The Divergence Lab

The Mutation Lab proves lasair rejects garbage. The live fuzzer is harder: it imports honest-looking blocks into lasair and a reference client and compares the whole state root after every block. The first byte that disagrees is a real protocol bug. Here are sixteen of them, each with the actual report diff where there is one, the trace that found the cause, and the fix. B1–B9 come from lasair's M1 campaign on Graypaper 0.7.2. C1–C4 come from the official fuzzer's first GP 0.8.0 batch. C5–C7 were found without any fuzzer, by reading lasair's accumulation against the Graypaper: no published vector or trace reaches them. Read one and you learn the protocol the way we did: by watching it break.

How to read a divergence

The state root gives no partial credit — one wrong byte fails the whole block. So a report is small and exact. Read it in three moves, shown on every card below:

  1. How deep did it get? Imported blocks before the stop. Never compare across seeds — depth is a property of the seed.
  2. Which key diverged? The first key byte names the state component — that is your first hypothesis.
  3. How wrong? Expected vs computed, byte for byte. And check the pre-state: if pre == exp ≠ got, you modified what the reference left alone.
Select a divergence…

How each one was fixed

Every card above ends with the real change — the before/after code under “The fix, in code” — and the root cause and lesson that explain why. The bug is the lesson; the fix is the proof you understood it.