Glossary

Archive — The compressed file produced by a codec. It includes format information as well as coded data.

Arithmetic coder — A way to turn a sequence of probability-based decisions into a compact bitstream by narrowing a numerical interval.

Back-reference — An instruction to copy bytes from an earlier part of the already decoded output. Used by LZ77.

Bit — A value of 0 or 1. Eight bits form a byte.

Binary — Base-two notation, using only 0 and 1; each digit of a byte is one bit.

Byte — A number from 0 to 255, represented by eight bits. Files in this project are byte sequences.

Codec — An encoder and its matching decoder, plus the archive format they agree on.

Decoder — The program logic that reads an archive and reconstructs the original bytes.

Context — Already-seen bytes or bits used to predict the next value. An order-3 byte context is the previous three bytes.

Entropy — Average ideal coding cost for a known probability distribution. It is a reference point, not the size of an actual archive.

Escape — A coded signal that the current model or context did not supply the actual value; the decoder should try a fallback.

Exclusion — Removing values seen at a longer context from the choices at a shorter context after an escape. Used by this repo’s PPMD-style codec.

Frequency — A count of how many times a symbol followed a particular context or appeared in a block. Counts are used to estimate probabilities.

Header — The initial archive fields that identify a format and carry information such as the original length.

Hexadecimal — Base-sixteen notation for bytes, using digits 0–9 and letters A–F. 0x41 is the same byte value as decimal 65.

Huffman code — A prefix-free variable-length code derived from symbol frequencies.

KiB — A kibibyte, equal to 1,024 bytes. A 4 KiB window holds up to 4,096 bytes.

Literal — A byte written directly rather than as a copy from earlier data.

Little-endian — Storing the least significant byte of a multibyte integer first; the number two as a two-byte value is 02 00.

Lossless — Decompression restores every original byte exactly.

Magic — A recognizable header signature at the start of an archive; the CLI uses its first four bytes to select a decoder.

Mixer — A model that combines several next-byte or next-bit predictions into one probability distribution.

Model — The state and rules used to predict the next byte or bit. Encoder and decoder maintain matching model states.

Order — How much recent history a context includes. Check whether the particular codec counts bytes or bits.

Probability — A number from 0 to 1 assigned to an outcome; larger values mean the model expects that outcome more strongly.

Renormalization — Expanding a narrowed integer coding range after settled leading bits have been emitted, so more symbols can be coded with finite precision.

Payload — The encoded data after a format’s header and length fields.

PPM — Prediction by partial matching: try a long context, then shorter contexts if needed.

Prefix-free — No complete codeword is the start of another codeword, so a decoder can split a bitstream unambiguously.

Round trip — Compress, decompress, then verify that the output is byte-for-byte identical to the input.

Symbol — One value coded by a model; in this project it is often a byte, sometimes a bit, and sometimes an escape event.

UTF-8 — A common text encoding in which a written character may occupy one or more bytes.

Window — The earlier portion of input or output in which an LZ or match model searches for repeated data.

Return to the reading path or the interactive examples.