UAE Inventor's Rule-Based Engine Holds a 54,796,312-Byte Array in 32 Bytes

The Knowledge Triangle Route (KTR) project, an independent research framework based in the United Arab Emirates, has published a reproducible benchmark showing how a 54,796,312-byte data array can be stored and reconstructed using a single 32-byte mathematical rule. Developed by independent inventor Roman Burtsev, the Information Engine (INEN) keeps the mathematical rule that generates an array rather than a snapshot of it, reconstructing the original data bit-for-bit on demand. A US provisional patent application for the method was filed in June 2026.

Benchmarked Against Production Standards

INEN was evaluated against production-grade differencing followed by LZMA, the Lempel-Ziv-Markov chain algorithm that underlies the standard baseline used in systems such as Apache Parquet, FLAC, and Gorilla, rather than against general-purpose compressors fed plain text.

"We evaluate against delta encoding, not against general-purpose compressors fed ASCII text. The delta baseline is what cloud engineers actually run in production, so that is the baseline worth publishing."

— Roman Burtsev, Inventor of the Information Engine

A Rule That Doesn't Grow

The stored rule remains 32 bytes at every array length tested, while the delta coder must record state changes for every element, causing its output to grow from 108 bytes at a thousand elements to 7,696 bytes at 6.5 million, widening the gap from 3-fold to 240-fold across that range. At full scale, the raw array occupies 54,796,312 bytes while the rule occupies 32, the same 32 bytes that hold a dataset of just a thousand elements.

The advantage becomes more pronounced on machine-generated data such as simulation output, synthetic AI datasets, and high-frequency logs, which is globally deterministic but locally random. On the output of a linear congruential generator, a 56-byte rule opened a 7,881-fold gap over the same delta baseline; on a Fibonacci mod p sequence, a 36-byte rule opened a 15,440-fold gap. According to the project, differencing cannot close these gaps by construction, and a stronger compressor would not help either, since LZ-family methods already operate at the entropy rate of a stationary source — optimal under Shannon, and blind under Kolmogorov for exactly that reason.

Three Things a Compressor Cannot Do

The method also demonstrated an ability to recover missing data: when 500 values were deleted from a test series, it restored 499 of them exactly and returned no incorrect values, a capability the project says a standard compressor cannot replicate, since a compressor can only return what it originally stored. The threshold that separates a found rule from a coincidence is derived, not chosen: a rule of degree d consumes d+1 points, so at least one more must remain to test it against. Without that margin, noise yields a rule every time, and every gap it fills is filled wrongly.

Where no underlying rule exists in a dataset, INEN returns a "NO_RETURN" status rather than a fabricated result, a distinction the project frames as important since a compressor always returns something, even absent genuine structure. The project also says its compression ceiling is computable before compression begins, from a single measurement of the stream: unbounded for integer arithmetic, but finite and nameable in advance for chaotic dynamics.

"A ratio can be disputed. A ceiling named before the work starts is a commitment."

— Roman Burtsev, Inventor of the Information Engine

Economics of Access Frequency

Because the rule is found once while the savings accrue on every subsequent read, the project says access frequency determines the economics of deployment. By its own estimate, a cold archive read once a year would never pay back the upfront cost of finding the rule, while a hot cache accessed a million times a year would pay back within days, placing the method's practical use in active data paths rather than long-term cold storage.

Built to Be Independently Verified

Every quantitative result in the benchmark is produced by a terminating script that reproduces the numbers bit-for-bit, and every comparison includes both a naive baseline and the standard production baseline. The full benchmark, codebase, and execution logs have been deposited on the Open Science Framework, allowing external researchers to confirm the hashes against the manifests and independently rerun the work.

INEN is the fifth in a sequence of methods filed and deposited by Burtsev since June 2026, with each verified before the next was built on it, following the project's initial disclosure of the broader KTR framework. More information is available at ktr-project.org, and media inquiries can be directed to Leila Salieva at press@ktr-project.org.