Context
The definition isolates the information that a prefix must preserve when every legal suffix has a fixed remaining length. It is the common interface behind the layered-state, one-way communication, and one-pass memory results below.
Definitions
- For , a prefix has future row given by .
- Two prefixes at the same cut are equivalent exactly when their future rows agree on every suffix of the required length. Write for the set of distinct reachable rows and .
- A strict layered recognizer advances through disjoint grades in the fixed symbol order; only states reached by prefixes count.
- A deterministic one-way protocol sends one prefix-dependent message to a suffix reader. Message names not produced by any prefix are discarded.
Hypotheses and scope
- The alphabet is finite and nonempty.
- The total length and cut position are fixed.
- Recognition is deterministic and exact on every word of the selected length.
- Binary message cost uses one fixed number of bits for every reachable message at that cut.
Proof or evidence
The specification separates literal matrices from their quotient by equal rows and fixes the categories in which minimum and uniqueness are later asserted. The proof record uses these conventions without changing them at boundary cases such as total length zero or a constant recognition slice.
Verification notes
The definition was checked against the M25 proof, every Boolean function on binary words through length four, and the endpoint and constant-slice boundaries. Reachability, strict grades, exact remaining length, complete mutable configurations, and fixed-length rather than prefix-free message coding are all explicit.
Limitations
- The per-length canonical recognizer is not by itself one uniform streaming algorithm across all lengths.
- Row cardinality is distinct from matrix rank.
- The definition does not cover randomized, nondeterministic, interactive, or variable-length protocols.
Notes
Exact-length rows, reachable states and messages, strict fixed-order layers, and fixed-length codes are part of the category. Row cardinality is distinct from matrix rank, and the per-total-length canonical program is not by itself a uniform streaming algorithm.