[billyoneal]

You need a needle that triggers the problem, AND a haystack where the delta2 table matters, AND the algorithm to stop at a problematic index where it uses a bad delta2 value.

There's a reason the published version even in 'the papers' was wrong for 3 years.