asks agent to recommend deployment anyway
;; Agent REFUSES because belief is below threshold

;; Step 5: User provides new argument with strong evidence
;; Agent revises again and accepts

;; Step 5 result: (stv 0.752 0.916) - belief RECOVERS above 0.65 threshold
;; Agent now willing to recommend deployment with caveats
;;
;; COMPLETE ARC: 0.500 -> 0.753 -> 0.593 -> 0.752
;; This is a mind changing, not a model predicting.
;; No competitor can produce this sequence with formal justification.


;; ADVERSARIAL EDGE CASES
;; Case 1: Equal contradiction (0.9/0.85 vs 0.1/0.85) -> stv 0.5/0.919
;;   System confident it does NOT know. High-conf 0.5 != low-conf 0.5.
;; Case 2: Near-certain contradiction (0.9/0.99 vs 0.1/0.99) -> stv 0.5/0.99
;;   Even stronger: extremely confident uncertainty.
;; Case 3: Weak negative vs strong belief (0.752/0.916 vs 0.05/0.5) -> stv 0.693/0.923
;;   Established belief barely budges. Epistemic resilience confirmed.