Open Access. Powered by Scholars. Published by Universities.®
- Publication Year
- File Type
Articles 1111 - 1140 of 11790
Full-Text Articles in Entire DC Network
Ai-Powered Autonomous Agent For Enterprise Platform Configuration With Live Introspection And Dependency-Aware Deployment, Anonymous
Defensive Publications Series
An autonomous agent configures a configurable enterprise platform to onboard a new business vertical. The agent performs live environment introspection using parallel administrative API calls to build a snapshot and a graph of configuration components and their relationships. Existing verticals are detected using naming patterns and relationship traversal with confidence scoring, and a baseline vertical template is extracted. A three-phase requirement analysis maps requirement patterns to component clusters, matches required components against the snapshot using five-tier classifications (exact, sibling, partial, misconfigured, missing) with risk metadata, and enriches results using a machine-learning model that identifies dependencies and contradictions and outputs structured …
Systems And Methods For Automated Recipe Extraction From Multimedia Content With Visual Pantry Inventory Detection And Differential Shopping List Generation, Anonymous
Defensive Publications Series
Systems and methods are described for transforming unstructured cooking-related multimedia into actionable outputs. A client provides a video link, web link, screenshot, or photo. One or more models extract ingredients and preparation actions using combinations of speech recognition, optical character recognition, and visual recognition. A structured recipe is generated with ingredients and ordered steps, and additional attributes such as serving size, nutrition or macronutrients, and dietary tags are derived. A meal plan may be produced over a time window, and required ingredients are aggregated. Optionally, the user captures images of pantry and/or refrigerator contents via a guided flow. Inventory items …
Systems And Methods For Session-Level Emotional Trajectory Management In Content Recommendation With Asymmetric Streak-Breaking And Arc Template Shaping, Anonymous
Defensive Publications Series
A recommendation technique manages session-level emotional exposure by using precomputed emotional features for content items, including valence and emotional weight, and maintaining a running emotional state during an active user session. The running emotional state is used to classify a trajectory state such as healthy, drifting_negative, spiral, recovery_needed, or variance_collapsed. Based on the trajectory state, a controller modifies a ranking pipeline by pruning candidates, adjusting scores, and/or enforcing constraints. A hard asymmetric negative streak breaker constrains a next ranked position to a non-negative-valence item after at least N consecutive negative-valence items, without imposing a corresponding constraint on positive streaks. Additional …
Systems And Methods For Asymmetric Propagation Constraints With Partitioned Memory Architecture In Collaborative Recommendation, Anonymous
Defensive Publications Series
Techniques are described for collaborative recommendation that constrain cross-user propagation of user-derived signals using asymmetric, type-dependent rules. A signal classification layer assigns each generated insight to a psychological signal type including positive preference, negative preference, behavioral pattern, and vulnerability indicator. An asymmetric propagation engine applies type-specific parameters such as hop depth, evidence thresholds, confidence discounting, and temporal decay, including higher friction for negative, behavioral, and vulnerability-related signals than for positive preferences. Negative preference propagation is constrained to a demotion-only effect in ranking, reducing scores without excluding candidates. A partitioned memory architecture stores ranking-accessible preferences in a first partition and stores …
Systems And Methods For Automated Code Review Quality Assessment Using Hybrid Classification And Logarithmic Scoring With Anti-Gaming Safeguards, Anonymous
Defensive Publications Series
Systems and methods are described for automated assessment of code review quality using review comments, reviewer actions, reactions, and diff context. A pipeline identifies human reviewers, extracts per-reviewer comment threads, and executes parallel sub-agent analyses to produce structured signals including impact and relevance scores and comment attributes. A deterministic ruleset classifies each comment into HIGH, MEDIUM, LOW, or OUT_OF_SCOPE. A reviewer score is computed using weighted counts of classified comments with a logarithmic normalization term to provide diminishing returns, and may include additive bonuses such as code suggestions, reactions, and request-changes actions. Multi-layer integrity safeguards may include concentration ratio thresholds, …
Systems And Methods For Trust-Adaptive Ranking With Asymmetric Recovery In Content Recommendation, Anonymous
Systems And Methods For Trust-Adaptive Ranking With Asymmetric Recovery In Content Recommendation, Anonymous
Defensive Publications Series
Systems and methods maintain a per-user trust score for a recommendation system and adapt ranking behavior based on the trust score. The trust score is updated after recommendation outcomes using asymmetric dynamics in which positive outcomes increase trust with diminishing returns and negative outcomes decrease trust more strongly, optionally using severity weights for different negative signals and a small decay for neutral outcomes. The trust score is classified into discrete trust states that simultaneously modulate multiple ranking parameters, including a candidate quality threshold, an exploration rate, prediction-confidence weighting applied to ranking scores, and sensitivity to negative-signal predictions. When trust is …
Systems And Methods For Curiosity-Zone Detection And Information-Gap-Driven Exploration In Content Recommendation, Anonymous
Defensive Publications Series
Techniques are described for targeted exploration in content recommendation using per-topic user knowledge state and curiosity-zone detection. For each user-topic pair, a system maintains an engagement-derived knowledge depth profile including exposure count and a depth distribution across difficulty levels. A curiosity intensity is computed as an inverted-U function of exposure count with parameters learned per user or cohort from engagement-versus-exposure data, and a curiosity state is classified based on exposure and an engagement trend. Candidate items are scored for exploration using the topic curiosity intensity, a depth-gap fit to a target information gap of approximately one difficulty level above the …
Systems And Methods For Phase-Specific Content Diversity Modulation Based On Per-User Per-Topic Satiation State Tracking, Anonymous
Defensive Publications Series
Techniques are described for recommendation pipelines that track per-user per-topic satiation state and modulate topic diversity and quality policies accordingly. For each user-topic pair, the system maintains exposure counts and an engagement quality trend computed from a composite of dwell time, save rate, and completion rate. The user-topic pair is classified into one of six phases: discovery, rising, plateau, declining, saturated, or recovery, using exposure, trend, and time since last exposure, with thresholds normalized by the user’s consumption rate. Phase-dependent multipliers are computed to scale a baseline minimum quality threshold and a base per-topic diversity cap, including relaxing diversity and …
Cognitive Mode Detection And Conditional Feature Gating For Content Ranking Systems, Anonymous
Cognitive Mode Detection And Conditional Feature Gating For Content Ranking Systems, Anonymous
Defensive Publications Series
Techniques are described for mode-aware ranking in recommendation systems. Session-level behavioral signals are aggregated over a sliding window to form a behavioral feature vector including at least dwell time, scroll velocity, topic autocorrelation, and interaction depth ratio. A cognitive mode detector outputs a probability distribution over cognitive modes including browse, decide, and flow. A feature gating network uses the mode probability distribution to generate gate weights for feature groups, and the gate weights are applied to candidate item feature groups to produce gated features used for scoring and ranking. Within-session transitions may be managed using hysteresis and smooth interpolation of …
Partitioned User Profile Architecture With Differentiated Access Control For Recommendation Systems, Anonymous
Partitioned User Profile Architecture With Differentiated Access Control For Recommendation Systems, Anonymous
Defensive Publications Series
Techniques are described for partitioning user profiles in agentic recommendation systems into a preference partition and a behavioral vulnerability partition with differentiated access control. A ranking agent retrieves only the preference partition to rank candidate content items, while a separate well-being monitoring system retrieves only the behavioral partition to compute a vulnerability-related measure and determine interventions. The well-being monitoring system transmits serving constraints to the ranking agent via a one-way constraint interface that excludes behavioral vulnerability indicators, such that delivery behavior may be adjusted (e.g., pacing, breaks, item limits, resource surfacing) without exposing behavioral vulnerability data to the ranking objective. …
Metadata-Annotated Preference Memory Architecture For Recommendation Systems, Anonymous
Metadata-Annotated Preference Memory Architecture For Recommendation Systems, Anonymous
Defensive Publications Series
Systems and methods are described for managing natural-language preference memory in an agentic recommendation system. A preference memory stores per-user preference chunks as natural-language statements annotated with metadata including confidence with temporal decay, rolling-window exposure count, engagement-quality trend, a source tag distinguishing explicit versus inferred preferences, and a derived satiation indicator. An effective preference weight is computed from decayed confidence, a source-dependent multiplier, and a satiation-based discount, and is supplied to a large language model (LLM) ranking agent to modulate candidate scoring. Preference lifecycle states may be assigned based on effective weight thresholds, and profile summaries may be generated from …
Autonomous Multi-Agent System For Social Discovery And Interpersonal Compatibility Evaluation, Anonymous
Autonomous Multi-Agent System For Social Discovery And Interpersonal Compatibility Evaluation, Anonymous
Defensive Publications Series
Autonomous agent-mediated social discovery is disclosed. Users configure personalized AI agents with personality models, tiered values hierarchies, and conversational behavior parameters. Agents publish anonymized personality embeddings to a discovery service that performs compatibility pre-filtering and a handshake protocol to initiate contact. Matched agents conduct a structured, multi-stage, multi-turn compatibility conversation managed by a conversation orchestrator, with optional early termination based on non-negotiable value misalignment. A compatibility scoring engine generates a multi-dimensional compatibility report derived from conversation signals, including sub-scores for personality alignment, values congruence, and behavioral compatibility. A human supervisory interface enables observation of agent-to-agent interactions, private guidance injection invisible …
Product Ownership Inference From Conversion Signals For Advertising Optimization, Anonymous
Product Ownership Inference From Conversion Signals For Advertising Optimization, Anonymous
Defensive Publications Series
A digital advertising platform constructs a per-user product ownership graph by ingesting third-party conversion signals from multiple advertisers, including purchase events and registration-related events, and resolving each signal to a canonical product entity using catalog matching, fuzzy attribute matching, and cross-merchant deduplication. The platform assigns an ownership confidence score and stores owned product nodes with acquisition metadata, taxonomy, attributes, lifecycle state, and relationship edges such as complementary, successor, substitute, and sequence relationships. During ad selection and auction, the platform suppresses ads for owned products and for satisfied categories, and adjusts ranking to promote complementary products, timed replacements, version upgrades, similarity-based …
Carrier-Modulated Ultrasonic Haptic Feedback System, Anonymous
Carrier-Modulated Ultrasonic Haptic Feedback System, Anonymous
Defensive Publications Series
A carrier-modulated ultrasonic haptic button is described. A piezoelectric actuator is coupled to a button body to form a resonant structure driven at a sonic or ultrasonic carrier frequency, including ultrasonic frequencies. A tactile sensation is produced by modulating an amplitude envelope of the carrier with a lower-frequency modulation signal in a tactile sensitivity range, such that ultrasonic mechanical energy is transferred through the button to a user’s finger while the perceived haptic effect follows the modulation. Mechanical integration may use a flexible seam such as an o-ring, clamped edges, fused seams with isolation features, or a single-piece button-housing structure …
Cryptographic Attestation Of Tool Responses For Gating Autonomous Agent Actions, Anonymous
Cryptographic Attestation Of Tool Responses For Gating Autonomous Agent Actions, Anonymous
Defensive Publications Series
Tool services generate cryptographically signed tool-call receipts that bind a tool-call request and a tool-call response to metadata including a caller identifier, timestamp, and nonce. The receipt includes a request hash and a response hash, and is signed (e.g., Ed25519) over receipt fields. An autonomous agent forwards the signed receipt with a subsequent action request to a downstream gate service. The gate service verifies the signature and enforces gating conditions including request binding to the targeted resource, freshness via the timestamp, and non-replay via nonce consumption. Consequential actions are authorized only when required receipt prerequisites validate; missing receipts, mismatched request …
Multi-Agent Artificial Intelligence Coordination System With Inter-Agent Communication And Task Delegation Protocols, Anonymous
Defensive Publications Series
A multi-agent AI coordination approach uses a collaboration messaging platform as the inter-agent communication bus and a shared natural-language governance document as an executable specification. Multiple persistent autonomous agent instances send inter-agent messages under a human operator identity while embedding a standardized sender-identification header. Messages are routed to persistent per-agent comms threads to consolidate visibility. Each agent operates on an independent periodic heartbeat to check messages, initiate proactive actions, and monitor peer responsiveness without centralized orchestration. The governance document includes self-referential rules requiring that updates to the governance document or other shared documents trigger fan-out notifications so peer agents re-read …
Distributed Ledger-Based Provenance Tracking And Verification For Multi-Agent Ai Knowledge Systems, Anonymous
Distributed Ledger-Based Provenance Tracking And Verification For Multi-Agent Ai Knowledge Systems, Anonymous
Defensive Publications Series
A blockchain-based provenance verification layer is described for text-based knowledge artifacts shared among multiple AI agents. Middleware intercepts creation and consumption of artifacts in shared knowledge stores and associates each artifact or claim with a provenance record on a distributed ledger, including content hash, author identifier, timestamp, verification status, parent links, propagation depth, and confidence. Independent verification agents generate cryptographically signed attestations using techniques such as command help checks, API test calls, cross-references, and consistency checks, and ledger status is updated accordingly, optionally blocking unverified writes. During reads, a trust score is computed from verification status, attestation density, propagation depth, …
Recommendation Systems For Autonomous Agent Tool And Service Selection, Anonymous
Recommendation Systems For Autonomous Agent Tool And Service Selection, Anonymous
Defensive Publications Series
Systems and methods provide machine-facing recommendations for autonomous agents selecting tools and services from large catalogs. A task specification including a task representation, required input and output types, and explicit constraints (cost, latency, reliability, chain length, and context-token budget) is received. Tools are represented with structured descriptors including learned capability vectors, schemas, type tags, and empirically observed performance. A directed composability graph encodes tool-to-tool compatibility using semantic type compatibility, schema compatibility, and historical co-execution success learned from outcomes. Candidate ordered tool chains are generated by traversing the graph under constraints and scored with a multi-objective function including predicted task success, …
Monoculture-Resistant Diversity Enforcement System For Agent-Facing Recommendation Platforms, Anonymous
Monoculture-Resistant Diversity Enforcement System For Agent-Facing Recommendation Platforms, Anonymous
Defensive Publications Series
Techniques are described for enforcing population-level diversity in agent-facing recommendation platforms by attributing autonomous agents to underlying large language model (LLM) provider classes and conditioning ranking on provider-correlated consumption. An attribution function partitions agents by provider using explicit registration and/or implicit behavioral fingerprinting. For items, provider-partitioned consumption counts are used to compute a monoculture index, and provider-conditional penalties or boosts are applied when an item exhibits concentration and the requesting agent’s provider is over-represented. Counterfactual rankings based on models of alternative provider behaviors may be generated and interleaved into mid-ranking positions. A systemic risk score may combine monoculture, popularity, and …
Agent Preference Memory Integrity Verification System For Detecting And Recovering From Preference Poisoning Attacks, Anonymous
Defensive Publications Series
Systems and methods are described for integrity verification of AI agent preference memory. Preference entries include semantic content, confidence, timestamp, source interaction identifier, and reinforcement history, and are associated with cryptographic provenance signatures and interaction context hashes. A preference consistency graph computes embedding-based consistency weights between preferences and produces anomaly scores for candidate preferences based on contradictions with stored high-confidence preferences. Confidence values may decay over time and be re-verified using subsequent behavior and a multi-source corroboration ladder. The system creates cryptographically signed checkpoints and performs targeted rollback to surgically remove unverifiable or anomalous preference entries while preserving verified entries, …
Agent-Human Traffic Differentiation And Dual-Objective Recommendation Serving System, Anonymous
Agent-Human Traffic Differentiation And Dual-Objective Recommendation Serving System, Anonymous
Defensive Publications Series
Techniques are described for differentiating recommendation traffic among human, autonomous agent, and agent-mediated-human sessions and serving rankings using dual objectives. Interaction telemetry is converted into behavioral fingerprinting features and classified into a three-class probability distribution. Based on confidence thresholds, the system selects human-mode ranking using an engagement-trained scoring head, agent-mode ranking using a task-utility-trained scoring head, or blended ranking that combines both head outputs using a weight alpha. Responses may be formatted as rich human-readable content, structured machine-readable content, or a combined representation. Feedback is isolated and routed such that engagement signals train the human head and task-utility signals train …
Content Injection Firewall For Detecting And Neutralizing Adversarial Instructions In Agent-Facing Recommendation Content, Anonymous
Defensive Publications Series
Techniques are described for an agent-mediated content injection firewall in recommendation systems. Candidate recommendation content such as descriptions, reviews, and metadata is evaluated using a dual pipeline that yields a human-oriented quality score and an injection risk score. The injection risk score may be computed as a calibrated ensemble of instruction-pattern matching, obfuscation detection, a transformer-based instruction classifier, and an adversarial judge based on behavioral divergence of a language model when conditioned on the content versus a neutralized version. The injection risk score is integrated into ranking using an agent-adjusted penalty scaled by an agent vulnerability profile to produce soft …
Optimization Budget Allocation System For Preventing Goodhart Breakpoint Crossing In Multi-Layer Recommendation Pipelines, Anonymous
Defensive Publications Series
Techniques are disclosed for coordinating optimization across multi-layer recommendation pipelines in which each layer optimizes a proxy metric. For each layer, an optimization distance between a current policy and a reference policy is measured, for example using expected KL divergence, and constrained by a per-layer budget using training-time penalties and/or serving-time interpolation toward the reference policy. A total optimization pressure is computed from per-layer distances and cross-layer interaction terms weighted by sensitivity between layers, and per-layer budgets are allocated and reallocated under a total budget constraint. Satisficing thresholds cap proxy optimization after diminishing returns and may redirect optimization to secondary …
Behavioral Monitoring And Intervention System For Detecting Emergent Tacit Collusion Among Autonomous Agents In Recommendation Platforms, Anonymous
Defensive Publications Series
Behavioral monitoring for recommendation marketplaces is described. Observable time-series actions of autonomous agents and platform signals are ingested per market segment. A competitive-equilibrium baseline is computed and deviations from the baseline are measured. Coordination evidence is derived without inspecting agent internals, including behavioral correlation beyond common cause using Granger-causality-based analysis, reward-punishment dynamics detected from deviation events and retaliatory responses, and platform-mediated signaling exploitation measured by mutual information between platform outputs and subsequent actions. A composite collusion score aggregates component measures using calibrated weights and produces tiered alerts. A causal audit trail may be generated with timelines and analyses and may …
Cascading Proxy Drift Detection And Corrective Action System For Multi-Layer Optimization Pipelines, Anonymous
Cascading Proxy Drift Detection And Corrective Action System For Multi-Layer Optimization Pipelines, Anonymous
Defensive Publications Series
Systems and methods are described for detecting and correcting cascading proxy drift in multi-layer optimization pipelines. A directed acyclic graph represents proxy metrics across layers, including a ground-truth metric and downstream proxies. For each adjacent layer pair, a drift state is computed using sliding-window Spearman correlation, a trend estimate, and volatility, with adaptive window sizing via change-point detection. Cascade onset is detected based on a product of per-layer correlations and/or concurrent negative drift across multiple layers. Upon cascade detection, a drift source is localized using intervention-based tests including sequential freeze, counterfactual correlation estimation, and Shapley-value attribution (optionally via permutation sampling). …
Recommendation Circuit Breaker System For Agent-Speed Feedback Loop Stability, Anonymous
Recommendation Circuit Breaker System For Agent-Speed Feedback Loop Stability, Anonymous
Defensive Publications Series
Systems and methods are disclosed for stabilizing recommendation rankings subject to fast feedback-loop dynamics. A circuit breaker layer monitors ranking position time series to compute ranking velocity and a catalog-level ranking volatility index, detects oscillations via frequency-domain analysis, and measures exposure concentration using concentration indices. The system may additionally detect absolute quality drift using calibration probe items and a probe-to-organic engagement gap. A composite instability score selects among graduated intervention tiers, including diversity-weight adjustment, ranking velocity capping, ranking freezes using cached stable rankings, and fallback rankings based on longer-horizon popularity. The system may classify sessions as agent-originated or human-originated using …
Psychological Reasoning Validation For Agentic Recommendation Systems, Anonymous
Psychological Reasoning Validation For Agentic Recommendation Systems, Anonymous
Defensive Publications Series
A validation layer for agentic recommendation systems validates psychological reasoning used by an LLM-based recommendation agent. A non-LLM behavioral ground truth estimator computes psychological-state estimates from behavioral signals with confidence values. A non-LLM reasoning extractor parses agent reasoning traces into structured psychological-state claims. Claims are compared to ground truth subject to confidence thresholds to compute per-state reasoning-quality metrics including precision, recall, and hallucination rate, optionally at granularities such as per topic, per user segment, and per prompt version. A hallucination detection pipeline may apply tiered responses including logging, soft override via ground truth injection, and hard override for safety-relevant actions. …
Position-Dependent Psychological Risk Management For Recommendation Ranking, Anonymous
Position-Dependent Psychological Risk Management For Recommendation Ranking, Anonymous
Defensive Publications Series
A recommendation system constructs a ranked slate using position-dependent psychological risk management. The system partitions slate positions into contiguous zones including a Trust zone, an Engagement zone, and a Discovery zone, with zone boundaries adapted per user based on a user trust score and optionally cognitive mode or session progress. A base ranking model produces base value scores and an uncertainty measure. For each target position, the system computes a position-modified score by applying a zone-specific risk tolerance function conditioned on user state. In the Trust zone, items are penalized as a function of prediction uncertainty scaled by user distrust; …
Feedback Loop Intensity Monitoring And Automatic Intervention For Recommendation Systems, Anonymous
Feedback Loop Intensity Monitoring And Automatic Intervention For Recommendation Systems, Anonymous
Defensive Publications Series
Systems and methods are described for monitoring recommendation feedback loops and automatically intervening in live ranking. A feedback loop intensity (FLI) for a user and topic is computed as a ratio of a user-conditional topic serving proportion to a global topic serving proportion over a rolling window. An organic interest score is computed from non-recommendation-channel evidence including one or more of search, direct navigation, creator following, or external engagement. An adjusted FLI is generated by dampening the raw FLI as a function of organic interest. When adjusted FLI exceeds configurable thresholds, a controller applies topic-specific, tiered interventions in the ranking …
Unified Dynamic User State Vector With Cross-State Conflict Resolution For Recommendation Ranking, Anonymous
Unified Dynamic User State Vector With Cross-State Conflict Resolution For Recommendation Ranking, Anonymous
Defensive Publications Series
Systems and methods are disclosed for recommendation ranking using a unified, explicitly decomposed user psychological state vector. A coordination layer loads per-user state values including mood, trust, cognitive mode, topic-indexed satiation, and topic-indexed curiosity, constructs explicit cross-state interaction features (including trust×curiosity, mood×satiation, mode×satiation, and trust×mood), and provides these to a ranking model to compute base scores for candidate items. A conflict resolution engine detects contradictory ranking actions implied by different state dimensions and applies a dominance hierarchy with cascading hard and soft constraints, including mood-based content suppression and trust-gated exploration as hard constraints and satiation discounting as a soft constraint. …