B92Finite

Appendix A's deviation bound, and the key length no B92 paper writes

6/6 passed in 0ms

Test	What it does	Result
test_azuma_inverts	Both widths put back into Azuma's inequality as the paper scales it return the failure probability asked for.	✅ pass
test_azuma_scales	The width falls as 1/sqrt(N) and rises as sqrt(log(1/eps)): martingale scaling, not Chernoff.	✅ pass
test_azuma_splits	The two widths cross at an even split, are smallest there, and each diverges as its own pool empties.	✅ pass
test_length_refused	Neither branch returns a key length, the block size is checked first, and the two refusals differ.	✅ pass
test_plain_sources	The plain refusal names three published finite-key analyses of a different B92, flags the journal-only one as unread, and separates them from the extended-B92 proofs.	✅ pass
test_strong_gaps	The strong-reference refusal enumerates four missing pieces and names b92_azuma, the paper it stops inside and the asymptotic argument its rate is read off.	✅ pass