A minimum-description-length instrument for measuring the determinate-to-stochastic ratio of structure across scientific domains

We describe a single instrument that, given any dataset reduced to an integer array, searches a fixed vocabulary of generating primitives ordered by description length, returns the shortest lossless description it can verify, and reports the fraction of the data's marginal entropy that this description removes. Applied unchanged across eleven domains — population genomics, equity-microstructure finance, scalp EEG, network telemetry, acoustics, turbofan physics, stellar photometry, natural language, natural images, natural video, and a cryptographic-randomness control — the instrument produces a FLOOR map: a per-domain measurement of how much of each signal was determined by a discoverable rule versus left to genuine chance. Every reported reduction is verified by byte-exact reconstruction; the instrument returns FLOOR (no rule) on four independent randomness sources, establishing that the map's structure is not an artifact of the method. Highlights: (i) individual genomes are a sparse deviation from a population reference (99.8% sparse), stable as the sample grows; (ii) cross-patient EEG deviation is dense and becomes less predictable as patients are added — opposite organizational regimes stated by one method. A pre-registered unification hypothesis was tested on real data and refuted; the negative result is reported. The instrument is best understood not as a compressor but as a hypothesis generator.