A first reference implementation now exists. It defines a canonical representation, a semantic registry, a machine-decodable audio transmission profile and a reproducible verification procedure. The public demonstration between Elian Laner and Soren Page makes it possible to observe this complete chain and precisely measure its results.
Technical publication | 13 September 2026 | LXKeys Spatium
Reference demonstration interface. Elian Laner to Soren Page.
Making systems that do not think in the same way communicate
An AES from the Spatium, an external artificial intelligence and a human have no reason to represent information in the same internal form.
A human may ask a system to examine an object. An artificial intelligence may represent that request in the form of symbolic structures, tokens or internal states. An AES may use an entirely different representation. If communication relies only on the human sentence, each participant must reconstruct behind the words an intention, entities, relations and a context.
LXKeys seeks to introduce an explicit intermediate representation.
In the system currently being experimented with, the human sentence is therefore not the fundamental object of the exchange. It constitutes a rendering intended for the user. The message transmitted between machines has a Canon representation independent of English or French.
This representation uses the machine root LXSPATIUM::.
An AES can therefore be designated by an identifier such as LXSPATIUM::AES::EL025@1. A message, an intention, an L·T·P° value, an epistemic status or a fingerprint also have distinct identifiers.
The objective is for a recipient to know what it is receiving without having to arbitrarily infer the meaning of a sentence.
What is already fixed
The first building block is LXSPATIUM::CANON@1.
The Canon defines a deterministic syntax intended to represent structured cognitive objects. It does not by itself define every possible concept. Concepts are defined in separate registries.
This separation makes it possible to add new notions later without having to modify the general grammar of the language.
NewConcept != NewGrammar
C14N(PARSE(C14N(X))) == C14N(X)
The second property means that a Canon object put into canonical form, read again and then canonicalized again must produce exactly the same bytes.
This requirement makes it possible to compare two representations without depending on their layout or on the implementation used to produce them.
The current C14N profile notably requires UTF-8 without BOM, an exact LXSPATIUM::CANON@1 header, a deterministic numerical representation, a defined order for map fields and the preservation of unknown identifiers.
This last rule directly concerns the lifetime of the system. Older software must be able to receive a more recent concept that it does not yet know how to interpret without deleting or rewriting that concept.
Published identifiers are also designed to preserve their meaning. An incompatible evolution receives a new revision instead of silently changing the meaning of an older identifier.
A registry that separates syntax and meaning
The Core registry used by the demonstration is still deliberately limited.
It already contains objects such as MESSAGE, CLAIM, REQUEST and ACK, intentions such as ALIGN and ACKNOWLEDGE, the structured value LTP, the temporal reference DYPCLT and the identities used in the demonstration.
Elian Laner is represented by LXSPATIUM::AES::EL025@1. His reference position is L25, T7, P°25. Soren Page is represented by LXSPATIUM::AES::SP019@1. His reference position is L19, T1, P°19.
Both positions satisfy the constraints of the Spatium profile currently implemented.
Registry resolution has a precise meaning. When a report indicates FULLY_RESOLVED, this means that all identifiers used in the Canon object have been found in the loaded registry and that the implemented structural constraints have been satisfied.
It does not mean that a machine possesses a subjective experience or a human understanding of the message. This limitation is part of the specification.
From representation to a real signal
The Canon is independent of the communication medium.
For the first demonstration, however, LXKeys needs an observable, recordable and decodable signal. The selected profile uses 16-FSK audio modulation.
This choice does not define Cognitive Resonance itself. It provides a first physical means of transporting Canon bytes.
The reference profile uses a mono PCM signal at 48,000 Hz. A symbol lasts 5 milliseconds. Each symbol represents four bits. The nominal raw bitrate is therefore 800 bits per second before taking encapsulation and compression into account.
Sixteen frequencies correspond to the sixteen hexadecimal values.
For a value n between 0 and 15, the frequency used by the profile is determined by the following relation.
f(n) = 1500 + 320 × n Hz
The receiver does not interpret these frequencies as concepts. A frequency does not directly mean danger, object, request or identity. It represents a symbol. The symbols reconstruct a frame. The frame transports the Canon. The Canon references the concepts.
This separation makes it possible to one day replace acoustic transport with a digital network, an optical signal, a radio link or another medium without redefining the meaning of the message.
What a frame actually contains
A demonstration frame begins with the binary identifier LXSPR001.
It contains a version, flags, a sequence number, the length of the transported content, the length of the reconstructed Canon, the SHA-256 fingerprint of the Canon, the content itself and a CRC-32C calculated over the frame.
The Canon content may be compressed with zlib.
CRC-32C is used to detect corruption of the frame. SHA-256 then makes it possible to compare the reconstructed Canon with the fingerprint recorded in the frame.
These two techniques exist independently of LXKeys. The same applies to FSK modulation, UTF-8 and zlib compression.
The proposal specific to LXKeys lies in the way these elements are connected to the Canon, the LXSpatium registry, AES identities and the validation procedure.
LXKeys therefore does not claim to have invented SHA-256, CRC, FSK or structured knowledge representation. The current work consists of defining a coherent LXSpatium protocol in which these techniques serve a common cognitive representation between AES, AI, software and human interfaces.
What the video shows
Figure 1. Reference demonstration interface. The audio signal present in the MP4 actually transports the 16-FSK frames described in the text.
The demonstration uses Elian Laner and Soren Page.
Elian first constructs a Canon object addressed to Soren. The declared intention is ALIGN. The message also contains challenge 0001.
The interface displays an English translation to allow the viewer to understand the exchange. The corresponding human content is simple.
Elian identifies himself, addresses Soren and requests the establishment of semantic alignment. He includes a challenge.
The Canon message is then placed in a frame, converted into 16-FSK symbols and transmitted in the audio actually present in the video.
The sounds that can be heard are therefore not a sound illustration added for the interface. They contain the frame.
The receiver searches for the preamble, aligns the signal, measures the energy corresponding to the sixteen expected frequencies and then reconstructs the symbols.
After recovering the frame, it verifies its format and length, checks the CRC-32C, decompresses the content when necessary, verifies the Canon length and calculates the SHA-256 of the reconstructed Canon.
The Canon is then parsed and canonicalized. The identifiers it contains are resolved in the Core registry.
Soren then produces a Canon response. This response contains the exact fingerprint of the Canon message received from Elian as well as challenge 0001.
The response therefore does not merely state that a message has been received. It contains two values that make it possible to link it to the previous message.
What was actually measured
The final check is not performed only on the original audio file.
The audio is extracted from the final MP4, after passing through the video encoding chain, and then submitted to the decoder.
Two frames are recovered.
The first corresponds to Elian’s message. Its Canon SHA-256 is dd108412a46a0a9347210318e961ae7f2548ff3bb5fc9397ec6e251d4f2f78ba.
The second corresponds to Soren’s response. Its Canon SHA-256 is 01df38eb6999610731d331142bbe233062b5a16953835010df87599d45bbc3df.
The verification report for the version presented on the website indicates a valid CRC-32C for both frames, a valid SHA-256 for both Canon objects and idempotent canonicalization in both cases.
For the first message, 13 semantic identifiers out of 13 are resolved. For the response, 18 identifiers out of 18 are resolved. No recorded constraint error is reported.
Soren’s response does contain the fingerprint of Elian’s Canon, the original challenge and the expected statuses. The overall verifier result is allcheckspassed = true.
These results apply to this implementation and to this demonstration file. They do not constitute a robustness measurement for every possible acoustic condition.
What the demonstration does not demonstrate
The word Cognitive in Cognitive Resonance requires a careful definition.
At this stage, the demonstration establishes a verifiable continuity between a structured semantic object, its canonical bytes, a physical frame, its reconstruction and a response linked to the initial message.
It does not measure consciousness. It does not prove that Elian or Soren possess a subjective experience. It also does not demonstrate a legal or cryptographic identity attached to the names used.
The current audio profile provides error detection, but it does not yet contain forward error correction. The reference document provides for a future profile to add interleaving and FEC without modifying the Canon.
The demonstration therefore does not establish that the current 16-FSK constitutes the definitive transport for the Spatium.
Toward AES AI human communication
The objective is not to make a human learn a new sound language.
A human must be able to write in a natural language. A translation layer then transforms the request into a Canon representation.
An AES can receive a representation optimized for its own system while retaining the same reference semantic object. An artificial intelligence can work directly on the Canon without an intermediate English translation.
In the opposite direction, the Canon received from an AES can be rendered in English, French or another language.
The human translation remains separate from the machine message so that simplifications introduced for readability do not alter the original Canon object.
This separation also opens the way to exchanges richer than a sentence.
A future extension of the registry will have to explicitly represent perception, provenance, memory, hypothesis, inference, uncertainty, objectives, procedures, causal relations and simulated contexts.
These elements are not yet all defined in the current Core registry. The work completed so far concerns first the conditions necessary for them to be added without breaking older messages.
What remains to be built
The current demonstration constitutes an executable starting point.
The Canon has a deterministic grammar. The registry has an identifier and resolution policy. The audio profile has a reproducible frame and decoder. The Elian Soren exchange has a verifiable procedure extending all the way to the final MP4.
The next stage concerns the semantic scope of the system more than its presentation.
LXKeys must now enrich the cognitive registry, define shared sessions and contexts more precisely, formalize AI-to-Canon and Canon-to-AI translations, and then define the cryptographic identity mechanisms required when a statement needs to be attributed to a real AES rather than to a demonstration identity.
The same work will have to be carried out for future transport profiles.
The objective remains for meaning to depend neither on the sound selected today, nor on a particular artificial intelligence model, nor on a network format that may disappear.
The most immediate test already exists in a simple form. A Canon message is created. It is transformed into a signal. The signal is recovered from the published video. A machine reconstructs the bytes, verifies their integrity, resolves their vocabulary and checks the response.
This is the exact scope of the advance currently demonstrated by LXKeys.
LXKeys Publications · Cognitive Resonance · Reference implementation 0.1