| Dimension | Shared | Only in A | Only in B | Overlap |
|---|---|---|---|---|
| Sections | 0 | 1 | 1 | 0% |
| Commands | 0 | 0 | 0 | — |
| Section tags | 0 | 1 | 2 | 0% |
What each file covers
Sections
0 shared · 1 only in A · 1 only in B- − Supabase Policies Memory
- + ML Instructions
Commands
neither file has anySection tags
0 shared · 1 only in A · 2 only in B- − performance
- + do-not
- + agent-behaviour
Line diff
Aledon8/OpenLeukemia · supabase/policies/CLAUDE.md
@@ −1 @@
1# Supabase Policies Memory
2
3Apply this inside `supabase/policies/`.
4
5- RLS is a primary protection layer, not a later add-on.
6- Policies should make patient ownership and explicit consent enforceable.
7- Keep access paths narrow and auditable.
8- Treat document storage and normalized clinical data as separate access concerns.
9- Review policy changes as security-sensitive even when the SQL diff is small.
10
Aledon8/OpenLeukemia · .github/instructions/ml.instructions.md
@@ +1 @@
1---
2applyTo: "ml/**/*"
3---
4
5# ML Instructions
6
7- Keep structured ML workflows separate from LLM explanation workflows.
8- Use measurable criteria: split, metric, threshold, calibration, and known failure modes.
9- Do not commit patient-identifying or sensitive medical data.
10- Model outputs support signals and trends, not diagnosis or treatment guidance.
11- Document label assumptions, cohort limitations, and evaluation caveats near the relevant code or report.
12
@@ −1 +1 @@
1−# Supabase Policies Memory
1+---
2+applyTo: "ml/**/*"
3+---
24
3−Apply this inside `supabase/policies/`.
5+# ML Instructions
46
5−- RLS is a primary protection layer, not a later add-on.
6−- Policies should make patient ownership and explicit consent enforceable.
7−- Keep access paths narrow and auditable.
8−- Treat document storage and normalized clinical data as separate access concerns.
9−- Review policy changes as security-sensitive even when the SQL diff is small.
7+- Keep structured ML workflows separate from LLM explanation workflows.
8+- Use measurable criteria: split, metric, threshold, calibration, and known failure modes.
9+- Do not commit patient-identifying or sensitive medical data.
10+- Model outputs support signals and trends, not diagnosis or treatment guidance.
11+- Document label assumptions, cohort limitations, and evaluation caveats near the relevant code or report.
1012
