File size: 1,186 Bytes
9c91a7e
f0afaa5
 
 
9c91a7e
f0afaa5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
---
tags:
- classify-and-augment
- llm-annotated
---

# davanstrien/otb-cascade-phaseB-2tier

LLM-annotated dataset produced by [classify-and-augment](https://github.com/davanstrien/classify-and-augment).

## Configuration

- **Model**: Cascade: `LiquidAI/LFM2.5-1.2B-Instruct``Qwen/Qwen3-4B-Instruct-2507`
- **Labels**: `jim_crow`, `no_jim_crow`
- **Input rows**: 200
- **Output rows**: 200

## Label distribution

| Label | Real | Synthetic | Total |
|---|---:|---:|---:|
| `jim_crow` | 14 | 0 | 14 |
| `no_jim_crow` | 186 | 0 | 186 |

## Cascade audit

| Tier | Model | Processed | Entropy | Parse fail % | `jim_crow` count | `no_jim_crow` count |
|---|---|---|---|---|---|---|
| 1 | `LiquidAI/LFM2.5-1.2B-Instruct` | 200 | 0.515 | 0.0 | 23 | 177 |
| 2 | `Qwen/Qwen3-4B-Instruct-2507` | 200 | 0.366 | 0.0 | 14 | 186 |

**Pairwise tier agreement:**

- Tier 1 vs Tier 2: 82.5% (165/200 examples)

**Escalation:** 35 examples (17.5%) had cross-tier disagreement; tier-3 arbitration not used (no tier-3 configured; tier-2 wins).

See [CASCADE_DESIGN.md](https://github.com/davanstrien/classify-and-augment/blob/main/CASCADE_DESIGN.md) for the disagreement-driven cascade strategy.