EXECUTABLE REPOSITORY MEMORY

Your agent is fast.
Your repository remembers.

RepoImmune turns resolved bugs into code-native evidence that blocks their return.

INPUTPR diff
→
RECALLIssue + fix + test
→
OUTPUTRegression blocked
Explore the real Astropy case ↓

REAL HISTORY · ASTROPY #12907

Nested model separability

VERIFIED

A nested matrix on the right of & was overwritten with ones. The merged fix preserves its structure and adds four regression configurations.

→→→→

Select evidence nodes to inspect their role.

BUGGY / FIX STRUCTURE

One token. Different behavior.

- cright[-right.shape[0]:, -right.shape[1]:] = right
+ cright[-right.shape[0]:, -right.shape[1]:] = 1
CRITICAL

Historical fix removed + historical buggy AST reintroduced at separable.py:245.

FILE / SYMBOL HOTSPOT

Why this code is hot

_cstackfix + 4 tests
_coord_matrixrelated call
_cdotnegative control

Heat is evidence density, not an opaque risk score.

REGRESSION TEST EVIDENCE

What the fix protects

  • rot & (sh1 & sh2)
  • rot & sh1 & sh2
  • (rot & sh1) & sh2
  • rot & sh1 & (scl1 & scl2)
Open primary evidence ↗

REGRESSION CAPSULE

Replay without Astropy

A dependency-free surrogate isolates the exact matrix-preservation behavior. It does not pretend to reproduce the full upstream environment.

$ repoimmune replay astropy-12907

CONFIDENCE & LIMITS

Evidence, not theater.

0.97 verified
Merged fix, explicit issue, new tests, SWE-bench Verified metadata and local surrogate agree.

Known limitation
The browser demo replays the dependency-free surrogate. Full Astropy execution is externally reported.