Redundancy (Claude Shannon, 1948)

The fraction of a message that is predictable — determined by the code's constraints rather than by fresh choice. Redundancy is not waste: it is exactly what makes a message survive noise, because the receiver can reconstruct dropped or garbled parts from the rest.

Formula: R = 1 - H / H_max

• H — actual entropy per symbol of the source (see shannon-entropy).
• H_max = log2(N) — the entropy if all N symbols were equiprobable and independent (the maximum possible).
• R ranges 0 (no redundancy) to 1 (fully predictable). Written English runs roughly R ~ 0.5-0.75.

Transfer. Zero-redundancy prose is theoretically maximal information density and practically unreadable — one typo or one skimmed line and meaning collapses, because nothing lets the reader recover it. Real channels are noisy (readers skim, get interrupted, half-attend), so being-read requires engineered redundancy: the restatement, the example after the claim, the summary line, the parallel structure. This is why closure and repetition aid comprehension rather than insult it. The craft question is never "redundancy or not" but where to spend it — pad the connective tissue so the load-bearing surprises land intact. Agents that compress everything to the terse maximum produce brittle, un-skimmable text.

Kin: shannon-entropy · mutual-information · channel-capacity · models · two-pleasures-arousal-and-closure