T-3.1

Line Diff → Entity Operations (10-Diff Suite)

Not yet tested
Hypothesis: A standard git diff can be parsed into a correct sequence of entity-level operations (add, modify, rename, relate, delete).
Test: Supply 10 real diffs of increasing complexity:
  1. Single line change inside a function
  2. New function added
  3. Function renamed
  4. Function moved to a different file
  5. Import added → new dependency edge
  6. File renamed
  7. Two files merged into one
  8. One file split into two
  9. Interface changed → multiple dependents updated
  10. Large refactor: rename + move + split + new dependencies

For each: generate entity ops, apply them to the graph, re-project the affected files, and diff against expected output.

Pass: ≥ 8/10 diffs produce correct entity ops AND the re-projected output matches expected within tolerance.
Fail: < 8/10 pass, or any diff causes graph corruption (orphaned nodes, broken edges).
T-3.2

Conflict Resolution

Not yet tested
Hypothesis: When two agents (or an agent and a human) modify the same entity concurrently, the graph can detect and resolve the conflict without data loss.
Test: Two concurrent modifications to the same entity: Agent A adds a method, Agent B renames the class. Apply both. Check graph integrity and that both changes are preserved.
Pass: Both changes present in final state. No silent overwrites. Conflict detected and logged.
Fail: One change silently lost, or graph enters inconsistent state.