Distributed ledger-based cognitive inference model feature evaluation method, device, equipment, medium and system

Abstract

The invention provides a cognitive inference model feature evaluation method, device, equipment, medium and system based on a distributed account book. Obtaining a target cognitive inference model, and mapping and packaging the target cognitive inference model into an on-chain digital certificate entity in a distributed ledger network; synchronizing execution characteristic data of the target cognitive inference model in the operation process to the distributed ledger network by using an oracle assembly; using an evaluation contract operator deployed in the distributed account book network to call a preset dynamic evaluation algorithm to process the execution feature data so as to generate a feature evaluation score for the on-chain digital certificate entity; and in response to an ownership state transition trigger signal, executing numerical value rewriting on an ownership mapping parameter based on the feature evaluation score by using an intelligent contract so as to complete state transition processing for the target cognitive inference model. The technical problem that execution features and asset state updating are disjointed in a traditional evaluation scheme is relieved.

Description

Technical Field

[0001] This application relates to the fields of distributed ledger and artificial intelligence technology, and more specifically, to a method, apparatus, device, medium and system for evaluating the features of a cognitive reasoning model based on distributed ledger. Background Technology

[0002] With the popularization of artificial intelligence technology, various cognitive reasoning models (such as industry expert models and strategy generation models) have become core digital assets. In cross-platform and cross-institutional intelligent agent collaboration scenarios, due to the highly dynamic and evolving value of cognitive reasoning models, how to perform a high-fidelity quantitative assessment of their contribution to actual tasks and thereby achieve the synchronous transfer of ownership status has become a key requirement for the development of distributed cognitive scenarios.

[0003] In existing cognitive model management solutions, a centralized database is typically used to record model call logs and performance data. This solution first assigns a unique asset identifier to each model on a centralized server and establishes an associated performance statistics table. Then, when a model is called externally, the centralized gateway intercepts the call request and records the execution result. Finally, based on periodically summarized statistical reports, administrators manually update the model's reputation score or ownership information in the database.

[0004] However, this centralized approach has significant technical drawbacks. Because call logs are stored on a single entity's private server, the authenticity and integrity of execution characteristic data are difficult to verify across institutions and are susceptible to data tampering risks. Furthermore, since the contribution of cognitive models to collaborative tasks often exhibits complex non-linear characteristics, simple summary reports struggle to establish a real-time technical correlation between execution characteristics and asset value. This results in lags in updating ownership status, a lack of automated closed-loop verification mechanisms, and an inability to adapt to the high-frequency, dynamically evolving needs of cognitive asset valuation. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a method, apparatus, device, medium, and system for evaluating the features of a cognitive reasoning model based on a distributed ledger, thereby at least alleviating the aforementioned technical problems.

[0006] A feature evaluation method for a cognitive reasoning model based on distributed ledger, comprising: Step 1: Obtain the target cognitive reasoning model, map and encapsulate it as an on-chain digital credential entity in the distributed ledger network; use the oracle component to synchronize the execution feature data of the target cognitive reasoning model during operation to the distributed ledger network; Step 2: Using the evaluation contract operator deployed in the distributed ledger network, a preset dynamic evaluation algorithm is invoked to process the execution feature data in order to generate a feature evaluation score for the on-chain digital certificate entity. Step 3: Monitor the ownership state transition trigger signal in the distributed ledger network. In response to the ownership state transition trigger signal, use a smart contract to retrieve the ownership mapping parameters associated with the on-chain digital certificate entity, and perform numerical rewriting on the ownership mapping parameters based on the feature evaluation score to complete the state transition processing for the target cognitive reasoning model.

[0007] Optionally, the execution feature data is collected from the off-chain environment through the oracle component, and includes the behavioral trajectory components of the target cognitive reasoning model being called by the external agent program, the execution performance indicators for the preset task, and the associated environmental feedback parameters.

[0008] Optionally, step 2 specifically includes: using the evaluation contract operator based on marginal contribution attribution analysis logic to perform association mapping on the behavioral trajectory components, the execution efficiency index, and the environmental feedback parameters to determine the feature association contribution weight of the target cognitive reasoning model in a multi-model collaborative scenario; and performing weighted calculation by combining the initial attribute score of the target cognitive reasoning model with the feature association contribution weight to generate the feature evaluation score.

[0009] Optionally, mapping and encapsulating it into an on-chain digital certificate entity in a distributed ledger network specifically includes: extracting the unique identifier fingerprint and model metadata of the target cognitive reasoning model; and using a non-fungible token protocol to solidify the unique identifier fingerprint and the model metadata in the distributed ledger network to generate the corresponding on-chain digital certificate entity.

[0010] Optionally, the step of synchronizing the execution feature data of the target cognitive reasoning model during operation to the distributed ledger network using the oracle component specifically includes: using multiple independent verification nodes in the oracle component to obtain the original running records of the target cognitive reasoning model from multiple off-chain data sources; performing multi-party consistency verification on the original running records to generate the execution feature data with a trusted evidence identifier.

[0011] Optionally, step 2 further includes: obtaining historical evaluation records stored in the distributed ledger network and retrieving a preset time decay operator; performing a weighted decay mapping on the historical evaluation records using the time decay operator to generate effective feature components that characterize the logical evolution state of the target cognitive reasoning model, and correcting the feature evaluation score based on the effective feature components.

[0012] Optionally, before synchronizing the execution feature data of the target cognitive reasoning model during operation to the distributed ledger network using the oracle component, the method further includes: performing a privacy confidence transformation on the original message during operation using a zero-knowledge proof protocol to generate a de-identified execution component that does not contain model logic details, and then using the de-identified execution component as the execution feature data to be synchronized by the oracle component.

[0013] Optionally, in step 3, monitoring the ownership state transition trigger signal in the distributed ledger network includes: performing anomaly detection on the fluctuation slope of the feature evaluation score using a preset behavior monitoring contract; and generating the ownership state transition trigger signal representing the locked state in response to the fluctuation slope exceeding a preset threshold.

[0014] A feature evaluation device for a cognitive reasoning model based on a distributed ledger, comprising: The credential mapping module is used to obtain the target cognitive reasoning model, map it, and encapsulate it into an on-chain digital credential entity in the distributed ledger network. The oracle synchronization module is used to synchronize the execution feature data of the target cognitive reasoning model during its operation to the distributed ledger network using oracle components; The score evaluation module is used to calculate and generate feature evaluation scores for the on-chain digital certificate entities using evaluation contract operators deployed in the distributed ledger network. The state transition module is used to monitor the ownership state transition trigger signal and, in response to the ownership state transition trigger signal, use a smart contract to perform a numerical rewrite of the ownership mapping parameters associated with the on-chain digital certificate entity based on the feature evaluation score.

[0015] An electronic device includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the program to implement the steps of the method as described in any one of the embodiments of this application.

[0016] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method as described in any of the embodiments of the present application.

[0017] A feature evaluation system for a cognitive reasoning model based on a distributed ledger, comprising: The cognitive reasoning model layer is used to run the target cognitive reasoning model and generate running records. An oracle gateway is used to transform the execution records into trusted execution characteristic data using oracle components and synchronize them to the distributed ledger network. A distributed ledger network is used to maintain the on-chain digital credential entities generated by the target cognitive reasoning model mapping, and to deploy evaluation contract operators and smart contracts; The evaluation contract operator is used to calculate and generate a feature evaluation score based on the execution feature data; the smart contract is used to monitor the ownership state migration trigger signal and perform rewriting processing on the ownership mapping parameters associated with the on-chain digital certificate entity based on the feature evaluation score.

[0018] This application presents a feature evaluation scheme for cognitive reasoning models based on distributed ledgers. Addressing the technical shortcomings of traditional centralized evaluation schemes, such as the susceptibility of execution feature data to tampering, delayed updates to ownership status, and a lack of credibility, this scheme maps the target cognitive reasoning model to an on-chain digital certificate entity within a distributed ledger network and utilizes an oracle component to achieve trusted synchronization of off-chain execution feature data to on-chain data. This process alleviates the problem of the disconnect between physical execution data and asset certificates in traditional schemes. Compared to centralized log recording, this application leverages the immutability of distributed ledgers and the multi-party verification mechanism of oracles to ensure the objectivity and authenticity of the evaluation basis.

[0019] The evaluation contract operator deployed in the distributed ledger network quantifies the execution feature data to generate feature evaluation scores, and drives the smart contract to rewrite the ownership mapping parameters numerically, realizing an automated closed loop for cognitive asset evaluation and value transfer. Compared with the traditional manual update mode, this application establishes a real-time logical connection between execution features and ownership status through smart contracts, improving the real-time performance and logical rigor of cognitive asset state transitions. In addition, by combining marginal contribution attribution analysis logic and anomaly detection mechanism, it ensures that in complex environments with multi-model collaboration, the individual contribution of a single model can be accurately identified and quantified, thereby supporting the accurate pricing and efficient transfer of cognitive ability assets in decentralized networks. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a feature evaluation method for a cognitive reasoning model based on a distributed ledger, as described in an embodiment of this application. Figure 2 This is a structural block diagram of a cognitive reasoning model feature evaluation device based on distributed ledger, according to an embodiment of this application. Figure 3 This is a schematic diagram of an electronic device structure according to an embodiment of this application; Figure 4 This is an architecture diagram of a cognitive reasoning model feature evaluation system based on distributed ledger, according to an embodiment of this application. Detailed Implementation

[0021] like Figure 1As shown in the figure, this application provides an embodiment of a method for evaluating the features of a cognitive reasoning model based on a distributed ledger, comprising: Step 1: Obtain the target cognitive reasoning model, map and encapsulate it as an on-chain digital credential entity in the distributed ledger network; use the oracle component to synchronize the execution feature data of the target cognitive reasoning model during operation to the distributed ledger network; Step 2: Using the evaluation contract operator deployed in the distributed ledger network, a preset dynamic evaluation algorithm is invoked to process the execution feature data in order to generate a feature evaluation score for the on-chain digital certificate entity. Step 3: Monitor the ownership state transition trigger signal in the distributed ledger network. In response to the ownership state transition trigger signal, use a smart contract to retrieve the ownership mapping parameters associated with the on-chain digital certificate entity, and perform numerical rewriting on the ownership mapping parameters based on the feature evaluation score to complete the state transition processing for the target cognitive reasoning model.

[0022] Optionally, the execution feature data is collected from the off-chain environment through the oracle component, and includes the behavioral trajectory components of the target cognitive reasoning model being called by the external agent program, the execution performance indicators for the preset task, and the associated environmental feedback parameters.

[0023] Preferably, the specific technical implementation of obtaining the execution feature data using the oracle component in step 1 is as follows: Using a designed sequence feature parsing component, the original inference messages generated by the target cognitive reasoning model in an off-chain execution environment (such as an automated financial trading simulation system or a complex multi-agent collaborative game platform) are obtained. The sequence feature parsing component performs instruction flow reconstruction and nonlinear feature extraction on the original inference messages, identifying the decision evolution path, neuron activation distribution, and inference response latency faced by the target cognitive reasoning model, thereby generating the behavioral trajectory components. This processing action realizes the digital reconstruction of the model's decision-making "thought process," providing underlying behavioral data support for subsequent quantification of the model's logical robustness.

[0024] Preferably, the specific implementation process of step 1 is as follows: Using the designed performance status quantification unit, the execution result data associated with the target cognitive reasoning model for a preset task (such as a risk hedging task or a complex path planning task within a demand period) is obtained. The performance status quantification unit maps the execution result data with the benchmark target feature execution comparison stored in the local cache, and calculates quantified components including task achievement rate, decision accuracy deviation, and unit computing power contribution value to generate the execution performance index. The generated execution performance index directly characterizes the physical utility of the target cognitive reasoning model at the functional implementation level, alleviating the technical deficiency in traditional solutions that cannot accurately quantify the actual output of the black-box model.

[0025] Preferably, in one scenario, step 1 is specifically implemented as follows: An environmental perception and recognition module is used to capture external feedback signals (such as market instantaneous volatility index, environmental noise intensity, or adversarial interference parameters) from the off-chain environment in real time. The environmental perception and recognition module uses a preset spatiotemporal anchoring algorithm to perform time alignment and spatial topological association on the external feedback signals, determining the external constraint background of the target cognitive reasoning model when performing decision-making actions, and producing the environmental feedback parameters characterizing the difficulty of the decision. This processing ensures that the evaluation logic can perceive the dynamic changes in the decision-making environment, effectively avoiding inaccurate evaluation results due to ignoring environmental factors.

[0026] Preferably, the generated behavioral trajectory components, execution performance indicators, and environmental feedback parameters are aggregated into the trusted message encapsulation logic of the oracle component. The trusted message encapsulation logic uses a preset asymmetric encryption algorithm to perform digital signature and structured encapsulation processing on each of the above components, producing execution feature data with a globally unique timestamp and source verification identifier. The generated execution feature data is then synchronized to the distributed ledger network via the oracle component's transmission interface, serving as a key input for subsequently driving the allocation of execution weights for the evaluation contract operator.

[0027] Preferably, the multi-dimensional collection and processing logic for the behavioral trajectory components, execution performance indicators, and environmental feedback parameters alleviates the technical monitoring bottleneck caused by the highly indirect and dynamically evolving nature of cognitive asset value. Compared to the traditional single-result recording mode, this application constructs a comprehensive, multi-dimensional cognitive feature profile by introducing behavioral trajectories reflecting the reasoning path, performance indicators reflecting execution effectiveness, and environmental parameters reflecting external constraints. This off-chain data trusted anchoring mechanism based on oracle components ensures that the distributed ledger network can obtain a high-fidelity, tamper-proof, and logically deep execution foundation, thereby supporting accurate feature evaluation and automatic migration of ownership status for cross-platform cognitive assets.

[0028] Optionally, step 2 specifically includes: using the evaluation contract operator based on marginal contribution attribution analysis logic to perform association mapping on the behavioral trajectory components, the execution efficiency index, and the environmental feedback parameters to determine the feature association contribution weight of the target cognitive reasoning model in a multi-model collaborative scenario; and performing weighted calculation by combining the initial attribute score of the target cognitive reasoning model with the feature association contribution weight to generate the feature evaluation score.

[0029] Preferably, the specific implementation process of the association mapping using the evaluation contract operator based on marginal contribution attribution analysis logic in step 2 is as follows: The evaluation contract operator obtains the behavioral trajectory components, the execution efficiency index, and the environmental feedback parameters that are synchronized to the distributed ledger network. Using a designed marginal utility simulation unit, the participation state of the target cognitive reasoning model in multiple heterogeneous model combination subsets is simulated, and the marginal increment of the execution efficiency index caused by injecting the behavioral trajectory components into a set subset under the constraints of the environmental feedback parameters (such as preset task complexity or external interference intensity) is calculated. The evaluation contract operator performs a weighted average processing based on probability weights on the marginal increments generated under different subset combinations to eliminate nonlinear coupling interference between models, thereby determining the feature association contribution weight of the target cognitive reasoning model in a multi-model collaborative scenario. The generated feature association contribution weight, in the form of numerical weights, accurately characterizes the degree of independent technical contribution of the target cognitive reasoning model to the overall task objective, alleviating the technical problem of the difficulty in decomposing the contribution of a single model in a multi-model hybrid decision-making scenario.

[0030] Preferably, in the specific technical implementation of step 2: the feature-related contribution weights calculated above are obtained, and the initial attribute scores (e.g., static parameters representing the initial quality or historical benchmark capability of the model) of the target cognitive reasoning model are retrieved from the metadata of the on-chain digital certificate entity. Using a designed feature weighted fusion operator, the initial attribute scores and the feature-related contribution weights are mapped to a unified numerical dimension space. The feature weighted fusion operator performs a weighted calculation on the initial attribute scores representing the model's basic capability and the feature-related contribution weights representing its real-time contribution potential according to a preset weight allocation logic (e.g., using linear proportional synthesis or nonlinear normalization mapping), producing a physical scalar value representing the model's current comprehensive value state, thereby generating the feature evaluation score. This processing action realizes the logical integration from static attributes to dynamic performance, ensuring that the generated feature evaluation score can reflect the feature evolution state of the model in a real-time and objective manner in the actual execution environment.

[0031] Preferably, in one scenario, step 2 is specifically implemented as follows: When the evaluation contract operator executes the marginal contribution attribution analysis logic, it uses a built-in contribution smoothing component to perform temporal filtering on the generated feature-related contribution weights. The contribution smoothing component obtains the feature-related contribution weights generated in the current period and historical weight records stored in the distributed state database. By performing exponential smoothing or moving average mapping, it weakens the weight jumps caused by instantaneous environmental fluctuations, generating corrected feature-related contribution weights. Subsequently, the corrected feature-related contribution weights are applied to the initial attribute score using the aforementioned feature weighted fusion operator. This processing action, by introducing temporal stability constraints, makes the generated feature evaluation score highly robust, effectively avoiding abnormal fluctuations in model weight parameters caused by short-term noise in the off-chain environment.

[0032] Preferably, the above-described process for generating the feature evaluation score alleviates the technical shortcomings of coarse-grained quantification of cognitive assets in complex collaborative tasks and the lack of causal basis. By introducing attribution analysis logic based on marginal contribution, this application can extract the pure contribution increment of the target model relative to the environmental background and other models from multi-dimensional execution data. Compared with the traditional global apportionment evaluation mode, this application establishes a deterministic logical chain from the underlying execution trajectory to the high-level ownership parameters through the underlying operation of the evaluation contract operator. This evaluation mechanism based on individual marginal increments ensures that the feature evaluation score can accurately capture the logical advantages of the cognitive reasoning model, providing credible physical evidence for the subsequent automated resource state migration in the distributed ledger network.

[0033] Preferably, the generated feature evaluation score is physically written to a temporary state buffer in the distributed ledger network. Upon detecting an update in the feature evaluation score, subsequent rewriting logic for the ownership mapping parameters is triggered. Since the feature evaluation score is calculated based on the marginal contribution reflecting the actual behavioral trajectory of the model, it has extremely high technical penetration in representing the value of the target cognitive reasoning model. This deep feature mapping from "execution features" to "asset valuation" constructs a dynamically evolving credit game environment, enabling high-value model features to obtain corresponding resource weights in the distributed network, thereby supporting fair evaluation and efficient collaboration of cross-platform cognitive assets at the technical architecture level.

[0034] Optionally, mapping and encapsulating it into an on-chain digital certificate entity in a distributed ledger network specifically includes: extracting the unique identifier fingerprint and model metadata of the target cognitive reasoning model; and using a non-fungible token protocol to solidify the unique identifier fingerprint and the model metadata in the distributed ledger network to generate the corresponding on-chain digital certificate entity.

[0035] Preferably, the specific technical implementation of mapping and encapsulating the target cognitive reasoning model into an on-chain digital credential entity in a distributed ledger network is as follows: Using a designed feature fingerprint extraction operator, the underlying code logic, neural network hierarchical architecture, weight distribution matrix, or operator execution path components of the target cognitive reasoning model are obtained. The feature fingerprint extraction operator uses a preset secure hash algorithm or irreversible feature mapping algorithm to perform fingerprint compression processing on the above-mentioned physical elements of the model, producing a globally unique and collision-resistant physical identifier component, thereby determining the unique identifier fingerprint. Simultaneously, using a designed metadata definition module, the development version, developer's encrypted signature, applicable task dimensions, initial performance calibration value, and preset logical description information of the target cognitive reasoning model are obtained to generate the model metadata. This processing action establishes the physical uniqueness and logical traceability of the cognitive reasoning model in the digital mapping process.

[0036] Preferably, in the specific technical implementation of step 1: the unique identifier fingerprint and the model metadata produced above are obtained, and they are aggregated into a cognitive asset feature package to be solidified. Using the designed token protocol adaptation unit, a pre-defined non-fungible token protocol interface (e.g., a digital credential standard protocol conforming to adaptation requirements) in the distributed ledger network is invoked. The token protocol adaptation unit maps each component in the cognitive asset feature package to the metadata storage field, extended attribute description domain, or external resource location handle of the token contract, and performs an immutable notarization record of the unique identifier fingerprint and the model metadata by initiating an on-chain state change request, thereby performing a solidification action for the aforementioned model features in the distributed ledger network. In this processing action, through a distributed consensus mechanism, a strong association mapping is performed between the logical identity of the cognitive model and the unique address on the ledger, eliminating the technical risk of the digital model identity being easily cloned or impersonated in traditional solutions.

[0037] Preferably, in one scenario, step 1 is specifically implemented as follows: Using the designed digital certificate instantiation logic, after the solidification action is completed on the distributed ledger network, a contract interaction receipt fed back by the smart contract and its corresponding asset mapping identifier are obtained. Based on the contract interaction receipt, the digital certificate instantiation logic activates a state object with asset-based interaction attributes for the target cognitive reasoning model in the ledger topology space, thereby generating the on-chain digital certificate entity. The generated on-chain digital certificate entity not only serves as a digital carrier of the cognitive reasoning model on the blockchain but also provides a unique logical anchor point for the execution feature data obtained in subsequent steps and the generated feature evaluation score through a built-in state index mechanism. This technical encapsulation mechanism based on feature fingerprints and metadata solidification alleviates the ownership confirmation problem of complex, dynamically evolving digital assets like cognitive models during multi-platform flow, providing a definite physical foundation for building a credible feature evaluation system.

[0038] Preferably, the above-described generation logic for the on-chain digital certificate entity, compared to a simple document storage scheme, constructs a digital mapping relationship with a "technical core" by deeply integrating the underlying logical fingerprint of the model with the metadata describing the model's capabilities and mapping it to the token protocol. This fine-grained encapsulation ensures that even when the model logic undergoes minor iterations, its physical continuity can still be identified through the unique identifier fingerprint. Simultaneously, the task dimensions and initial performance benchmarks defined in the model metadata provide an original reference system for subsequent marginal contribution attribution analysis, giving the generated feature evaluation scores causal logic support from the inside out, greatly enhancing the overall solution's technical sensitivity when dealing with heterogeneous and dynamic cognitive assets.

[0039] Preferably, the generated on-chain digital certificate entity is physically stored in the state layer of the distributed ledger network. Upon detecting the instantiation state of the on-chain digital certificate entity, the aforementioned oracle component synchronization process is automatically triggered. Since the on-chain digital certificate entity is anchored with a unique fingerprint of its respective model, the oracle component can accurately identify and collect the corresponding execution records in the off-chain environment using this fingerprint, producing highly directional execution characteristic data. This logical cascading from "digital certificate encapsulation" to "off-chain feature collection" ensures the consistency of characteristics and ownership of cognitive assets throughout their entire lifecycle, laying the foundation for automated resource state migration processing.

[0040] Optionally, the step of synchronizing the execution feature data of the target cognitive reasoning model during operation to the distributed ledger network using the oracle component specifically includes: using multiple independent verification nodes in the oracle component to obtain the original running records of the target cognitive reasoning model from multiple off-chain data sources; performing multi-party consistency verification on the original running records to generate the execution feature data with a trusted evidence identifier.

[0041] Preferably, in step 1, the specific implementation of obtaining raw running records from multiple off-chain data sources using multiple independent verification nodes in the oracle component involves: utilizing a distributed feature extraction operator deployed in the oracle component to establish feature connections with off-chain data sources deployed in different physical environments or management domains via an encrypted remote procedure call protocol. The distributed feature extraction operator drives each associated independent verification node to send parallel instructions to its associated off-chain data source for extracting running logs of the target cognitive reasoning model within its adaptation requirement execution cycle. Upon receiving the data message from the corresponding off-chain data source, each independent verification node performs preliminary format alignment and feature extraction operations, thereby producing the raw running records characterizing the model's execution path and performance indicators. This processing action, through a multi-point distributed acquisition strategy, ensures the physical diversity of data acquisition paths, laying the foundation for subsequent credible verification.

[0042] Preferably, in step 1, the specific technical implementation of performing multi-party consistency verification on the original running records involves: obtaining the original running records produced by the aforementioned independent verification nodes; using a designed multi-party consistency consensus component, performing cross-comparison mapping on the decision hash value, input / output components, and timestamp information contained in each original running record; the multi-party consistency consensus component using a preset threshold signature logic or majority voting decision mechanism to determine whether the original running records provided by different nodes fall within a preset deviation tolerance range; and confirming the credibility of the original running records upon detecting that the number of records reaching consensus meets a preset proportion (e.g., more than two-thirds of the verification nodes report consistency). During this process, a designed abnormal data removal operator is used to identify and physically shield illegal record components with numerical jumps or signature verification failures, ensuring that all data ultimately used for evidence storage has undergone physical endorsement from multiple parties.

[0043] Preferably, in one scenario, the specific implementation of generating the execution feature data with a trusted evidence identifier in step 1 involves: obtaining the original running record after passing the aforementioned multi-party consistency verification; using a designed evidence anchoring operator, concatenating and encapsulating the original running record with the unique identifier fingerprint of the target cognitive reasoning model and the aggregate signature of the independent verification node executing the verification task; the evidence anchoring operator using a preset one-way hash mapping function to generate an electronic fingerprint component with a globally unique index attribute for the encapsulated data packet, thereby determining the trusted evidence identifier; subsequently, using the evidence anchoring operator to map the data packet carrying the trusted evidence identifier into the execution feature data, and pushing it to the distributed ledger network through an asynchronous write interface; the generated execution feature data constructs a trusted bridge connecting the off-chain physical execution trajectory and the on-chain digital asset state, alleviating the problem of data authenticity doubt caused by the black box nature of the cognitive model execution environment.

[0044] Preferably, the above-described logic for collecting, verifying, and storing the original execution records alleviates the technical problem of easily manipulated or tampered execution evidence during the dynamic evolution of cross-platform cognitive assets. Compared to traditional single oracle node forwarding schemes, this application introduces a multi-party game and consistency verification mechanism with the independent verification nodes, enabling the generated execution feature data to possess extremely high technical certainty and non-repudiation. This evidence storage method based on multi-source heterogeneous record alignment ensures that even if some off-chain data sources or individual independent verification nodes experience operational anomalies, accurate credible evidence identifiers can still be produced through the consensus mechanism. The generated execution feature data, through logical association with the on-chain digital certificate entity, provides credible physical evidence for subsequent feature evaluation based on marginal contribution attribution analysis logic, enhancing the overall evaluation system's resistance to attacks and operational robustness.

[0045] Optionally, step 2 further includes: obtaining historical evaluation records stored in the distributed ledger network and retrieving a preset time decay operator; performing a weighted decay mapping on the historical evaluation records using the time decay operator to generate effective feature components that characterize the logical evolution state of the target cognitive reasoning model, and correcting the feature evaluation score based on the effective feature components.

[0046] Preferably, the specific implementation process for obtaining the historical evaluation record in step 2 is as follows: Using the designed historical feature extraction module, the asymmetric storage address associated with the on-chain digital certificate entity is located through the state query interface in the distributed ledger network. Within this storage address space, the historical feature extraction module extracts the notarized score records generated by the target cognitive reasoning model within a preset historical period (e.g., the past thirty natural days or five hundred block heights) according to the time series or block height series, thereby generating the historical evaluation record. This processing action realizes the physical retrieval of the model's value evolution trajectory, providing objective data samples for subsequent cross-time series stability analysis.

[0047] Preferably, in the specific technical implementation of step 2: the historical evaluation records generated by the aforementioned processing actions are obtained, and the time decay operator (e.g., using a preset half-life decay logic or a linear decreasing weight model) is simultaneously obtained from the global parameter configuration table. Using the designed time feature weight generation unit, based on the time difference between the storage timestamp corresponding to each historical evaluation record and the current settlement time, the time decay operator is used to calculate the decay weight coefficient characterizing the effectiveness of each historical sample. The time feature weight generation unit physically applies the decay weight coefficient to the score component in the corresponding historical evaluation record and performs a weighted summation operation based on the time-series dimension to generate the effective feature component. The generated effective feature component, through logical weakening of outdated data, accurately maps the logical activity and decision reliability of the target cognitive reasoning model at the current moment.

[0048] Preferably, in one scenario, step 2 is specifically implemented as follows: The feature evaluation score of the initial state produced by the aforementioned marginal contribution attribution analysis logic is obtained, and the effective feature components are retrieved simultaneously. A preset temporal smoothing coefficient is obtained using a designed dynamic score smoothing operator. The dynamic score smoothing operator uses the temporal smoothing coefficient to perform a proportional linear synthesis of the feature evaluation score representing the current execution performance and the effective feature components representing historical cumulative performance, producing a target evaluation result constrained by temporal stability, thereby achieving a correction action on the feature evaluation score. This processing action ensures that the evaluation system can both capture the latest performance leaps of the model and prevent drastic fluctuations in evaluation caused by the randomness of a single execution through the constraint of historical data, thus improving the noise resistance of the feature evaluation score.

[0049] Preferably, the aforementioned time-series decay and correction logic for the historical evaluation records alleviates the technical problem of the difficulty in smoothing value fluctuations of cognitive assets during long-term evolution. Compared to traditional single-point-of-time evaluation schemes, this application constructs a feature evaluation framework with time memory and a focus on key aspects by introducing the time decay operator. This weighted decay mapping based on effectiveness weights ensures that the final generated feature evaluation scores objectively reflect the true state of the target cognitive reasoning model's evolution from occasional performance to sustained robust performance. The generated effective feature components are not merely a simple summation of historical performance, but a physical simulation of the model's value evolution law, ensuring that the evaluation results can serve as a highly reliable logical driver for numerical rewriting of ownership mapping parameters in the distributed ledger network.

[0050] Preferably, the corrected feature evaluation score is then transmitted to the subsequent ownership mapping parameter rewriting link. Utilizing the correction result generated by the dynamic score smoothing operator, the smart contract is driven to perform a numerical rewriting action on the ownership mapping parameters when it detects the ownership state transition trigger signal. Through this dynamic correction mechanism based on historical real-time performance, this application constructs a value assessment benchmark with fault tolerance and logical depth for distributed cognitive scenarios, effectively guiding cognitive asset holders to continuously optimize and iterate the target cognitive reasoning model, thereby promoting the healthy evolution of cognitive capabilities within the distributed ledger network ecosystem.

[0051] Optionally, before synchronizing the execution feature data of the target cognitive reasoning model during operation to the distributed ledger network using the oracle component, the method further includes: performing a privacy confidence transformation on the original message during operation using a zero-knowledge proof protocol to generate a de-identified execution component that does not contain model logic details, and then using the de-identified execution component as the execution feature data to be synchronized by the oracle component.

[0052] Preferably, in this scenario, the specific technical implementation of the privacy confidence transformation using the zero-knowledge proof protocol in step 1 is as follows: A privacy circuit construction operator is used to obtain the original message reflecting the internal logical jumps, parameter weight distribution, and intermediate layer output state of the target cognitive reasoning model during operation. For the original message, the privacy circuit construction operator uses the non-interactive zero-knowledge proof architecture in the zero-knowledge proof protocol (e.g., using a zero-knowledge verification operator logic adapted to specific needs) to map the private logical details in the original message into encrypted witness components, and simultaneously generates a validity proof trajectory representing the conformity of the aforementioned private logical details with preset execution rules, thereby producing the desensitized execution component. This processing action physically achieves low-level isolation of the model's intellectual property rights, ensuring that execution evidence with high confidence can still be provided to external systems without revealing the internal black-box logic.

[0053] Preferably, in the specific technical implementation of step 1: the de-identified execution component produced by the aforementioned privacy confidence transformation process is obtained. Using the designed data attribute encapsulation unit, the de-identified execution component undergoes metadata mapping and structure alignment based on on-chain data specifications to generate a trusted data carrier representing the de-identified execution state, and this carrier is used as the execution feature data. Subsequently, the data attribute encapsulation unit physically transfers the execution feature data to the oracle component. After obtaining the execution feature data, the oracle component initiates the aforementioned distributed synchronization process to synchronize it to the distributed ledger network. During this process, the de-identified execution component serves as the sole data source for subsequent evaluation in the consensus phase, alleviating the technical problem of protecting the core algorithmic logic of cognitive assets in an open distributed system.

[0054] Preferably, the aforementioned privacy desensitization and feature synchronization logic for the original message alleviates the technical challenge of balancing privacy security and audit transparency during the cross-platform transfer of cognitive assets. Compared to the conventional approach of directly transmitting the original execution log, this application constructs a cognitive feature filtering layer with privacy protection attributes by introducing the zero-knowledge proof protocol. This evidence storage mechanism based on the desensitized execution components enables the distributed ledger network to verify that the model has indeed executed the predetermined logic and produced the corresponding performance results, without needing to analyze sensitive parameters within the model. The generated execution feature data, having eliminated sensitive and redundant information that could lead to logical reverse engineering, greatly enhances the security and logical robustness of the target cognitive reasoning model in heterogeneous environments, providing a pre-processing guarantee that balances data security and technical depth for building a credible, decentralized feature evaluation system.

[0055] Optionally, in step 3, monitoring the ownership state transition trigger signal in the distributed ledger network includes: performing anomaly detection on the fluctuation slope of the feature evaluation score using a preset behavior monitoring contract; and generating the ownership state transition trigger signal representing the locked state in response to the fluctuation slope exceeding a preset threshold.

[0056] Preferably, the specific technical implementation of step 3, which utilizes a preset behavior monitoring contract to perform anomaly detection on the fluctuation slope of the feature evaluation score, is as follows: The feature evaluation score generated by the evaluation contract operator in the current period is obtained, and the historical score component corresponding to the previous period is simultaneously retrieved from the state database of the distributed ledger network. Using the designed numerical fluctuation slope operator, a numerical difference operation is performed between the feature evaluation score and the historical score component, and a division mapping process is performed in conjunction with the time step of the current collection period to produce a physical scalar value representing the rate of score evolution, thereby determining the fluctuation slope. This processing action realizes the digital quantification of the dimension of cognitive asset value change, providing a real-time numerical criterion for identifying abnormal disturbances in the evaluation logic.

[0057] Preferably, in the specific technical implementation of step 3: the volatility slope calculated above is obtained, and the corresponding preset threshold (e.g., a safe numerical boundary determined based on the historical volatility standard deviation of the domain to which the cognitive reasoning model belongs) is retrieved from the contract storage space. Using the comparison and judgment logic in the designed behavior monitoring contract, the volatility slope and the preset threshold are compared and mapped based on amplitude. In response to detecting that the volatility slope exceeds the preset threshold (e.g., due to adversarial interference or logical collapse causing a sudden change in the evaluation value), the comparison and judgment logic outputs a logical judgment result representing the abnormal behavior. In this process, the behavior monitoring contract constructs an automatic defense barrier against the cognitive asset credit assessment system by dynamically monitoring the slope of the evaluation trajectory.

[0058] Preferably, in one scenario, step 3 is specifically implemented as follows: The logical judgment result produced above is obtained. Using a designed signal logic generation component, in response to the logical judgment result, the ownership state transition trigger signal, containing a pending lock state identifier and an anomaly feature fingerprint, is encapsulated and generated. The signal logic generation component pushes the generated ownership state transition trigger signal to the instruction listening queue of the smart contract. After obtaining the ownership state transition trigger signal, the smart contract initiates a preset security redundancy processing procedure, performs write protection processing on the ownership mapping parameters associated with the on-chain digital certificate entity, and synchronously interrupts the corresponding resource ownership update action to complete the locking of the target cognitive reasoning model.

[0059] Preferably, the aforementioned fluctuation slope detection and locking trigger mechanism for the feature evaluation score alleviates the technical problem of the lack of automated defense capabilities in the cognitive model evaluation system when facing malicious attacks or sudden changes in the operating environment. Compared with traditional offline auditing solutions, this application achieves physical parallel processing of evaluation result generation and anomaly detection by integrating the monitoring operator into the contract execution link of the distributed ledger network. This technique based on numerical evolution slope ensures that every change in the ownership mapping parameter is within the preset behavioral safety boundary. Since the generated ownership state migration trigger signal is produced based on rigorous numerical comparison logic, its intervention in ownership state has extremely high technical certainty, thereby supporting the logical robustness and ownership security of cross-platform cognitive assets in a decentralized game environment.

[0060] Preferably, the locked on-chain digital certificate entity will enter a verification buffer. The ownership state transition trigger signal and associated feature evaluation score in the verification buffer are obtained, triggering the aforementioned oracle component to initiate deep verification logic. This involves introducing higher-dimensional off-chain execution feature data to verify the rationality of the volatility slope. This logic degradation and deep verification mechanism based on anomaly detection triggers not only prevents malicious manipulation of asset value but also provides technically oriented negative samples for subsequent optimization of the dynamic evaluation algorithm's parameter strategies through source analysis of anomaly features, ensuring the continuous evolution and logical self-correction of the cognitive asset evaluation system.

[0061] like Figure 2 As shown, this embodiment of the present application provides a feature evaluation device for a cognitive reasoning model based on a distributed ledger, comprising: The credential mapping module is used to obtain the target cognitive reasoning model, map it, and encapsulate it into an on-chain digital credential entity in the distributed ledger network. The oracle synchronization module is used to synchronize the execution feature data of the target cognitive reasoning model during its operation to the distributed ledger network using oracle components; The score evaluation module is used to calculate and generate feature evaluation scores for the on-chain digital certificate entities using evaluation contract operators deployed in the distributed ledger network. The state transition module is used to monitor the ownership state transition trigger signal and, in response to the ownership state transition trigger signal, use a smart contract to perform a numerical rewrite of the ownership mapping parameters associated with the on-chain digital certificate entity based on the feature evaluation score.

[0062] like Figure 3 The image shows an electronic device including a processor and a memory; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the steps of the method as described above.

[0063] like Figure 4 The figure shows a feature evaluation system for a cognitive reasoning model based on a distributed ledger, which includes: The cognitive reasoning model layer is used to run the target cognitive reasoning model and generate running records. An oracle gateway is used to transform the execution records into trusted execution characteristic data using oracle components and synchronize them to the distributed ledger network. A distributed ledger network is used to maintain the on-chain digital credential entities generated by the target cognitive reasoning model mapping, and to deploy evaluation contract operators and smart contracts; The evaluation contract operator is used to calculate and generate a feature evaluation score based on the execution feature data; the smart contract is used to monitor the ownership state migration trigger signal and perform rewriting processing on the ownership mapping parameters associated with the on-chain digital certificate entity based on the feature evaluation score.

[0064] Preferably, this application provides a cognitive reasoning model feature evaluation system that focuses on hardware architecture implementation. The system includes: Physical sensing message access front end: corresponding Figure 4 The oracle gateway in the system is visualized as a hardware access layer consisting of a multi-source serial communication bus and a Trusted Execution Environment (TEE) secure memory. The access front end acquires off-chain execution records and performs multi-party signature verification in the TEE environment, producing encrypted execution feature data packets.

[0065] Distributed ledger consensus acceleration engine: corresponding Figure 4 The underlying hardware of the distributed ledger network consists of a high-speed hash array and a distributed state database. The engine receives the encrypted messages through a physical high-speed backplane bus, performs consensus verification using the hash array, and stores the mapped on-chain digital credential entity in a designated address space of the state database.

[0066] Contract instruction execution unit: corresponding Figure 4 The smart contract engine and evaluation contract operator are visualized as a programmable gate array (FPGA) or a dedicated smart contract processing core. The execution unit retrieves execution feature data from the state database, performs marginal contribution attribution analysis in the instruction pipeline, and produces corresponding feature evaluation scores. In response to the trigger pulse (i.e., the ownership state transition trigger signal) captured by the hardware monitoring circuit, the execution unit performs numerical rewriting on the memory unit corresponding to the ownership mapping parameters, thereby physically completing the state transition of resource ownership.

[0067] The aforementioned hardware entities achieve data interconnection through a system-level backplane switching bus. By hardware coupling between FPGA operators adapted to specific needs and secure memory, a complete physical closed loop from off-chain message access to on-chain state rewriting is achieved, alleviating consensus latency and data processing bottlenecks present in pure software implementations, and improving the timing determinism and resistance to physical attacks of the cognitive asset valuation system.

[0068] The above Figures 2-4 For an exemplary description, please refer to the above. Figure 1 .

Claims

1. A method for evaluating the features of a cognitive reasoning model based on a distributed ledger, characterized in that, Includes the following steps: Step 1: Obtain the target cognitive reasoning model, map and encapsulate it as an on-chain digital credential entity in the distributed ledger network; use the oracle component to synchronize the execution feature data of the target cognitive reasoning model during operation to the distributed ledger network; Step 2: Using the evaluation contract operator deployed in the distributed ledger network, a preset dynamic evaluation algorithm is invoked to process the execution feature data in order to generate a feature evaluation score for the on-chain digital certificate entity. Step 3: Monitor the ownership state transition trigger signal in the distributed ledger network. In response to the ownership state transition trigger signal, use a smart contract to retrieve the ownership mapping parameters associated with the on-chain digital certificate entity, and perform numerical rewriting on the ownership mapping parameters based on the feature evaluation score to complete the state transition processing for the target cognitive reasoning model.

2. The feature evaluation method for a cognitive reasoning model based on distributed ledger as described in claim 1, characterized in that, The execution characteristic data is collected from the off-chain environment through the oracle component, and includes the behavioral trajectory components of the target cognitive reasoning model being called by the external agent program, the execution performance indicators for the preset task, and the associated environmental feedback parameters.

3. The feature evaluation method for a cognitive reasoning model based on distributed ledger as described in claim 1, characterized in that, Step 2 specifically includes: The evaluation contract operator based on marginal contribution attribution analysis logic is used to perform correlation mapping on the behavioral trajectory components, the execution efficiency index, and the environmental feedback parameters to determine the feature correlation contribution weight of the target cognitive reasoning model in a multi-model collaborative scenario. The initial attribute scores of the target cognitive reasoning model are combined with the feature association contribution weights to perform a weighted calculation to generate the feature evaluation score.

4. The feature evaluation method for a cognitive reasoning model based on distributed ledger as described in claim 1, characterized in that, The mapping and encapsulation of this data into an on-chain digital credential entity in a distributed ledger network specifically includes: Extract the unique identifier fingerprint and model metadata of the target cognitive reasoning model; By using a non-fungible token protocol, the unique identifier fingerprint and the model metadata are embedded in the distributed ledger network to generate the corresponding on-chain digital certificate entity.

5. The feature evaluation method for a cognitive reasoning model based on distributed ledger as described in claim 1, characterized in that, The step of synchronizing the execution feature data of the target cognitive reasoning model during its operation to the distributed ledger network using an oracle component specifically includes: Using multiple independent verification nodes in the oracle component, the original running records for the target cognitive reasoning model are obtained from multiple off-chain data sources respectively; Perform multi-party consistency verification on the original execution record to generate the execution feature data with a trusted evidence identifier.

6. The feature evaluation method for a cognitive reasoning model based on distributed ledger according to claim 1, characterized in that, Step 2 also includes: Obtain historical evaluation records stored in the distributed ledger network and retrieve a preset time decay operator; The time decay operator is used to perform a weighted decay mapping on the historical evaluation records to generate effective feature components that characterize the logical evolution state of the target cognitive reasoning model, and the feature evaluation score is corrected based on the effective feature components.

7. A feature evaluation device for a cognitive reasoning model based on distributed ledger, characterized in that, include: The credential mapping module is used to obtain the target cognitive reasoning model, map it, and encapsulate it into an on-chain digital credential entity in the distributed ledger network. The oracle synchronization module is used to synchronize the execution feature data of the target cognitive reasoning model during its operation to the distributed ledger network using oracle components; The score evaluation module is used to calculate and generate feature evaluation scores for the on-chain digital certificate entities using evaluation contract operators deployed in the distributed ledger network. The state transition module is used to monitor the ownership state transition trigger signal and, in response to the ownership state transition trigger signal, use a smart contract to perform a numerical rewrite of the ownership mapping parameters associated with the on-chain digital certificate entity based on the feature evaluation score.

8. An electronic device, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the program to implement the steps of the method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

10. A feature evaluation system for a cognitive reasoning model based on distributed ledgers, characterized in that, include: The cognitive reasoning model layer is used to run the target cognitive reasoning model and generate running records. An oracle gateway is used to transform the execution records into trusted execution characteristic data using oracle components and synchronize them to the distributed ledger network. A distributed ledger network is used to maintain the on-chain digital credential entities generated by the target cognitive reasoning model mapping, and to deploy evaluation contract operators and smart contracts; The evaluation contract operator is used to calculate and generate a feature evaluation score based on the execution feature data; the smart contract is used to monitor the ownership state migration trigger signal and perform rewriting processing on the ownership mapping parameters associated with the on-chain digital certificate entity based on the feature evaluation score.

Images