[Research] What was abandoned: From programming language to trace format

### 3. Explicit Language Syntax
`archive/old-syntax.md`, `archive/hello-world.clair` show actual programming syntax:
```clair
-- Old CLAIR was a language you wrote in
let sorted = belief {
value = quicksort xs,
confidence = 0.91,
justification = complexity_proof(O(n log n))
}
```

## What Changed

**Then**: CLAIR was a programming language with epistemic types
**Now**: CLAIR is a data format for reasoning traces

**Then**: You programmed in CLAIR
**Now**: LLMs emit CLAIR; humans/agents consume it

## Research Questions

1. **Why the pivot?** What made the language approach untenable?
2. **What was lost?** Can the type-theoretic foundations inform v2?
3. **Hybrid approach?** Statically-typed belief composition + dynamic traces?
4. **Executable CLAIR?** Can traces be "run" or only audited?

## Hypothesis: The Two-Layer Model

The old approach conflated:
- **Layer 1**: Computational language (functions, types, evaluation)
- **Layer 2**: Epistemic tracking (confidence, provenance, justification)

CLAIR v1 kept only Layer 2—the trace of reasoning, not the reasoning itself.

**Question**: Is this separation permanent, or should they reunite?

## Archaeological Value

The archive contains sophisticated mathematics:
- Graded monad laws with proofs
- Categorical semantics
- Type-theoretic foundations
- Turing-completeness arguments

Even if the language approach was abandoned, the formalism may be:- Reusable for INTENT (negotiation protocol types)
- Implementable in dependent type systems (Lean, Idris)
- Valuable for proving properties of CLAIR traces

## Action Items

1. Document why the pivot happened (interview/notes?)
2. Extract reusable formalism from archive
3. Determine if any language features should be resurrected
4. Clarify CLAIR's identity: language, format, or both?