58 lines
1.1 KiB
Markdown
58 lines
1.1 KiB
Markdown
# Review Agent Regression Test Workspace
|
|
|
|
This workspace isolates regression testing for review-oriented CCPE artifacts.
|
|
|
|
## Purpose
|
|
|
|
Evaluate whether migrated CCPE System Lite prompts preserve or improve the review quality of mature CCPE 2.0 prompts.
|
|
|
|
Primary question:
|
|
|
|
```text
|
|
Should mature review agents be migrated by full CCPE System rewrite, or by original-kernel + minimal Lite wrapping?
|
|
```
|
|
|
|
## Directory Layout
|
|
|
|
```text
|
|
articles/
|
|
User-provided full article texts for testing.
|
|
|
|
prompts/
|
|
Prompt variants under test.
|
|
|
|
results/
|
|
Raw model outputs from ChatGPT/Codex-side tests and Gemini-side tests.
|
|
|
|
rubrics/
|
|
Evaluation rubric and scoring protocol.
|
|
|
|
reports/
|
|
Final comparison reports.
|
|
```
|
|
|
|
## Test Families
|
|
|
|
```text
|
|
1. 巨人认知 / Giant Cognition
|
|
2. 认知显影 / Cognitive Imaging
|
|
```
|
|
|
|
Each family should test three prompt variants:
|
|
|
|
```text
|
|
A. original-ccpe-2
|
|
B. ccpe-system-lite
|
|
C. original-kernel-minimal-lite
|
|
```
|
|
|
|
Each variant should be tested in two environments:
|
|
|
|
```text
|
|
1. ChatGPT / Codex-side model
|
|
2. Gemini 3.1 Pro
|
|
```
|
|
|
|
Use a fresh thread or subthread for each test run to avoid context contamination.
|
|
|