ReconcileCascade

Cascade block sizes, the published parameter table and a counted simulation

9/9 passed in 197ms

Test	What it does	Result
test_block_rules	Martinez-Mateo's near-optimal rule reproduces the first three block sizes of nine of Table 3's twelve rows.	✅ pass
test_cascade_misfits	In the three rows it misses, the closed form is one power-of-two step too large: k2 at 1% and 4%, k1 at 8%.	✅ pass
test_frame_bound	DECLARED GAP. Zero failures in 32 frames bound the frame error rate at 0.0894 with 95% confidence, 961 times Martinez-Mateo's 9.3e-5, so the simulation's count cannot stand in for a correctness parameter.	✅ pass
test_leak_floor	Cascade's disclosed parities exceed the Slepian-Wolf minimum $nh(Q)$, the measured f_EC staying above one.	✅ pass
test_replay_identity	On one error pattern and pass permutations the Rust Cascade discloses the same parities and leaves the same residual errors as cascade_ref, over four variants and both subblock-reuse settings.	✅ pass
test_reuse_gain	Without subblock reuse the same block sizes and fourteen passes leak a tenth more, taking Cascade from 1.04 to about 1.13.	✅ pass
test_simulated_fec	Published anchor: counted Cascade parities over 32 frames of 2^14 bits reproduce Table 3's measured f_EC, 1.04006 at 2% and 1.04313 at 5%.	✅ pass
test_table_anchor	cascade_point returns Martinez-Mateo Table 3 verbatim, 1.04006 inefficiency and 9.3e-5 frame error rate at 2%.	✅ pass
test_variant_sizes	Each variant's first block size follows its own rule: 0.73/Q doubling for the original, 0.8/Q with a fivefold second block for Yan, a power of two for the two optimised sets.	✅ pass