SP 800-90B · entropy assessment

Entropy assessment

The SP 800-90B non-IID estimator suite and restart tests, run entirely in this page. Sample files are read locally and never leave this machine — there is no server behind this. Results match NIST's SP800-90B_EntropyAssessment tool, v1.1.8; that tool remains the tool of record for submissions.

The non-IID track (§6.3): all ten estimators over the raw symbols and, for multi-bit symbols, over the derived bitstring. The result is the initial entropy estimate min(H_original, bits per symbol × H_bitstring), before H_submitter.

drop raw sample files here · or click to choose (several files run as a batch)

sample packing packed samples unpack MSB-first, one zero-padded byte each bits per symbol the low bits of each byte are the sample samples to use the reference tool reads all samples by default

estimator	§	raw symbols · H/symbol	bitstring · H/bit
most common value	6.3.1
collision	6.3.2
Markov	6.3.3
compression	6.3.4
t-tuple	6.3.5
longest repeated substring	6.3.6
MultiMCW prediction	6.3.7
lag prediction	6.3.8
MultiMMC prediction	6.3.9
LZ78Y prediction	6.3.10

Restart testing (§3.1.4): each file is one row dataset of 1000 restarts × 1000 samples. The sanity check compares the largest per-row and per-column symbol count against a cutoff derived from H_I, then the estimator suite runs on the rows and on the columns. Several files run as a batch against the same H_I.

drop restart matrix files here · or click to choose (several files run as a batch)

sample packing packed samples unpack MSB-first, one zero-padded byte each bits per symbol the low bits of each byte are the sample Hᵢ · initial estimate per-symbol entropy from the initial assessment · optional: blank runs the suites without sanity check or validation

estimator	§	rows · H/symbol	columns · H/symbol
most common value	6.3.1
collision	6.3.2
Markov	6.3.3
compression	6.3.4
t-tuple	6.3.5
longest repeated substring	6.3.6
MultiMCW prediction	6.3.7
lag prediction	6.3.8
MultiMMC prediction	6.3.9
LZ78Y prediction	6.3.10