# The lex records whose category is model-architecture

This index as JSON: https://banes-lab.com/json/api/facets/lex/category/model-architecture

## Entries

- [Ad-Hoc Notebook-to-Production](https://banes-lab.com/records/lex/ad-hoc-notebook-to-production.md): Promoting exploratory notebook code straight to production without engineering it into a reliable pipeline.
- [Deploy-and-Forget Models](https://banes-lab.com/records/lex/deploy-and-forget-models.md): Deploying a model and never monitoring it, so degradation as the data drifts goes unnoticed.
- [Exact Keyword Search Only](https://banes-lab.com/records/lex/exact-keyword-search-only.md): Relying solely on exact keyword matching for retrieval, missing semantically related results.
- [Flat Document-Only Knowledge](https://banes-lab.com/records/lex/flat-document-only-knowledge.md): Representing knowledge as unlinked flat documents, losing the relationships a graph would capture.
- [Opaque Black-Box Decisions](https://banes-lab.com/records/lex/opaque-black-box-decisions.md): Producing model decisions with no explanation, so their reasoning cannot be inspected or trusted.
- [Opaque Ungoverned Model Use](https://banes-lab.com/records/lex/opaque-ungoverned-model-use.md): Using models with no governance or oversight, leaving their behavior and risks unmanaged.
- [Training-Time-Only Model Logic](https://banes-lab.com/records/lex/training-time-only-model-logic.md): Building logic that exists only during training, with no counterpart to serve predictions at inference.
- [Unapproved Model Deployment](https://banes-lab.com/records/lex/unapproved-model-deployment.md): Deploying a model to production without passing the required review and approval gates.
- [Ungrounded Generation](https://banes-lab.com/records/lex/ungrounded-generation.md): Generating output from a model alone without grounding it in retrieved facts, inviting hallucination.
- [Unguarded Model Autonomy](https://banes-lab.com/records/lex/unguarded-model-autonomy.md): Letting a model act autonomously with no safety guardrails on what it can do.
- [Untested Model Deployment](https://banes-lab.com/records/lex/untested-model-deployment.md): Deploying a model without evaluating it, so its real-world quality is unknown until it fails.
- [Approval Policy](https://banes-lab.com/records/lex/approval-policy.md): The declared rules and gates a model must pass before it may be deployed.
- [Document Store](https://banes-lab.com/records/lex/document-store.md): The body of documents a retrieval system searches to ground a model's generation.
- [Data Pipeline](https://banes-lab.com/records/lex/data-pipeline.md): The stages that ingest, clean, and transform data into a form suitable for training or inference.
- [Data/Model Boundaries](https://banes-lab.com/records/lex/data-model-boundaries.md): The lines separating data preparation, model training, and serving so each concern stays isolated.
- [Dataset](https://banes-lab.com/records/lex/dataset.md): A curated collection of examples used to train or evaluate a model.
- [Embeddings](https://banes-lab.com/records/lex/embeddings.md): Numeric vector representations of data that place semantically similar items near each other.
- [Entities](https://banes-lab.com/records/lex/entities.md): The distinct things, such as people, places or concepts, that a knowledge graph represents as nodes.
- [Grounding Strategy](https://banes-lab.com/records/lex/grounding-strategy.md): A scheme for anchoring a model's output in retrieved, authoritative sources rather than its parameters alone.
- [Guardrails](https://banes-lab.com/records/lex/guardrails.md): Constraints and filters that bound what a model is permitted to output or do at runtime.
- [Input/Output Contract](https://banes-lab.com/records/lex/input-output-contract.md): The agreed schema of the inputs a model accepts and the outputs it returns.
- [Model Artifact](https://banes-lab.com/records/lex/model-artifact.md): The trained model file, with its learned weights, that is loaded to serve predictions.
- [Model Registry](https://banes-lab.com/records/lex/model-registry.md): A catalog that tracks model versions, their metadata, and their deployment status.
- [Rationale/Evidence](https://banes-lab.com/records/lex/rationale-evidence.md): The reasons and supporting evidence recorded for a model's decision.
- [Relations](https://banes-lab.com/records/lex/relations.md): The typed connections between entities that a knowledge graph represents as edges.
- [Retriever](https://banes-lab.com/records/lex/retriever.md): A component that finds and returns the most relevant documents for a query.
- [Schema/Ontology](https://banes-lab.com/records/lex/schema-ontology.md): A formal definition of the entity types and relationship types a knowledge graph may contain.
- [Tool Interface](https://banes-lab.com/records/lex/tool-interface.md): The defined contract through which an agent invokes external tools and receives their results.
- [Training/Inference Separation](https://banes-lab.com/records/lex/training-inference-separation.md): The requirement that model training and prediction serving be distinct, separately-managed phases.
- [Vector Index](https://banes-lab.com/records/lex/vector-index.md): A data structure that organizes embedding vectors for fast nearest-neighbor lookup.
- [Retraining Triggers](https://banes-lab.com/records/lex/retraining-triggers.md): Signals that automatically initiate model retraining when measured drift crosses a threshold.
- [Model-Integrated Systems](https://banes-lab.com/records/lex/model-integrated-systems.md): The ability of a software system to incorporate models as integral parts of its behavior.
- [Audit and Debugging](https://banes-lab.com/records/lex/audit-and-debugging.md): The ability to inspect and trace a model's decisions for auditing and debugging.
- [Bounded Tool-Using Agents](https://banes-lab.com/records/lex/bounded-tool-using-agents.md): The ability to run agents that use external tools within defined, safe limits.
- [Contextual Generation](https://banes-lab.com/records/lex/contextual-generation.md): The ability to generate output informed by retrieved, task-specific context.
- [Controlled Model Deployment](https://banes-lab.com/records/lex/controlled-model-deployment.md): The ability to release models through a governed, approved process.
- [Degradation Detection](https://banes-lab.com/records/lex/degradation-detection.md): The ability to detect when a model's accuracy declines as data shifts.
- [Governed Autonomy](https://banes-lab.com/records/lex/governed-autonomy.md): The ability to let an agent act autonomously within enforced governance limits.
- [Model Selection/Regression Detection](https://banes-lab.com/records/lex/model-selection-regression-detection.md): The ability to compare models and catch quality regressions before deployment.
- [Relationship-Aware Retrieval/Reasoning](https://banes-lab.com/records/lex/relationship-aware-retrieval-reasoning.md): The ability to retrieve and reason over the relationships between entities, not just isolated facts.
- [Reliable Model Lifecycle](https://banes-lab.com/records/lex/reliable-model-lifecycle.md): The ability to manage a model's data, training, deployment and monitoring reliably and repeatably.
- [Runtime Prediction/Generation](https://banes-lab.com/records/lex/runtime-prediction-generation.md): The ability to produce predictions or generated output from a trained model at runtime.
- [Safe Model Deployment](https://banes-lab.com/records/lex/safe-model-deployment.md): The ability to deploy models with safeguards that bound their behavior.
- [Semantic Search](https://banes-lab.com/records/lex/semantic-search.md): The ability to find results by meaning and similarity rather than exact keyword match.
- [Similarity Retrieval](https://banes-lab.com/records/lex/similarity-retrieval.md): The ability to retrieve items nearest to a query in an embedding space.
- [Structured, Versioned Prompts](https://banes-lab.com/records/lex/structured-versioned-prompts.md): Prompts authored as structured, version-controlled artifacts rather than ad-hoc strings.
- [Knowledge Freshness](https://banes-lab.com/records/lex/knowledge-freshness.md): The degree to which a system's knowledge reflects current rather than stale information.
- [Trust](https://banes-lab.com/records/lex/trust.md): The degree to which users are willing to rely on a system's outputs.
- [Capability/Utility](https://banes-lab.com/records/lex/capability-utility.md): The degree of usefulness a model offers, which strict safety limits can constrain.
- [Curation Cost](https://banes-lab.com/records/lex/curation-cost.md): The degree of ongoing effort required to build and maintain a curated knowledge graph.
- [Experiment Velocity](https://banes-lab.com/records/lex/experiment-velocity.md): The rate at which model experiments can be run and iterated, which governance can slow.
- [Experimentation Speed](https://banes-lab.com/records/lex/experimentation-speed.md): The rate at which new modeling ideas can be tried and evaluated.
- [Explainability/Recall](https://banes-lab.com/records/lex/explainability-recall.md): The degree to which retrieval stays explainable and complete, traded against pure similarity ranking.
- [Latency/Cost](https://banes-lab.com/records/lex/latency-cost.md): The degree of latency and expense incurred to serve model predictions.
- [Metric Completeness](https://banes-lab.com/records/lex/metric-completeness.md): The degree to which evaluation metrics capture every dimension of a model's quality.
- [Model Complexity](https://banes-lab.com/records/lex/model-complexity.md): The degree of intricacy in a model, which raises accuracy but lowers explainability.
- [Monitoring Cost](https://banes-lab.com/records/lex/monitoring-cost.md): The degree of ongoing expense of continuously monitoring a deployed model.
- [Retrieval Quality/Latency](https://banes-lab.com/records/lex/retrieval-quality-latency.md): The degree to which retrieval must trade result quality against speed.