Advanced and Deep effort levels lack 'When' criteria

## The Issue

The effort level table in OBSERVE defines 8 tiers, but Advanced and Deep have no guidance on **when** to use them:

| Tier | Budget | When |
|------|--------|------|
| Instant | <10s | "right now", trivial lookup, greeting |
| Fast | <1min | "quickly", simple fix, skill invocation |
| Standard | <2min | Normal request, no time pressure stated |
| Extended | <8min | Still needed relatively fast, but quality must be extraordinary |
| **Advanced** | <16min | *(no "When" criteria — description same as Deep)* |
| **Deep** | <32min | *(no "When" criteria — description same as Advanced)* |
| Comprehensive | <120m | Don't feel rushed by time |
| Loop | Unbounded | External loop, PRD iteration |

Both Advanced and Deep have identical descriptions: "Full phases, checkpoints every 1 min" — making it impossible to distinguish when one should be selected over the other.

### Suggested Fix

Add "When" guidance. For example:
- **Advanced**: "Multi-system changes, cross-cutting concerns, significant refactors"
- **Deep**: "Architectural decisions, novel problem domains, research-heavy work"

### Location
- Lines ~40-41 in `versions/TheAlgorithm_Latest.md`