Author: Jack Kowalski https://entropment.com Patent Pending 1. What exactly is RIPA QCO, if it is neither a cipher, a cryptographic primitive, nor a cryptographic tool? In reality, it is a concept arising from observations at the intersection of analysis, numerics, and research (including work related to astrophysics). The idea is that an accumulating numerical error in a finite representation is irreversible, and classical “ordering” operations (e.g., orthonormalization) do not recover the original state-they only restore local matrix order. Instead of minimizing this effect, we decided to maximize and control it. The algorithm is a consequence of this reasoning: a rapid and strong focus on the error itself, followed by formal anchoring and closing of the loop so that nothing analytically meaningful can be done with the object. Detailed explanation for programmers: https://entropment.com/media/downloads/PAPERS_programmer_friendly_Entroper_explanation.txt 2. Why does the product, by design, not meet the conditions for recognition as a cryptographic tool? Because the requirements imposed on cryptography (reversibility, confirmation oracle, reduction to a known problem, etc.) create attack surfaces. The goal was to make the message so “secret” that it is practically unrecoverable-even for the legitimate recipient, who receives only an approximation. This has more in common with recovering approximate preimages from a deformed observation than with classical cryptography. 3. What does the statement that “the FPU residual becomes an element of the protocol, not noise to be minimized” mean in practice? Analysis is effective only when the objects under study have some sense (coherence). We therefore decided to deliberately produce nonsensical arithmetic. Finite bit representation guarantees deviations from ℝ that are normally minimized. We needed a catastrophic avalanche of such deviations (deviations of deviations…) so that the trace of the input would be erased by the accumulating error. The result is a generator of pathological arithmetic dependent on the key, whose analysis is devoid of sense because the preimages of the representations do not fit in the machine register. 4. In what way is the deliberate breaking of classical analytical benchmarks a design feature here, not a flaw? Analytics is the fundamental way of attacking any object. We therefore created an object so absurd that analytics becomes pointless. The only realistic attack remaining is iteration over keys. The key size was chosen so as to encourage such an attack (because computational power grows and we cannot base security solely on increasing key length). Successfully carrying out this iteration is unconfirmable, pointless, and only multiplies further problems-constantly returning to key iteration and hypotheses that cannot be positively confirmed. 5. What does the “omnipotent Eve” adversary model look like, and what specifically remains undecidable? In the test harness, the adversary is given the key for every block (we do not inquire how she obtained it-she simply has it). She therefore receives the entire stream full of fakes. The fakes may, for example, be identically generated ciphertexts passed through ECC so that they have matching statistics. What remains for her is puzzle assembly: which fragment belongs to what, and which of the ciphertexts is the real one worth further attack. We deliberately adopt the worst possible scenario (Eve receives the keys for free for all blocks, true and false) and even then she is left with arranging an open set of hypotheses. The model is extremely favorable to the adversary. We took the catastrophic scenario as the baseline, and its assumptions are unattainable for a real adversary. 6. Why is the impossibility of positive confirmation of the single correct solution treated as a security property? A signature would require a computational loop that does not serve information deformation and is counterproductive relative to the primary goal. All logics are inverted: they confirm the absence of inconsistency, never confirm consistency. Absence of inconsistency is a very broad field-broad enough that in the fake-generation loop collisions occur extraordinarily often (roughly 2^{1/8}%). Signature and positive confirmation are weaknesses amenable to analysis. 7. In what way does the length of communication become a carrier of difficulty, and why is the set of equivalent reconstructions open? Because of the rapid growth of the 0|1 choice tree when rounding 1/2 (half the maximum key noise + current noise in the integer representation). There are many such operations that yield the same result-none is “true,” and only one trajectory through the tree gives an approximation with correctable BER. Each subsequent message causes further growth of the tree in all directions. 8. What exactly happens to the genealogy of the trajectory in the q_FPU projection, and why can it not be recovered? Any arithmetic in a finite representation has this property. You can test it on your own computer: open a calculator, divide 1 by 7, copy the result to a notepad, reset the calculator, paste and multiply by 7. You do not get 1. 9. What does a typical pipeline look like? The ciphertext yields a buffer. The buffer is enlarged by ECC in such a way that the original buffer can be recovered at a certain percentage of errors. RIPA deforms this buffer and injects fakes (a good source of randomness is, for example, similarly prepared buffers containing ciphertexts of pictures of cute kittens). The recipient opens each packet and if it does not match the pattern of the explicit header-rejects it. Accepted blocks go into the result buffer, which is recovered by ECC and passed onward. A ciphertext is not necessary-it can be an ordinary buffer-but we assume that the client has their own requirements and will not rely solely on the channel itself. 10. What are the requirements for the external ECC, and how is BER ≈ 2–2.5% controlled? BER is controlled by an analogue of QCO that ensures deformation below the bit representation (the original QCO has graded sensitivity under u1024; for u64 it is too coarse). These operations keep BER on the edge of acceptability. The external ECC must deliver a buffer that tolerates such irreversible lossiness. The goal is to obtain, in the bit representation, a flat valley (hill climbing) for the entire range of keys, with a peak of width smaller than a bit for the proper key. Manipulating settings makes sense only when changing the representation (u32, u128). At u64 (52+1) a change causes either a sharp rise in BER or the possibility of hill-climbing on nearby keys. This is an observational property of the bit representation under a given toolchain and calibration, not a defensive mechanism against optimization attacks in the cryptographic sense. This is exactly how we calibrate an oscilloscope. Precise calibration for a given representation is an engineering problem (i.e., how to obtain a fractional effective exponent in the functor), discussed observationally in the analysis: https://entropment.com/media/downloads/PAPERS_RIPA_QCO_EN.txt 11. Is an air-gap possible, and under what conditions? In principle, the HSM is a separate module. The source of the buffer to be processed can also be communicatively separated (unidirectionally: buffer_placed → perform operations). On the other side, RIPA unidirectionally passes the recovered buffer to ECC, and ECC to the decrypting unit. In this context the HSM serves to protect the key bunch and state, not to perform cryptographic operations in the classical sense. 12. What happens if the internal encrypting system is compromised? In principle, nothing changes. Even when the ciphertext is not compromised, fishing it out of the deformed channel is an open computational problem that grows with the length of communication (including the fake communication). RIPA acts as a shield of access to the ciphertext itself. When the ciphertext is compromised, the unestimable cost of attack becomes an additional deterring factor (the risk of burning resources without a reliable success condition). 13. How do the false blocks (decoys) and mutation of the key bunch work? Fakes are rejected by the recipient after checking against the explicit header (they do not match the obtained error results). If a block is accepted, the same block is “decrypted” with the next key, producing gibberish that serves as the source of mutation of all keys. The session breaks every block-the keys are new. The adversary must iterate over keys every 480 bits (including fake ones), because she does not know which block is real. 14. Why does even the correct seed K give only an approximation ~M, and not the original M? Because the message is to be so secret that no one can read it-including the legitimate recipient. This solves the problem of the possibility of reading by anyone else. 15. What are the requirements regarding determinism of the FPU toolchain? The same FPU (with the same rounding settings) and libm are needed on both sides. In a transmitter–receiver pair configuration this presents no difficulty. Libm is needed because some operations (mainly pow, exp and similar) are not part of the IEEE-754 standard. 16. How does the product behave with very short vs. very long sessions? A very short, single-block session without a fake provides practically no protection (iteration over 2^{53} combinations is feasible, and knowing the 480-bit format one can fairly easily find a matching approximation). Each subsequent block (real or fake) sharply increases the combinatorics. With one buffer per second, after an hour the number of hypotheses becomes absurd. 17. On what basis is it claimed that the logic is not subject to export restrictions? The channel logic does not implement functions that are standardly subject to export restrictions concerning cryptographic tools; any cryptography in the payload is subject to separate assessment. 18. What information is public, and what is covered by license and consultations? Everything. The algorithm is fully open together with the implementation in a prototype language (GameMaker Studio), with function and loop names given directly and additional analytical functions (Pearson, Kolmogorov, etc.). Private use, play, and research-no problem. The license covers business or institutional application. Consultations concern analysis of how to implement, what to watch out for, and how to verify-due to the level of complexity and the inverted (non-positive) logics. We recommend them for the client’s good, but we do not block independent attempts. 19. What is the status of the patent pending? It concerns solely the implementation of the object (mathematics cannot be patented). One may analyze freely-we even encourage it. That is why there is open code and educational materials (including the original version for encoding images and the quasi-attractor). 20. What exactly does the implementation consultation cover? Analysis of architecture (if needed), design of the pipeline (if the client needs it), verification of the logic of the RIPA implementation in a more serious version (with memory protection, physical HSM), and design of the HSM itself if required. Plus answers to questions about what can, what is worth, and what is not worth overcomplicating in one’s own pipeline (hashes, map generators, quasi-attractors, etc.). 21. In what way is the license for the logic adapted to the specific architecture of the client? In no way. We do not ask what the client intends to use it for. They purchase a license for the use of a solution tailored to their needs based on the protected algorithm. 22. What are the expected efforts on the client’s side? The pipeline is on the client’s side-they will use and service it. We can consult the logic of state resynchronization (we have it worked out, but do not publish it); however, we answer the questions that are asked. Resync is non-negative, not positive. 23. Are there scenarios in which using this layer makes no sense or may worsen the security situation? Yes-if you have no adversary, there is no sense in concealing data at the cost of computational power. There is no scenario in which the layer worsens security. The goal is not to increase security, but to create danger for the adversary (an enormous cost of forcing a way through the channel). It is de facto a passive offensive tool attacking the adversary’s computational power. 24. What are the main limitations of the construction from the point of view of classical security analysis? None in the classical sense-this tool does not increase security. It buys time by occupying the adversary’s computational resources. It is a combinatorial minefield in front of the shield, not the shield. The effect occurs exclusively in the absence of the current state/seed (a full and ordered key bunch) and does not constitute a closed guarantee. 25. What must be ensured on the implementation side? Identical FPU and libm for the sender and the recipient. 26. What is the situation when the adversary has access to the current state/seed of the RIPA layer? If she has access to K_0 (because she iterated), she has one block-whether real or fake is her problem. She has no access to K_1 other than by dismantling the HSM. If the block was fake, K_0 remains the same. If K_0 changed, the block was real and we have new keys. The number of keys in the starting bunch is whatever someone decides. 27. Does the product provide any guarantees of confidentiality, integrity, or authentication? None. We supply a cable. We do not ask what travels through that cable. We ensure that the “physics” of this cable causes everything to look like white noise (this can be measured-there are tools in the code). All properties expected of cryptography are provided by the cryptography traveling through this cable. 28. In what way can the client independently verify the declared properties? In the demonstrative prototype one can run a test loop and a report generator (input buffer vs. output). There is also a toy ECC for tests on text. One can enable the omnipotent Eve harness (she receives a duplicate of the key from every block) and see what comes out in the report and where the chain breaks relative to the legitimate user. 29. What does the test harness with omnipotent Eve show? The course of the entire operation step by step (the level of detail of the report can be regulated). The text file shows the differences between Eve and Bob. It serves for assessment and analysis of usefulness-to see what a mess arises for the adversary. 30. What security questions should we ask ourselves before deployment that classical cryptographic checklists do not cover? What computational power our adversary may have and how strongly we want to tempt them to burn resources-i.e., where to set the “paranoia slider” when generating fake buffers. This is one of the settings and has no upper bound. 31. “What exactly does ‘quasi-attractor’ mean and how does it affect trajectories.” - a quotient process on alternating inverses of primes as the base of the root of the key; the quasi-attractor has contraction in ℝ guaranteed, in F the same contraction exists but its course is not convergent with ℝ and is non-monotonic with respect to it; simply the curve ℝ does its own thing, and the broken line F jumps around it without order and composition resulting from the numerical errors of such operations; 32. “What does state resynchronization look like in practice (since it is non-negative).” - effective signal of rollback to the last confirmed HSM history, possibility of automatic operation in the absence of “history forget”, a second parallel channel on another bunch. 33. What are the concrete requirements regarding identity of libm (which functions, which roundings). - The principle “preferably identical machines and libraries, and later check what can be broken” is practical. Differences between libm are usually small; differences between FPU types in the processor already matter. 34. Can and how can one measure “white noise” on the output (tools in the code). - with Pearson, Kolmogorov, Lyapunov; there are L2 plots, Hamming by bits, by float values, and somewhere I even have a video on how to use it: https://www.youtube.com/watch?v=GfVSAT-NkI0 35. What happens when precision is changed or when FMA is used vs. separate operations. - BER either becomes too small (wider peak, hill-climbing on neighbors) or catastrophic. 36. Can a backdoor based on keys be inserted? - so that it is not visible in the algorithm directly, it cannot; so that it is not visible in the algorithm by tinkering with the valley-flattening and peak relations so as to obtain simple-to-optimize hill-climbing, yes, but it is very easy to detect with a simple test on one block using a key differing by one bit; if it works-someone has tinkered with the curves and weakened the algorithm; it is detectable in a trivial way; additionally the proper key would give zero BER in such a case every time; in the normal process such behavior is an exceptional artifact for BER to drop to zero and possible on one, long-sought block of special key configuration and open buffer; ----------------------------------------------------- YouTube playlist with this tool (technical, test loops, source code): https://www.youtube.com/playlist?list=PLlfGldDGbUS8qAknuU8wGe4mzdfU5_NMO ----------------------------------------------------- Formally, RIPA QCO satisfies none of the classical definitions of a cryptographic tool. There is no reversible mapping given knowledge of the key, no secret in the cryptographic sense, it does not reduce to a known computational problem, it provides no confidentiality, integrity, or authentication, and it possesses no confirmation oracle. All these absences are deliberate and follow from the construction. From this point of view the statement “this is not cryptography” is correct. At the same time the construction deliberately attacks those places where classical cryptography is analytically weak or where attack surfaces appear, such as the possibility of identifying the correct key, the possibility of confirming that the right solution has been found, the possibility of meaningful algebraic or optimization analysis, and the existence of a closed (finite) solution space. Instead of building these properties and then protecting them, they were removed from the specification, deliberately. The price is the absence of classical cryptographic guarantees. The gain (in the design intent) is an open set of hypotheses and loss of genealogy, which means that even with full knowledge of the generative material one cannot positively confirm the single solution. And even that solution does not exist. This complicates several formal issues: From the side of definition and specification-it is not cryptography. From the side of intention and effect-it deliberately destroys the surfaces that in cryptography serve analysis and confirmation of correctness. From the side of legal / regulatory- a dispute may arise whether “a tool that deliberately hinders analysis and recovery of data through controlled, irreversible degradation based on floating-point arithmetic” should not be treated analogously to cryptographic tools, even if it formally does not meet their definition. If someone starts splitting hairs, the argument will look roughly like this: “It does not meet the definition, but it was designed to hinder access to data in a way that goes beyond ordinary encoding or compression, and at the same time cannot be easily classified as something else.” In such a situation the resolution ceases to be purely technical and enters the area of expert interpretation and intention. Therefore we consistently hold to three things. A precise description of what the object does not do and which properties it does not possess. An open admission that the goal is to hinder analysis through irreversible degradation and an open set of hypotheses (and not to deliver classical cryptographic guarantees). A clear separation of the arithmetic channel logic itself from any cryptography that may be present in the payload. It is impossible to completely eliminate the risk that, under formal regulatory assessment, someone will consider intention and effect sufficient to treat this differently from an ordinary signal-modulation layer. As of 2026 the formal status is that it falls into no regulated category.