| Dimension | Shared | Only in A | Only in B | Overlap |
|---|---|---|---|---|
| Sections | 0 | 2 | 1 | 0% |
| Commands | 0 | 0 | 0 | — |
| Section tags | 0 | 4 | 1 | 0% |
What each file covers
Sections
0 shared · 2 only in A · 1 only in B- − ML Workspace Instructions
- − Rules
- + ML Inference Memory
Commands
neither file has anySection tags
0 shared · 4 only in A · 1 only in B- − code-style
- − monorepo
- − do-not
- − agent-behaviour
- + performance
Line diff
Aledon8/OpenLeukemia · ml/CLAUDE.md
@@ −1 @@
1# ML Workspace Instructions
2
3This directory is for training pipelines, inference experiments, evaluations, and notebooks.
4
5## Rules
6
7- Keep structured ML workflows separate from LLM explanation workflows.
8- Prefer reproducible scripts and documented evaluation outputs over one-off notebook state.
9- Do not treat model output as diagnosis, urgent clinical judgment, or treatment recommendation.
10- Track assumptions about input data, labels, validation, and limitations close to the code or report that uses them.
11- Avoid committing patient-identifying or sensitive medical data.
12- Prefer measurable success criteria: metric, split, threshold, calibration, and known failure modes.
13
Aledon8/OpenLeukemia · ml/inference/CLAUDE.md
@@ +1 @@
1# ML Inference Memory
2
3Apply this inside `ml/inference/`.
4
5- Keep inference interfaces typed, deterministic, and easy to validate.
6- Return explainable scores or signals with limitations, not clinical conclusions.
7- Preserve compatibility with `ai-service/` consumers when inference code is promoted.
8- Validate feature order, units, and missing-value behavior explicitly.
9
@@ −1 +1 @@
1−# ML Workspace Instructions
1+# ML Inference Memory
22
3−This directory is for training pipelines, inference experiments, evaluations, and notebooks.
3+Apply this inside `ml/inference/`.
44
5−## Rules
6−
7−- Keep structured ML workflows separate from LLM explanation workflows.
8−- Prefer reproducible scripts and documented evaluation outputs over one-off notebook state.
9−- Do not treat model output as diagnosis, urgent clinical judgment, or treatment recommendation.
10−- Track assumptions about input data, labels, validation, and limitations close to the code or report that uses them.
11−- Avoid committing patient-identifying or sensitive medical data.
12−- Prefer measurable success criteria: metric, split, threshold, calibration, and known failure modes.
5+- Keep inference interfaces typed, deterministic, and easy to validate.
6+- Return explainable scores or signals with limitations, not clinical conclusions.
7+- Preserve compatibility with `ai-service/` consumers when inference code is promoted.
8+- Validate feature order, units, and missing-value behavior explicitly.
139
