File size: 1,671 Bytes
8c10cf2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
41
42
43
# BookWriter-A2A (RQ1)

This folder contains evaluation scripts and the reference trajectory configuration for the **BookWriter-A2A** runs under the `RESULTS` directory.

## Files

- `evaluate_trajectory.py`
  - Evaluates multiple trajectory metrics (e.g., exact match, in-order match, any-order match, precision, recall).
  - Input: per-session `execution_path.md` files under each model's `test_results/<session>/` directory.
  - Output: a CSV summary (and optionally a CSV of failure reasons when enabled).

- `evaluate_success.py`
  - Computes per-model success rate based on generated artifacts in each session folder.
  - Output: JSON (detailed) and CSV (summary).

- `analyze_retry_patterns.py`
  - Analyzes retry patterns and error locations from `execution_path.md`.
  - Output: JSON/CSV summaries.

- `reference_trajectory.yaml`
  - Defines the reference (ground-truth) trajectory and evaluation configuration.

## Usage

Run from this directory:

- Trajectory evaluation:
  - `python3 evaluate_trajectory.py --config reference_trajectory.yaml --output evaluation_results.csv`

- Trajectory evaluation with failure diagnosis (CSV only):
  - `python3 evaluate_trajectory.py --config reference_trajectory.yaml --output evaluation_results.csv --diagnose-failures`

- Success-rate analysis:
  - `python3 evaluate_success.py`

- Retry-pattern analysis:
  - `python3 analyze_retry_patterns.py`

## Notes

- These scripts assume the default `RESULTS` layout used in this repository (see the `BASE_DIR` / `PROJECT_NAME` constants in each script).
- Markdown report generation is intentionally disabled in the trajectory evaluator; outputs are written as CSV/JSON only.