All digital data, whether stored or communicated, can be subject to corruption that results in the received information differing from the original. In some circumstances, no additional information is available to the receiver, and that is called a hard detection situation. In others, it could be that the bits are received with a reliability measure, called the soft detection.

The engineering solution to this problem is to code the data. A code adds redundancy to the data, and one fruitful way of thinking about it is one takes the original data and appends a hash of it. The hash itself is, of course, potentially subject to the damaging effects of corruption, but this additional redundancy can help inform the receiver about the integrity of the data they have received. There are a wide variety of codes - that is, rules for the creation of that hash - and they were all co-designed with specific decoders that use the code-book structure, the hash structure, to try and guess what was stored or communicated based on what was received.

Most of these codes or hashes are individually only suitable for the hard or soft detection setting because they only have a hard or soft detection decoder suitable for them. Because the decoder is entirely coupled to the code structure, this typically means one needs a distinct implementation to decode each of these codes and often, when done in hardware, even distinct pieces of hardware for different levels of redundancy.

Guessing Random Additive Noise Decoding (GRAND) is a methodology based on an innovative paradigm. It is agnostic to the code structure as it aims to identify the noise effect that has impacted the data, solely using the codebook for what it is: a hash of that data. GRAND can directly decode any moderate redundancy code, including non-linear ones. The efficiency of both hard or soft input versions of GRAND have been demonstrated in published circuits and chips.

The accepted paradigm behind the design of long, high-redundancy codes for soft detection settings is to create them by composing shorter component codes. The decoding of such codes then proceeds iteratively with a soft-input, soft-output (SISO) decoder. It has been established that soft input GRAND algorithms can readily produce highly accurate soft output. Using that feature, soft output GRAND can be used to decode long, powerful codes whose performance exceeds those in standards in a highly parallelised, low latency architecture.