Direct answer
Word-length steganography is a linguistic hidden-information technique that encodes data in the measured lengths of words. A simple binary scheme can define odd-length words as 0 and even-length words as 1, then read those states in a fixed order.
How parity encoding works
The sender and decoder must share an exact segmentation rule. One deterministic rule is to scan left to right, keep only alphabetic word runs, count the letters in each word, map odd lengths to one bit value and even lengths to the other, and then group the bitstream into fixed-size units such as 8-bit ASCII bytes.
Worked benign example
Suppose a four-word demonstration uses lengths 3, 4, 5, and 6. Under the rule odd = 0 and even = 1, the parity sequence is 0, 1, 0, 1. The visible words can still form ordinary prose; the secondary channel exists only because the decoder measures the agreed property.
Capacity and naturalness
A binary parity rule carries at most one raw bit per eligible carrier word before synchronization or error-correction overhead. In practice, forcing every word to a required parity can make writing awkward, so natural-looking text often trades payload rate for better linguistic quality.
Why the channel is fragile
Word-length channels are structural and position-dependent. Inserting or deleting a word can shift every later bit, while replacing a word with a synonym of different parity can flip a bit. Ordinary editing, paraphrasing, translation, or rewriting can therefore destroy the payload even when the visible meaning remains similar.
Punctuation and tokenization rules matter
A decoder must know whether punctuation is ignored, whether contractions or hyphenated forms are split, which scripts count as letters, and whether segmentation uses human word boundaries, a regular expression, or a model tokenizer. FugitiveAI CASE 011 deliberately uses alphabetic A-Z word runs so the counting rule remains independently reproducible.
How to solve word-length puzzles
First identify the carrier order and the exact definition of a word. Then count lengths consistently, write the mapped parity bits without skipping positions, group the bits according to the hint or expected encoding, and decode the resulting values. If the output is nonsense, recheck segmentation before changing the bit mapping.
Detection and defensive analysis
A word-length channel can be tested by extracting the same word sequence and analyzing parity, periodicity, and position. The method is not inherently invisible to software; once the extraction rule is suspected, decoding is straightforward and deterministic.
Frequently asked questions
How much data can word-length parity hide?
A binary odd/even rule has a nominal ceiling of one bit per carrier word, but usable capacity is lower when naturalness, synchronization, and robustness are considered.
Does punctuation count toward word length?
Only if the scheme defines it that way. A reliable challenge must state the segmentation and counting rule explicitly; CASE 011 counts alphabetic A-Z word runs and ignores punctuation.
Can paraphrasing remove a word-length hidden message?
Yes. Paraphrasing commonly changes word choice, length, order, and count, so it can flip bits or desynchronize the entire channel.
Is a language-model token the same thing as a word?
No. Model tokenizers can split or combine surface text differently from human word boundaries, so a word-length scheme must not assume tokenizer parity unless that is explicitly part of the design.