tine diff

tine diff loads two runs, finds the common ancestor of their main refs when possible, then prints graph steps that only appear in the first or second run.

Usage

tine diff <run_a> <run_b>

Example

tine diff failed.tine retry.tine
tine diff a3f8c1d2 b7e2f4a1

Diff: a3f8c1d2e9f0 vs b7e2f4a1c033

base  6d4a0b270a5f  6d4a0b270a5f  =
      > 3a9dbd19aa20 tool search(query="too broad")      only A
      x 69b7192c7fde error "No relevant sources"         only A
      > c2278d01bb2a tool search(query="peer reviewed")  only B
      + e88991a12caf done "Found stronger evidence"      only B

What It Compares

The current beta diff is graph-oriented. It reports the common ancestor and step IDs present only in run A or only in run B. It does not yet perform a semantic similarity comparison between changed step payloads.

When to Use It

• After tine fork, to see what the retry changed.
• After harness replay, to compare the recorded source with the rerun output.
• For regression checks where shared content-addressed steps should remain unchanged.