Field

Lossless data compression — general-purpose, with specialist paths for mathematically-structured data (numeric sequences, sampled signals) and for executable machine code.

Methods disclosed (to establish prior art)

The following methods, and their combination within a single lossless codec, are disclosed publicly as of the date above:

1. Best-of-modes lossless codec. Dividing input into blocks (or treating a whole file as one unit) and, per unit, evaluating several candidate reversible transforms, emitting whichever yields the smallest output with a small tag identifying it — such that output is never larger than a plain entropy-coded baseline (a never-expand guarantee).
2. Integer arithmetic-sequence (“generator”) detection. Recognising that a run of fixed-width little-endian integers (e.g. 32- or 64-bit) forms an exact arithmetic progression, and replacing the entire run with a compact generating formula (first value, step, count) that regenerates it exactly on decode — applied to real-world monotonic data such as timestamps and sequence identifiers.
3. Wrapping integer delta of multi-byte words. Replacing each fixed-width integer word with its modular (wrapping) difference from the previous word — a true multi-byte subtraction rather than a byte-wise one — so near-monotonic sequences reduce to small, highly-compressible residuals.
4. Fixed linear prediction with byte-plane separation for sampled audio. For multi-channel 16-bit PCM: de-interleaving into channels; applying a per-channel fixed-order linear predictor (repeated differencing, orders 0–4) chosen per channel; splitting each residual into low/high byte planes; entropy-coding the planes; and storing the container envelope verbatim so the whole file round-trips bit-exact.
5. Branch-target stream separation for x86 code (BCJ2-style). Scanning code for CALL/JMP/conditional-jump opcodes; converting qualifying relative operands to absolute addresses; routing those targets into separate streams distinct from the remaining code; recording, in a per-candidate decision bitstream, which opcodes were converted; and entropy-coding the streams separately — so repeated targets and the cleaned code stream each compress better apart than together, with exact reconstruction on decode.
6. Size-minimising selection across the above. Choosing, per block or per whole file, among the general path and any applicable specialist path (numeric, audio, executable) the encoding producing the smallest archive, recording only enough metadata to reverse the chosen transform.

Statement of prior art

The methods above, individually and in the combinations described, are hereby disclosed to the public as of 15 July 2026 to establish prior art, intended to prevent any party from obtaining valid patent claims over these methods or their described combination.

Reservation of rights

© 2026 Adamantware OÜ. This defensive publication discloses methods for prior-art purposes only. It does not waive copyright in the MIKE source code or in this document, trade-secret protection over implementation details, parameter values, qualification tests and selection heuristics not disclosed herein, or any trademark rights. M.I.K.E.™, puzzleCompressor™ and Adamantware™ are trademarks of Adamantware OÜ.