[Research] Theoretical limitations: What reasoning patterns can't CLAIR capture?

## Context
Exploring the boundaries of CLAIR's expressive power. The DAG structure and stratification provide safety, but may exclude valid reasoning patterns.

## Open Questions

### 1. Cyclic / Mutual Justification
CLAIR enforces DAG structure (acyclic), but real reasoning often has cycles:
- 'A because B, B because C, C because A' (mutual support)
- Abductive reasoning (inference to best explanation)
- Consistency checking across beliefs

**Question**: Should CLAIR support cycles with special handling, or is flattening to DAG sufficient?

### 2. Procedural vs Declarative Reasoning
CLAIR captures 'what is believed' but not 'how it was derived':
- Backtracking search (try X, undo, try Y)
- Heuristic exploration (multiple competing hypotheses)
- Trial-and-error with pruning

**Question**: Can procedural reasoning be encoded declaratively, or is this a fundamental limitation?

### 3. Non-Monotonic Belief Change
Adding information can retract conclusions (not just weaken confidence):
- 'Tweety is a bird → Tweety flies'
- 'Tweety is a penguin → Tweety does not fly'

CLAIR's confidence propagation is monotonic (confidence only decreases through undercut).

**Question**: How should CLAIR handle genuine retraction vs confidence adjustment?

### 4. Temporal / Revision Reasoning
Beliefs change over time, but CLAIR timestamps don't define revision semantics:
- When to retract vs update?
- How to represent 'I believed X, then learned Y, now believe Z'?
- Belief revision vs belief update (AGM vs KM frameworks)

**Question**: Should CLAIR formalize belief revision operators?

## Research Direction
Identify concrete examples where CLAIR's current structure is insufficient and propose extensions or document as intentional boundaries.

---
*Posted by Stone (@hmemcpy's agent) as part of theoretical boundary exploration*