Implementation method of intelligent non-player character in interactive game

Abstract

The invention provides a method for realizing an intelligent non-player character in an interactive game, which relates to the field of data processing and comprises the following steps of: packaging a decision intention into an immutable action package in a predefined decision period and placing the action package in a time domain trusteeship queue to realize time decoupling of a decision and a runtime load; the world state projection is divided into an opaque decision phase and a reconciliation phase, the decision phase only exposes the event category projection and the historical abstract, and the reconciliation phase reflows detailed trigger details in an independent window for posteriori maintenance but does not backtrack submitted actions; submission follows an atomization criterion and the submitted abstract does not contain time information; during resource contention, accumulating in response to a debt, and repaying through display shaping on the premise of not exposing a time sequence so as to keep the semantic consistency of main games; the posterior processing updates the action banks and event mappings offline based on the backflow details to improve subsequent decisions.

Description

Technical Field

[0001] This invention relates to the field of data processing, specifically to a method for implementing intelligent non-player characters in interactive games. Background Technology

[0002] In interactive games, intelligent non-player characters undertake a variety of interactive and environmental response tasks, and their complexity increases with the development of hardware networks; modern games require high concurrency and real-time performance to ensure immersion, so data processing and scheduling have become key foundations; decision-making systems for large-scale scenarios are shifting from rule-based to learning-based and hybrid models. Existing projects often employ compromises such as server-based authoritative deployment, reduced sampling, lockstep synchronization, rollback replay, and client-side prediction to cope with network and performance fluctuations. They commonly use behavior tree finite state machines or learning models to implement role decisions, and combine them with a layered event processing framework to coordinate the middleware and rendering layer to participate in the collaboration. The shortcomings of existing technologies are that they generally rely on the direct coupling between the decision-making process and the operating environment. Fluctuations in external conditions are transmitted to the behavior of the role through time delays or differences in data mapping. This makes the reaction patterns of the role predictable under high concurrency or resource contention conditions. This structural limitation results in the lack of necessary isolation and shielding mechanisms for the role's performance, making it difficult to maintain the independence and stability of behavior in complex environments. Summary of the Invention

[0003] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for implementing intelligent non-player characters in interactive games. This method solves the problem mentioned in the background that the decision-making process is directly coupled with the operating environment, and fluctuations in external conditions are transmitted to the character's behavior through time delays or data mapping differences, making the character's reaction patterns predictable under high concurrency or resource contention conditions.

[0004] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a method for implementing an intelligent non-player character in an interactive game, comprising the following steps: within a predefined decision-making cycle, the non-player character's decision-making process forms an intent and outputs the intent as an immutable action envelope; the action envelope is submitted to a time-domain managed process for storage; the action envelope includes an action category identifier, an applicable event category identifier, a priority category, and a lifetime window label, but does not contain frame-level input or resource usage information; The time-domain hosting process places the action envelopes into a hosting queue and arranges them in an orderly manner according to predefined unpredictable arrangement rules. When an external trigger event category is detected to match the applicable event class identifier of the action envelope, the action envelopes are submitted and made effective item by item according to the atomic submission criteria. The world state projection is divided into an opaque decision phase and a reconciliation phase; in the opaque decision phase, a read-only view containing only event category projections and historical submission summaries is provided to the non-player character decision-making process, and the exposure of fine-grained inputs, load metrics and concurrent request details of the current frame is prohibited; After the atomic submission, the actual triggering details of this round are fed back to the post-processing flow within an independent, unrelated time window for sample accumulation and action seal library update, and retrospective rewriting of submitted action seals is prohibited. When resource contention exists, the output processing phase maintains the atomic effective semantics defined by the atomic submission criteria, accumulates the contention as a reaction debt, and presents the submission result in a non-temporally reversible display manner according to the display shaping rules. The presentation process is display shaping. The post-processing flow maintains the action seal library based on the reflow details and updates the event category mapping table to improve the generation of subsequent action seals.

[0005] Preferably, within a predefined decision-making cycle, non-player characters generate action intentions according to their inherent intention model and map these intentions to specific action category identifiers. Each action category identifier corresponds to a unique operation type number, and each action envelope is assigned an immutable, globally consistent identifier to ensure it cannot be tampered with or reordered during subsequent hosting and submission. Each action envelope also has a priority category identifier and a lifetime window label. Priority categories are divided into high, medium, and low to clearly define the submission order priority. The lifetime window is the validity period of the action envelope, counted in milliseconds from the submission time. After the lifetime window expires, the action envelope will be terminated. The time-domain managed process automatically performs the deprecation operation; the action envelope does not contain any frame-level input, load metrics, or concurrent request details to ensure that non-player character decision-making processes cannot utilize the underlying state for unintended optimizations; the generated action envelope is submitted to the time-domain managed process through a controlled output interface, which is limited to accepting only a predefined set of event classes to ensure input atomicity. Before submission, no external observer can know the action envelope queue status or submission order; after receiving the action envelope, the time-domain managed process generates an immutable reception digest for each action envelope and records the action envelope reception order for subsequent reconciliation, while also following unpredictable arrangement rules. Action seals are placed in a managed queue to prevent submission patterns from being statistically predicted; unpredictable arrangement rules are provided by random perturbations from multiple independent entropy sources, combined with action seal priority categories and diversity constraints to prevent consecutive action seals from exhibiting stable dequeue patterns; when the external trigger event category matches the event class identifier applicable to the action seal, the time-domain managed process executes the action seal submission operation item by item according to the atomic submission criterion, prohibiting any intermediate state from being observed externally during the submission process, and only returning an immutable submission summary and confirmation flag to the non-player character decision-making process, ensuring that non-player characters cannot obtain timing information through submission feedback; After an action seal is submitted, the actual trigger details of this round are fed back to the post-processing flow within an independent, unrelated time window. The post-processing flow is used for sample accumulation and action seal library updates, and any retrospective modifications to submitted action seals are prohibited to ensure the consistency and immutability of action seal submissions. Under resource contention conditions, the output processing phase still maintains the atomic effectiveness semantics defined by the atomic submission criteria, and accumulates contention in the form of reaction debt. At the same time, the submission results are presented in a timeless and reversible manner according to the display integer rules to ensure that the main semantics of the game are not destroyed, and each operation generates an immutable log for post-analysis.

[0006] Preferably, within each decision cycle, the non-player character's decision-making process processes the current environment input and historical event summaries based on the intrinsic intent model, generates action intents, maps these intents to specific action category identifiers, and attaches an immutable, globally consistent identifier to each action seal. This globally consistent identifier is unique and fixed within the system, ensuring global consistency of the action seal during hosting and submission. During the mapping process, the non-player character's decision-making process categorizes actions into high-priority, medium-priority, and low-priority categories based on their priority, allowing the time-domain hosting process to prioritize actions according to priority in unpredictable arrangements. A lifetime window is also attached, counted in milliseconds, automatically starting from the seal generation time. If the lifetime window expires and the action seal is not matched and submitted, the hosting process automatically performs a discard operation and writes the discard record to an immutable log. The action seal output interface is limited to only accepting... Subject to a predefined set of event classes, the interface only allows controlled submissions of operations, prohibiting external access to frame-level input sequences, load metrics, or concurrent request details, thus ensuring the atomicity of action seal inputs. During the action seal's hosting period, any externally observable resource status or concurrent request details are prohibited from being used for rewriting or reordering. After each action seal is received and confirmed, the time-domain hosting process generates an immutable summary of the received data and records the receiving order for subsequent reconciliation operations, ensuring the immutability and consistency of the action seal throughout its lifecycle. During the hosting period, if the action seal's lifespan window expires or the number of matching failures reaches a predefined limit, the time-domain hosting process performs expiration or abandonment operations according to the lifespan window and matching strategy, and writes the abandonment record to an immutable log for post-analysis. The process does not expose the number of retries, submission order, or specific abandonment time to non-player characters or external systems, thus ensuring that non-player characters cannot use the submission timing for strategy optimization or external observation.

[0007] Preferably, after receiving an action envelope during the time-domain hosting process, the action envelope is placed into the hosting queue according to unpredictable arrangement rules, and a receiving order summary is generated for each action envelope for subsequent reconciliation operations. The unpredictable arrangement rules are driven by multi-source unassociative entropy sources and combined with action envelope priority categories and diversity constraints to avoid the hosting queue exhibiting a stable and statistically predictable dequeue pattern, thereby ensuring that the action envelope submission order is unpredictable. When an external trigger event category matches the event class identifier applicable to the action envelope, the time-domain hosting process performs an atomic state update operation on the single action envelope according to the atomic submission criteria, and simultaneously generates a submission immutable summary. The submission immutable summary records the action envelope identifier, the matching event class identifier, and the submission decision, but does not record the submission time or load indicators to ensure that non-player characters cannot obtain submission timing information. During the submission execution process, external observation of any intermediate state is prohibited, and after the submission is completed, the immutable summary is written to the immutable log for later use. The verification and reconciliation process does not allow retrospective modification or real-time intervention of submitted action seals. At any stage of the hosting process, the queue arrangement, submission time, and intermediate state of action seals are not exposed. Non-player characters only receive immutable submission digests and confirmation flags during submission, thus maintaining the global consistency and atomicity of action seals. If a submission fails or a match is unsuccessful, the time-domain hosting process performs internal retry or discard operations according to the action seal assignment window and matching strategy, and writes the retry record to the immutable log, but does not disclose the number of retryes, retry time, or action seal arrangement status to ensure that the submission behavior cannot be predicted or exploited externally. The entire hosting process ensures that action seals strictly follow priority ordering, diversity constraints, and atomic submission rules throughout their lifecycle, and forms a complete security closed loop through unpredictable arrangement and immutable digests, ensuring that the action seals output by non-player characters in the decision-making process are completely unknowable in terms of timing, state, and observable information.

[0008] Preferably, the world state projection is divided into an opaque decision phase and a reconciliation phase, and strict access constraints are imposed on the projection interface. In the opaque decision phase, the non-player character's decision-making process only receives a read-only view. This read-only view contains only the event category hierarchy, historical commit summaries, and necessary environmental topology mappings to provide environmental reference and decision-making basis. It prohibits the inclusion of any frame-level input sequences, load metrics, or concurrent request details, thereby ensuring that the non-player character cannot optimize or predict the action seal submission results through low-level fine-grained information. Within the opaque decision box, the non-player character can only generate action seals based on event categories and historical commit summaries. The action seals output by their decision-making process are submitted to the time-domain managed process through a controlled interface. The interface is prohibited from directly accessing the queue arrangement state or commit timing information, ensuring the atomicity of the action seals. In the reconciliation phase, [the process is] independent and non-associated. Within the time window, the actual trigger details of this frame are fed back to the post-processing flow for action seal library sample accumulation and action seal adaptation tag statistical calibration. The event category mapping table is also updated to improve the generation of the next round of action seals. During the feedback, rewriting of submitted action seals or immediate driving of new submission operations is prohibited to ensure the consistency of the action seal lifecycle. After the post-processing flow processes the feedback details, the updated results replace the event category mapping table at the decision cycle boundary for future use. This is not visible to the outside world to ensure the consistency of the opaque decision phase view and global unpredictability. The entire process ensures strict separation between the opaque decision phase and the reconciliation phase. Non-player characters cannot obtain any underlying state information of the reconciliation phase during the decision cycle. At the same time, the reconciliation phase update operation is completed in an independent channel, and an immutable maintenance summary is generated and written to the audit log to ensure data integrity and traceability.

[0009] Preferably, when event category matching is triggered, the time-domain managed process performs an atomic state update operation on a single action seal according to the atomic commit criterion. The atomic state update requires that no intermediate state visible to external observers be allowed during the commit process, and the update operation must be completed within a predefined consistency boundary to ensure the integrity and indivisibility of the action seal during commit. Simultaneously with the update operation, an immutable commit digest is generated, recording the action seal identifier, the matching event class identifier, and the commit decision. The commit time, load indicators, or queue arrangement status are not recorded, ensuring that non-player characters and external systems cannot obtain timing information or the status of available resources. After the commit is completed, the immutable commit digest is written to the immutable log as a post-verification reconciliation basis. The post-verification processing flow receives the immutable digest for sample processing. The system accumulates and updates action seals, but does not use them for immediate decision-making to modify them or trigger the generation of new action seals. If an action seal submission fails, the time-domain managed process performs an internal retry operation based on the action seal's lifespan window and matching strategy. The retry operation is completed in an internal loop, and the number and timing of retryes are not publicly disclosed. If the retry still fails, the action seal is discarded, and the discard record is written to an immutable log for post-analysis. At the same time, it ensures that submitted action seals cannot be backtracked or modified. The entire atomic submission process ensures that action seals have global consistency and irreversibility during submission. The action seal submission result is only fed back to non-player characters through an immutable digest, and any intermediate states, queue arrangement information, and submission time are not exposed, thereby maintaining the unpredictability and global consistency of the submission process for non-player characters.

[0010] Preferably, when resource contention is detected during the output processing phase, the submission result is recorded in the reaction debt ledger, maintaining atomic validity semantics. The reaction debt ledger records the amount of each resource contention according to entity ownership and sets a repayment priority, while limiting the ledger's upper limit to prevent long-term accumulation from leading to the formation of observable patterns. During the repayment process, the reaction debt is presented in a timeless and reversible manner according to the display shaping rules. Specific operations include adjusting the presentation order or display effect of non-critical visual elements and auxiliary presentation elements to ensure that the main game semantics are not destroyed. At the same time, presentation diversity constraints are introduced into the repayment strategy to prevent repeated display patterns from being statistically reproduced. Throughout the entire output processing phase, the timing of reaction debt entry, repayment order, and accumulation details are not changed. By exposing resource contention to the outside world in a sequential or delayed manner, no non-player character or external system can obtain the original time information of resource contention. After the repayment operation is completed, the repayment summary is written to the synchronous log to ensure that it can be used for post-analysis and auditing. At the same time, the repayment mode is prohibited from being used as a real-time queryable runtime indicator to ensure the stability and unpredictability of the game experience. In the case of multiple rounds of resource contention, the output processing stage processes each reaction debt in sequence according to the repayment priority and the ledger limit, ensuring that the action seal effective semantics defined by the atomic submission criteria are continuously maintained. The digestion and feedback of resource contention are completed through a timeless and irreversible display method, realizing the global consistency and unpredictability of the non-player character action seal submission and the external display effect.

[0011] Preferably, after receiving the current loop details in the post-processing flow, incremental maintenance is performed on the action seal library. This includes statistical calibration of action seal adaptation tags and preparation for replacing the event category mapping table. Incremental maintenance is conducted in an independent offline processing channel, completely isolated from non-player character decision-making processes and managed queues, and uses an unobservable model update channel to prevent exploitable time or load indications during maintenance. During incremental maintenance, the action seal priority and applicable event class mapping are adjusted according to the statistical results of the action seal adaptation tags, and an immutable maintenance summary is generated to record the details of each update operation. This immutable maintenance summary is written to the audit log for post-analysis, but does not expose any intermediate state or timing information to non-player characters or external systems. After maintenance is completed... Within the isolated verification environment, consistency and diversity checks are performed to verify whether the updated event category mapping table meets the constraints of global consistency, readability of the opaque decision phase, and submission diversity. Only after passing the checks is the updated result used to replace the event category mapping table at the decision cycle boundary for use in the next decision cycle, while remaining invisible to the outside world. The post-processing flow only uses the maintenance results as input for the generation of action seals in the next decision cycle. The entire process prohibits the immediate driving of new action seal submissions or modification of submitted action seals during the maintenance period, thereby ensuring the immutability and global consistency of the action seal library and event category mapping table within the decision cycle, and ensuring that non-player characters cannot obtain any exploitable underlying information in the opaque decision phase, thus maintaining the integrity and unpredictability of the operation.

[0012] Preferably, the entire non-player character action envelope generation and managed submission process first involves the non-player character decision-making process generating action intents based on an intrinsic intent model within each decision cycle, mapping these intents to specific action category identifiers. Each action envelope is appended with an immutable globally consistent identifier, priority category, and lifetime window label to ensure global consistency and atomicity. The action envelope is submitted to the time-domain managed process through a controlled output interface. During the managed process, the action envelope is received and confirmed, and an immutable reception digest is generated. Simultaneously, it is placed in the managed queue according to unpredictable arrangement rules. These unpredictable arrangement rules combine multi-source, unassociative entropy sources, priority categories, and diversity constraints to prevent the queue dequeue pattern from being statistically predictable. When an external trigger event category matches the event category identifier applicable to the action envelope, the time-domain managed process performs a single atomic submission of the action envelope according to atomic submission criteria, generating an immutable submission digest for post-verification reconciliation. During the submission process, exposure of intermediate states, queue arrangements, or submission sequence information is prohibited. Within an independent, unassociative time window... In this round, the actual trigger details are fed back to the post-processing flow for incremental maintenance of the action seal library and updating of the event category mapping table. At the same time, it is prohibited to write back to the submitted action seal or trigger new submission operations. In the output processing stage, when resource contention is detected, the contention is accumulated to the ledger in the form of reaction debt and repaid in a non-temporally reversible manner according to the display shaping rules, while maintaining the main game semantics and diversity constraints. The repayment summary is written to the synchronous log for post-analysis. The post-processing flow performs consistency and diversity verification on the maintenance results in an isolated verification environment. Only after passing the verification is the event category mapping table replaced at the decision cycle boundary to ensure the consistency of the opaque decision-making view. The entire process forms a closed-loop logic to ensure that the action seal generation, hosting, submission, reflow, and maintenance operations maintain atomicity, non-retrogression, and unpredictability in the global scope. An immutable summary and log record are generated at each stage to support post-reconciliation and sample accumulation, while preventing non-player characters or external systems from obtaining usable underlying state or temporal information.

[0013] Preferably, each action seal is assigned a defined lifetime window upon generation. This lifetime window is counted in milliseconds and takes effect from the moment the action seal is submitted to the time-domain hosting process. If the action seal is not matched by an externally triggered event category within the lifetime window, the time-domain hosting process executes an internal retry operation according to the matching strategy. The retry operation is executed cyclically at fixed time intervals. Before each retry, the lifetime window of the action seal is verified to be still valid. If the lifetime window expires or the number of matches reaches a predefined upper limit, the action seal automatically enters a discard state, and the discard operation record is written to an immutable log for post-analysis. The retry operation is completed in an independent internal channel, without exposing the retry count, retry time, or action seal queue status to non-player characters or external systems, thereby ensuring the unpredictability and atomicity of the submission behavior. During the retry process, each matching attempt still follows the atomic commit principle, meaning that any state update operation is prohibited from external observation of the intermediate state during execution, and an immutable commit summary is generated after a successful match for post-verification reconciliation. After the deprecation operation is completed, the action seal can no longer be backtracked or reactivated. The deprecation record, along with metadata such as the end time of the lifetime window and the number of matching failures, is written to an immutable log for subsequent sample accumulation and action seal database optimization. The entire lifetime window, matching strategy, retry, and deprecation mechanism form a complete closed loop, ensuring that the action seal has strict atomicity, non-retrogression, and global consistency within the hosting and commit cycle. At the same time, it provides traceable operation records for the post-verification process and ensures that non-player characters cannot use lifetime window or retry strategy information to predict commit behavior in advance.

[0014] Preferably, after action seals generated by non-player characters are submitted to the time-domain managed process through a controlled interface, the action seals wait in the managed queue for matching with external trigger event categories. When an external event category matches the event class identifier applicable to the action seal, the managed process submits the action seals item by item according to the atomic submission criteria and generates an immutable submission digest for post-processing reconciliation. Simultaneously, it ensures that queue arrangement, submission sequence, and intermediate state information are not exposed during the submission process. The world state projection is divided into an opaque decision phase and a reconciliation phase. In the opaque decision phase, non-player characters only receive a read-only view containing event category hierarchy, historical submission digests, and necessary environmental topology mappings. Access to frame-level input sequences, load metrics, or concurrent request details is prohibited to ensure the unpredictability and global consistency of action seal generation and submission. In the reconciliation phase, within an independent, unrelated time window, the actual trigger details of this round are fed back to the post-processing flow for incremental maintenance of the action seal library. The event category mapping table is updated and samples are accumulated. During the reflow period, rewriting of submitted action seals or triggering new submission operations is prohibited. The post-processing process performs consistency and diversity checks in the isolated verification environment, replaces the event category mapping table for use in the next decision cycle, and generates an immutable maintenance summary record update operation to ensure that non-player characters cannot access the underlying data of reconciliation. The entire process forms a complete closed loop. From the generation, management, atomic submission, lifespan window management, retry and discard operations to reflow processing and post-maintenance, action seals maintain irreversibility, atomicity, and unpredictability. At the same time, immutable summaries and log records provide traceable evidence for sample accumulation, reconciliation analysis, and action seal library optimization, ensuring that non-player characters cannot obtain exploitable underlying states, timing information, or submission order at any time during the decision-making process. This achieves global consistency and a secure closed loop for intelligent non-player character action seal management in interactive games.

[0015] Beneficial effects This invention provides a method for implementing intelligent non-player characters in interactive games. It has the following beneficial effects: This invention achieves security and stability for non-player character interactions in the game environment by generating action seals within a predefined decision cycle and submitting them to a time-domain managed process; ensuring the immutability and atomic submission of action seals; maintaining consistency in state updates when triggering event category matching; prohibiting external observation of intermediate states and submission timing information; and providing traceable evidence through immutable summaries and log records.

[0016] This invention divides the world state projection into an opaque decision phase and a reconciliation phase; in the opaque decision phase, only event category projections and historical submission summary views are provided; in the reconciliation phase, real trigger details are fed back and the action database is updated; a complete closed loop is maintained for lifetime windows and abandonment records; during resource contention, resources are accumulated in the form of reactive debt and repaid without time sequence through the display of shaping rules. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1: This invention provides a method for implementing intelligent non-player characters in an interactive game. The method includes deploying interactive non-player characters in a multiplayer online role-playing game. First, within each decision cycle, the non-player character decides to process current environment information and historical event summaries based on an intrinsic intent model to generate action intents. These action intents are then mapped to specific action category identifiers, each corresponding to a unique operation type number. Simultaneously, an immutable, globally consistent identifier is assigned to the generated action envelope to ensure its uniqueness and immutability during the hosting and submission process. The action envelope also includes a priority category identifier and a lifetime window label. The priority category is divided into high, medium, and low to clarify the submission order priority. The lifetime window is counted in milliseconds from the moment the action envelope is generated. Action envelopes that fail to match within the lifetime window are discarded by the time-domain hosting process, and the discard record is written to an immutable log. The action envelope output interface is limited to accepting only a predefined set of event classes to ensure the atomicity of the action envelope input. Non-player characters are prohibited from accessing frame-level input sequences, load metrics, or concurrent request details to prevent unintended optimization.

[0019] After an action envelope is submitted to the time-domain hosting process, a receipt confirmation is performed for each action envelope, and an immutable receipt digest is generated. At the same time, the action envelopes are placed into the hosting queue according to unpredictable arrangement rules. The unpredictable arrangement rules are provided by random perturbations from multiple sources of unassociative entropy sources, combined with priority categories and diversity constraints to avoid consecutive action envelopes exhibiting a stable and statistically measurable dequeue pattern. During the hosting process, external observation of the action envelope queue arrangement or submission timing information is prohibited. When an external trigger event category matches the event class identifier applicable to the action envelope, the atomic state update operation of the action envelope is performed item by item according to the atomic submission criteria, and an immutable submission digest is generated to record the action envelope identifier, the matching event class identifier, and the submission decision. The submission time or load index is not recorded, and external observation of the intermediate state is prohibited.

[0020] After submission, within an independent, unrelated time window, the actual trigger details of this round are fed back to the post-processing flow for incremental maintenance of the action seal library, updating of the event category mapping table, and sample accumulation. During the feedback period, rewriting of submitted action seals or triggering new submission operations is prohibited to maintain the consistency of the action seal lifecycle. In the output processing stage, when resource contention is detected, the submission results are recorded in the reaction debt ledger, and the contention amount and repayment priority are recorded according to entity ownership. At the same time, the upper limit of the ledger is limited to prevent long-term accumulation and the formation of statistical patterns. The repayment operation presents a timeless and reversible effect by adjusting non-critical visual elements and auxiliary presentation elements, while introducing presentation diversity constraints to prevent duplicate display patterns. The repayment summary is written synchronously. Logs are used for post-analysis and do not expose the original entry time to non-player characters or external systems. The post-processing flow performs consistency and diversity checks in an isolated verification environment, performs statistical calibration of action seal adaptation tags, and generates an immutable maintenance summary. Only after the verification passes can the event category mapping table be replaced at the decision cycle boundary for use in the next decision cycle. The entire process forms a closed-loop operation from action seal generation, hosting, atomic submission, lifetime window management, retry and discard operations, reflow processing to post-maintenance, ensuring the global consistency, atomicity, non-retrogression and unpredictability of action seals, and providing traceable evidence through immutable summaries and log records, providing a safe and reliable execution environment for non-player character decision-making.

[0021] Example 2: This embodiment, based on Embodiment 1, differs in that it optimizes and adjusts the processing order of action seal hosting and submission, as well as the use of the reflow time window, to improve overall submission consistency and post-processing efficiency under multi-event triggering conditions. It maintains the closed-loop operation process of action seal generation, hosting, atomic submission, lifetime window management, retry and discard, reflow, and post-maintenance as in Embodiment 1. Within each decision cycle, non-player character decisions still process current environment information and historical event summaries based on the internal intent model to generate action intents, mapping these intents to specific action category identifiers. Each action seal is appended with an immutable globally consistent identifier, priority category, and lifetime window label. Priority categories are divided into high, medium, and low to clarify the hosting and submission order. The lifetime window is counted in milliseconds from the moment the action seal is generated. Action seals that fail to match within the lifetime window are discarded by the time-domain hosting process, and the discard record is written to an immutable log. The action seal output interface is limited to accepting only a predefined set of event classes to ensure input atomicity, and non-player characters are prohibited from accessing frame-level input sequences, load metrics, or concurrent request details.

[0022] After an action envelope is submitted to the time-domain hosting process, a receipt confirmation is executed, and an immutable receipt digest is generated. Simultaneously, it is placed into the hosting queue according to unpredictable arrangement rules combined with priority categories and diversity constraints. Unlike Implementation 1, this implementation submits action envelopes of the same priority category in batches and segments within the hosting queue according to the receipt order. This ensures that when externally triggered event categories are matched, each batch of action envelopes updates its status and generates an immutable submission digest during the atomic submission process. External observation of intermediate states, queue arrangement, or submission sequence information is still prohibited. Within an independent, unrelated time window, the actual trigger details of this round are flowed back to the post-processing flow for incremental maintenance of the action envelope library, updating of the event category mapping table, and sample accumulation. During the flowback, writing back to already submitted action envelopes or triggering new submission operations is prohibited. Output processing... When resource contention is detected, the submission results are recorded in the reaction debt ledger, and the contention amount and repayment priority are recorded according to entity ownership. At the same time, the upper limit of the ledger is limited to prevent long-term accumulation and the formation of statistical patterns. The repayment operation presents a timeless and reversible effect by adjusting non-critical visual elements and auxiliary performance elements, and introduces presentation diversity constraints to prevent repeated display patterns. The repayment summary is written to the synchronous log for post-analysis, and the original entry time is not exposed to non-player characters or external systems. The post-processing process performs consistency and diversity checks in an isolated verification environment, performs statistical calibration on action seal adaptation tags, and generates an immutable maintenance summary. After the verification passes, the event category mapping table is replaced at the decision cycle boundary for use in the next decision cycle. Action seals of the same priority are submitted in batches and segments in the managed queue.

[0023] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for implementing an intelligent non-player character in an interactive game, characterized in that, Includes the following steps: Within a predefined decision-making cycle, the non-player character's decision-making process forms an intent and outputs the intent as an immutable action seal; the action seal is submitted to the time-domain managed process for sealing; the action seal includes an action category identifier, an applicable event category identifier, a priority category, and a lifetime window label, but does not contain frame-level input or resource consumption information. The time-domain hosting process places the action envelopes into a hosting queue and arranges them in an orderly manner according to predefined unpredictable arrangement rules. When an external trigger event category is detected to match the applicable event class identifier of the action envelope, the action envelopes are submitted and made effective item by item according to the atomic submission criteria. The world state projection is divided into an opaque decision phase and a reconciliation phase; The opaque decision-making process provides a read-only view to the non-player character's decision-making process, containing only event category projections and historical submission summaries, and prohibits the exposure of fine-grained inputs, load metrics, and concurrent request details for this frame; After the atomic submission, the actual triggering details of this round are fed back to the post-processing flow within an independent, unrelated time window for sample accumulation and action seal library update, and retrospective rewriting of submitted action seals is prohibited. When resource contention exists, the output processing phase maintains the atomic effective semantics defined by the atomic submission criteria, accumulates the contention as a reaction debt, and presents the submission result in a non-temporally reversible display manner according to the display shaping rules. The presentation process is display shaping. The post-processing flow maintains the action seal library based on the reflow details and updates the event category mapping table to improve the generation of subsequent action seals.

2. The method for implementing an intelligent non-player character in an interactive game according to claim 1, characterized in that: In each decision cycle, the non-player character decision-making process maps the intent to an action category based on the inherent intent model when outputting the action seal, and attaches an immutable globally consistent identifier, priority category, and lifetime window label to the action seal. The action seal output interface of the non-player character decision-making process is limited to accepting a predefined set of event classes, which are submitted to the time-domain managed process through a controlled interface to ensure input atomicity. The action block prohibits any externally observable resource status or concurrent request details from being used for rewriting or reordering during the hosting period; The time-domain managed process performs a receipt confirmation on the action seal and records the received immutable digest for subsequent reconciliation; the action seal is expired or discarded by the time-domain managed process according to the lifetime window and the number of matching failures, and the discard record is written to the immutable log for subsequent analysis.

3. The method for implementing an intelligent non-player character in an interactive game according to claim 2, characterized in that: After receiving the action envelope, the time-domain hosting process places the action envelope in a hosting queue according to an unpredictable arrangement rule and maintains a summary of the receiving order of each action envelope. The unpredictable arrangement rules are driven by multi-source, unassociative entropy sources and combined with priority categories and diversity constraints to avoid stable and statistically predictable dequeue patterns; when the external trigger event category matches the applicable event class identifier of the action envelope, the time-domain managed process executes a single commit according to the atomic commit criterion and life strategy, and simultaneously writes an immutable commit digest for subsequent reconciliation; the time-domain managed process does not expose any observable information that can indicate queue arrangement or commit time before committing; When performing a commit, the time-domain hosting process only returns an immutable commit summary and confirmation flag to the non-player character's decision-making process and prohibits the return of any commit time or load metrics, thereby preventing the non-player character from knowing the exploitable timing information at the time of commit.

4. The method for implementing an intelligent non-player character in an interactive game according to claim 3, characterized in that: The world state projection is divided into an opaque decision phase and a reconciliation phase, and access constraints are imposed on the projection interface. In the read-only view provided to the non-player character in the opaque decision phase, only the event category hierarchy, historical commit summary, and necessary environmental topology mapping are included, and frame-level input sequences, load metrics, or concurrent request details are prohibited. In the reconciliation phase, the actual trigger details of the current frame are flowed back to the post-processing flow within an independent time window for sample accumulation and action seal library updates, and rewriting of committed action seals or immediate driving of new commits is prohibited within the flowback window. The event category mapping table is replaced by the post-processing flow after the reconciliation phase update for future use and is not visible to the outside world.

5. The method for implementing an intelligent non-player character in an interactive game according to claim 4, characterized in that: The atomic commit criterion requires that when a matching event category is triggered, the time-domain managed process performs an atomic state update on the single action and simultaneously generates an immutable commit digest. The commit immutable digest records the action seal identifier, the matching event class identifier, and the commit resolution, but does not record the commit time; the atomic state update follows a predefined consistency boundary and prohibits any intermediate state visible to external observers during the update process; After submission, the immutable summary of the submission is written to the immutable log and transmitted to the post-processing flow as a reconciliation basis, but is not used for immediate decision-making and modification; in the event of submission failure, the time-domain managed process performs internal retry or abandonment operations based on the action seal lifetime window and matching strategy, and writes the retry record to the immutable log without exposing the retry time or number of times.

6. The method for implementing an intelligent non-player character in an interactive game according to claim 5, characterized in that: When resource contention is detected, the output processing phase records the submission result in the reaction debt ledger, maintaining atomic validity semantics. The reaction debt ledger records the contention amount and repayment priority according to entity ownership and limits the upper limit to prevent long-term observable accumulation patterns. The output processing phase repays the reaction debt in a non-temporally reversible manner by adjusting non-critical visual or auxiliary presentation elements according to presentation shaping rules, and introduces presentation diversity constraints in the repayment strategy to avoid statistically reproducible presentations. The output processing phase does not expose the original entry time of the reaction debt in any way through presentation order or delay. When repaying the reaction debt, the presentation shaping maintains the main game semantics unchanged and writes the repayment summary to the synchronization log for post-analysis, while prohibiting the use of repayment patterns as queryable runtime indicators.

7. The method for implementing an intelligent non-player character in an interactive game according to claim 6, characterized in that: After receiving the details of the current loop, the post-processing flow performs incremental maintenance on the action library, including statistical calibration of action library adaptation tags and preparation for replacing the event category mapping table; the incremental maintenance is performed in an independent offline processing channel and uses an unobservable model update channel to prevent the generation of exploitable time or load indications during maintenance. The post-processing procedure uses the maintenance results only as input for the action generation of the next decision cycle, and writes an immutable maintenance summary into the audit log when recording maintenance changes. Before replacing the event category mapping table with the maintenance results, the post-processing procedure performs consistency and diversity checks in an isolated verification environment, and performs replacement operations only at the boundary of the decision cycle to ensure the consistency of the opaque decision-making view.