compression

Compression needs encoded data to operate right now — without representation/encoding there is nothing to compress. This is constitutive, not merely facilitative.

Compression is the reduction of information redundancy — it is constitutively about information. Remove the concept of information and compression ceases to be a coherent practice. The direction test: information (epoch 0.10) predates compression (epoch 0.99), consistent with a DEPENDS_ON.

Compression requires algorithms — the specific procedures that define how data is mapped to compressed form. Removing algorithms leaves compression without its operative mechanism. Present-tense: every compression system runs an algorithm; no algorithm, no compression.

Compression (in audio, the process of reducing dynamic range) DEPENDS_ON gain: the removal test passes — compression operates by reducing gain (gain reduction) when the signal exceeds a threshold. Remove gain, and compression ceases to operate because gain reduction is the core mechanism; without gain there is no lever for compression to act on.

compression is a signal-processing technique. Remove signal processing and compression has no framework to operate within — its algorithms, theory, and practice all live inside signal processing. Without signal processing, compression cannot function.

A cheat sheet is a condensed reference whose entire identity rests on information compression — maximal useful content in minimal space. Remove compression (the practice of reducing information density while preserving meaning) and a cheat sheet stops being distinguishable from any other document. Removal test passes: X stops OPERATING without Y.

Compression as a concept (reducing data size while preserving information) predates neural network quantization by decades. Model quantization extends compression principles specifically to reducing numerical precision of neural network parameters. Which existed first? Compression came first and fed into the design of quantization techniques.

Test: is quantization a specific kind of compression? Yes — it reduces the information content and storage size of model parameters by mapping high-precision values to lower-precision discrete representations. A competent ML practitioner would call quantization 'a type of compression' — specifically, precision reduction as a compression technique. Per Law 9: specific→general.

lossy-compression IS a specific kind of compression. Per Law 9: a competent speaker would call lossy-compression 'a kind of compression' — it compresses by discarding data, distinguishing it from lossless compression. The nearest accepted kind is 'compression' itself.

The DCT is designed for compression — it concentrates signal energy into fewer coefficients, making it the core transform in JPEG, MP3, and video codec compression pipelines. Its purpose is to serve compression applications.

Dynamic range compression is a specific technique for reducing the amplitude range of a signal. A competent speaker would call DRC 'a compression method.' Direction tested: specific (DRC) → general (compression).

parallel-compression needs compression to operate: the technique blends a dry signal with a compressed signal. Remove compression and parallel compression cannot function — there is no compressed signal to blend. The removal test passes.

Codec SERVES compression — codecs are designed and maintained for the sake of compression (encoding data to reduce size, decoding to restore it). The codec is the servant; compression is the master purpose.