Build the authenticator's read-back plan for a value and mode (§15.5). rng is
a [0,1) source — the platform CSPRNG in the UI, a seeded LCG in tests — standing
in for the authenticator's live, unpredictable choice of which cells to ask for.
The strategy is size × type × mode:
big (>512-bit): the hash-derived fingerprint-middle cells (avalanche,
homoglyph-clean) — all of them voice-only, a couple to bind after a paste.
small, or medium programmable: all cells (no sound sample exists).
medium constrained: a run of consecutive cells from a random start (in
reading order, so it may span rows) — robust where a single row is too narrow.
paste-bind on a constrained value: a couple of cells to bind the machine's
exhaustive compare to the live reader (a fingerprint cell if big).
A sampled plan (consecutive, bind) on a confusable alphabet gets one extra
cell for homoglyph compensation (§14.5); an all-cells or fingerprint plan does not.
Build the authenticator's read-back plan for a value and mode (§15.5).
rngis a [0,1) source — the platform CSPRNG in the UI, a seeded LCG in tests — standing in for the authenticator's live, unpredictable choice of which cells to ask for.The strategy is size × type × mode:
A sampled plan (consecutive, bind) on a confusable alphabet gets one extra cell for homoglyph compensation (§14.5); an all-cells or fingerprint plan does not.