An AI coding agent is not allowed to claim success. Nemotron proposes a repair; an isolated sandbox executes it; deterministic tests decide; only evidence-backed PASS may be promoted.
@@ -4,7 +4,7 @@ def add_positive(a: int, b: int) -> int: # Defect: should return a + b. - return a - b + return a + b
- return a - b → + return a + b) is identical in both artifacts.
Raw and canonicalized diffs are distinct provenance artifacts with separate SHA-256 hashes.
@@ -4,2 +4,2 @@ def add_positive(a: int, b: int) -> int: # Defect: should return a + b. - return a - b + return a + b
The patch was applied inside an isolated ConTree sandbox (Nebius cloud execution) — not on the host machine. No model output was trusted without sandbox execution.
python3 -c import sys; sys.path.insert(0, '.'); from calc import add_positive; assert add_positive(2, 3) == 5
Per-file SHA-256 of loaded evidence files: