# Vercy AI instruction - YAML 1.2 (JSON-compatible) { "vercy": "1.0-draft", "publication": { "status": "published", "adjudicationStatus": "reviewable-draft", "publishableCanonical": false, "generatedAt": "2026-09-06T11:34:09Z", "synthesisSha256": "10b715cdbf465f6ed22318fb92b07208e1a7306b59337ab84540a26729f647b8", "providerMode": "single-provider-waiver", "providers": [ "Codex" ], "waivedProviders": [ "Claude", "Grok" ] }, "metaModel": { "id": "WM-AI-006", "registryId": "vr.wm-ai-006", "name": "Model Training / Fine-tuning Run", "version": "0.3.0-research.1", "previousVersions": [], "entryKind": "aggregate", "family": "World Models", "category": "Information and virtual systems", "industry": [ "Cross-industry" ], "domain": [ "INF.AI.TRN" ], "tags": [ "model", "training", "fine", "tuning", "run", "inf.ai.trn" ], "status": "published" }, "canonicalUrl": "https://ver.cy/models/wm-ai-006-model-training-fine-tuning-run/", "sourceUrl": "https://github.com/ver-cy/world-models/tree/feat/mega-model-registry/publications/wm-ai-006-model-training-fine-tuning-run", "model": { "registry_id": "vr.wm-ai-006", "model_id": "WM-AI-006", "name": "Model Training / Fine-tuning Run", "entry_kind": "aggregate", "purpose": "Represent one governed execution that transforms version-qualified model, data, code and configuration inputs into candidate model artifacts with reconstructable progress, resources, evidence, lineage and outcome.", "scope_statement": "Owns one model training or fine-tuning run identity; objective, method and authority; immutable bindings to base model, tokenizer, datasets, code, configuration and environment; topology, stages, attempts, progress, resources, checkpoints, metrics, validation and safety evidence; produced-artifact bindings, lineage, reproducibility limits, terminal outcome, access, correction, retention and projections. Dataset, source code, base model, trained model artifact, evaluation, registry, deployment, infrastructure, secret, policy, provenance, audit and records masters remain external.", "in_scope": [ "Run identity, experiment and job bindings, objective, method, risk, authority, input versions, configuration, environment, topology, stages, attempts and progress", "Resources, costs, energy, checkpoints, metrics, validation and safety references, candidates, derivation, reproducibility, outcome, access, correction, retention and projections" ], "out_of_scope": [ "Creating or mutating external dataset, code, base-model, trained-model, evaluation, registry, deployment, infrastructure, secret, policy, provenance, audit or records masters", "Equating a run with an experiment, checkpoint, trained model, registry entry or deployment, or equating requested resources with observed use", "Autonomous training, unrestricted compute allocation, protected-data or secret access, privacy or rights waiver, release, deployment or destructive cleanup" ], "boundary_notes": [ { "neighbor": "WM-DAT-001 Dataset", "distinction": "The candidate REFERENCE relation binds version-qualified dataset roles, splits, permissions and transformations. Dataset content, rights and lifecycle remain external.", "source_refs": [ "SRC-002", "SRC-004", "SRC-005", "SRC-006", "SRC-008", "SRC-009" ] }, { "neighbor": "WM-SFT-004 produced model artifact", "distinction": "The candidate PRODUCES relation records derivation and candidate selection. The artifact master, registry promotion, release and deployment remain external.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012" ] }, { "neighbor": "Experiment, pipeline, job, stage, task, attempt and checkpoint", "distinction": "The run is one execution aggregate; reusable definitions and independently addressable execution children retain distinct identities and provenance.", "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010" ] }, { "neighbor": "AI evaluation, model registry and deployment", "distinction": "The run may reference evaluations and emit candidates, but evaluation conclusions, promotion decisions, registry state and deployment state are external authorities.", "source_refs": [ "SRC-001", "SRC-002", "SRC-004", "SRC-008", "SRC-014" ] }, { "neighbor": "Infrastructure, telemetry, cost and environmental systems", "distinction": "External systems own allocation, billing, energy and carbon records. The run stores method-bound requested, allocated and observed references and summaries.", "source_refs": [ "SRC-010", "SRC-013", "SRC-015" ] }, { "neighbor": "MLflow, MLMD, OpenLineage, PROV, Kubeflow, SLSA, OCI, OpenTelemetry, MLPerf, SCI and PyTorch", "distinction": "These are versioned experiment, metadata, lineage, runtime, provenance, packaging, telemetry, benchmark, carbon and framework profiles. No mapping is universally applicable or assumed lossless.", "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-017" ] } ] }, "sources": [ { "id": "SRC-001", "title": "Artificial Intelligence Risk Management Framework (AI RMF 1.0)", "organization": "National Institute of Standards and Technology", "url": "https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10", "version_or_date": "NIST AI 100-1, 26 January 2023; revision in progress at access", "source_type": "public-authority", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Frames governed AI lifecycle risk, accountability, measurement and documentation." }, { "id": "SRC-002", "title": "Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile", "organization": "National Institute of Standards and Technology", "url": "https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf", "version_or_date": "NIST AI 600-1, July 2024", "source_type": "public-authority", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Adds training-data, privacy, security, content provenance, pre-deployment testing, incident and environmental risk considerations for generative AI." }, { "id": "SRC-003", "title": "Secure Software Development Practices for Generative AI and Dual-Use Foundation Models", "organization": "National Institute of Standards and Technology", "url": "https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-218A.pdf", "version_or_date": "NIST SP 800-218A, July 2024", "source_type": "public-authority", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Extends secure development practices to model, data, component and lifecycle evidence." }, { "id": "SRC-004", "title": "Regulation (EU) 2024/1689 Artificial Intelligence Act", "organization": "European Union", "url": "https://eur-lex.europa.eu/eli/reg/2024/1689/oj", "version_or_date": "13 June 2024 official journal text", "source_type": "legislation", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Provides an EU legal profile for data governance, technical documentation, records, transparency, risk and general-purpose AI obligations." }, { "id": "SRC-005", "title": "Regulation (EU) 2016/679 General Data Protection Regulation", "organization": "European Union", "url": "https://eur-lex.europa.eu/eli/reg/2016/679/oj", "version_or_date": "27 April 2016; applicable from 25 May 2018", "source_type": "legislation", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Provides an EU privacy profile for lawful processing, purpose, minimization, rights, security and accountability." }, { "id": "SRC-006", "title": "PROV-O: The PROV Ontology", "organization": "World Wide Web Consortium", "url": "https://www.w3.org/TR/prov-o/", "version_or_date": "W3C Recommendation 30 April 2013", "source_type": "ontology", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Defines entities, activities, agents, derivation, association, attribution, revision and time for run lineage." }, { "id": "SRC-007", "title": "OpenLineage Object Model", "organization": "OpenLineage", "url": "https://openlineage.io/docs/spec/object-model/", "version_or_date": "Specification 1.53.0 current at access", "source_type": "schema", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Separates Job design, Run execution, Dataset and RunEvent identities and recommends UUIDv7 run identifiers." }, { "id": "SRC-008", "title": "MLflow Tracking", "organization": "MLflow", "url": "https://mlflow.org/docs/latest/ml/tracking/", "version_or_date": "Latest first-party documentation at access", "source_type": "first-party-doc", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Defines experiments, runs, parameters, metrics, timestamps, datasets and artifacts while keeping registered models distinct." }, { "id": "SRC-009", "title": "ML Metadata", "organization": "TensorFlow", "url": "https://www.tensorflow.org/tfx/guide/mlmd", "version_or_date": "First-party documentation current at access", "source_type": "first-party-doc", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Defines Artifact, Execution, Event and Context metadata for ML pipeline lineage." }, { "id": "SRC-010", "title": "Kubeflow Trainer Overview", "organization": "Kubeflow", "url": "https://www.kubeflow.org/docs/components/trainer/overview/", "version_or_date": "Kubeflow Trainer 2.1.0 profile current at access", "source_type": "first-party-doc", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Provides a bounded distributed training runtime profile with train jobs, runtime, initializer, trainer and resources." }, { "id": "SRC-011", "title": "SLSA Terminology", "organization": "Open Source Security Foundation", "url": "https://slsa.dev/spec/v1.1/terminology", "version_or_date": "SLSA 1.1", "source_type": "standard", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Defines artifact, build, builder, dependency and provenance concepts for generated model and container artifacts." }, { "id": "SRC-012", "title": "Open Container Initiative Image Format Specification", "organization": "Open Container Initiative", "url": "https://github.com/opencontainers/image-spec/tree/v1.1.1", "version_or_date": "OCI Image Specification 1.1.1, 3 March 2025", "source_type": "standard", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Provides a release-pinned container image manifest and digest profile for runtime environments." }, { "id": "SRC-013", "title": "OpenTelemetry Specification", "organization": "OpenTelemetry", "url": "https://opentelemetry.io/docs/specs/otel/", "version_or_date": "Specification 1.60.0; semantic conventions 1.44.0 at access", "source_type": "standard", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Provides versioned traces, metrics, logs, resources and context propagation for run telemetry." }, { "id": "SRC-014", "title": "MLPerf Training", "organization": "MLCommons", "url": "https://mlcommons.org/benchmarks/training/", "version_or_date": "MLPerf Training v6.0 current at access", "source_type": "first-party-doc", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Provides a bounded benchmark profile based on dataset, quality target and time-to-quality with official rules as source of truth." }, { "id": "SRC-015", "title": "Software Carbon Intensity Specification", "organization": "Green Software Foundation", "url": "https://sci.greensoftware.foundation/", "version_or_date": "SCI 1.1.0", "source_type": "standard", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Defines a method-bound carbon intensity rate using operational emissions, embodied emissions and functional unit." }, { "id": "SRC-016", "title": "Date and Time on the Internet: Timestamps", "organization": "Internet Engineering Task Force", "url": "https://www.rfc-editor.org/info/rfc3339/", "version_or_date": "RFC 3339, July 2002", "source_type": "standard", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "Defines interoperable timestamps with seconds and an explicit UTC relationship." }, { "id": "SRC-017", "title": "Reproducibility", "organization": "PyTorch", "url": "https://docs.pytorch.org/docs/2.14/notes/randomness.html", "version_or_date": "PyTorch 2.14 documentation, updated 14 May 2026", "source_type": "first-party-doc", "primary_source": true, "authority_tier": 1, "accessed_at": "2026-09-06T14:35:00Z", "relevance": "States that full reproducibility is not guaranteed across releases or platforms and documents bounded controls for randomness and nondeterminism." } ], "structure": { "bundles": [ { "id": "run-identity-objective-method-and-authority", "name": "Run identity, objective, method and authority", "description": "Groups governed training-run context for run identity, objective, method and authority.", "rationale": "The run is one governed execution aggregate, not the experiment definition, data master or trained model master.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-006", "SRC-007", "SRC-008", "SRC-009" ], "layers": [ { "id": "run-root-experiment-parent-and-definition", "name": "Run root, experiment, parent and definition", "description": "Groups source-qualified training-run context for run root, experiment, parent and definition.", "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009" ], "findings": [ { "id": "run-root-identity-namespace-owner-revision-and-current-head", "name": "Run root identity, namespace, owner, revision and current head", "description": "Records run root identity, namespace, owner, revision and current head as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009" ], "questions": [ { "id": "run-root-identity-namespace-owner-revision-and-current-head-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish run root identity, namespace, owner, revision and current head?", "kind": "identity", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "run-root-identity-namespace-owner-revision-and-current-head-q02", "text": "Who may declare, execute, observe, review, correct or rely on run root identity, namespace, owner, revision and current head, under which authority and limits?", "kind": "composition", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "run-root-identity-namespace-owner-revision-and-current-head-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to run root identity, namespace, owner, revision and current head, and which evidence supports them?", "kind": "privacy", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "run-root-identity-namespace-owner-revision-and-current-head-data", "name": "Run root identity, namespace, owner, revision and current head data", "description": "Typed data for run root identity, namespace, owner, revision and current head with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009" ] } ], "artifacts": [ { "id": "run-root-identity-namespace-owner-revision-and-current-head-record", "name": "Run root identity, namespace, owner, revision and current head record", "description": "Immutable or successor-versioned training evidence for run root identity, namespace, owner, revision and current head.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for run-root-identity-namespace-owner-revision-and-current-head; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009" ] } ], "inline_only_rationale": null }, { "id": "experiment-parent-pipeline-job-definition-and-correlation-binding", "name": "Experiment, parent, pipeline, job definition and correlation binding", "description": "Records experiment, parent, pipeline, job definition and correlation binding as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009" ], "questions": [ { "id": "experiment-parent-pipeline-job-definition-and-correlation-binding-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish experiment, parent, pipeline, job definition and correlation binding?", "kind": "relationship", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "experiment-parent-pipeline-job-definition-and-correlation-binding-q02", "text": "Who may declare, execute, observe, review, correct or rely on experiment, parent, pipeline, job definition and correlation binding, under which authority and limits?", "kind": "evidence", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "experiment-parent-pipeline-job-definition-and-correlation-binding-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to experiment, parent, pipeline, job definition and correlation binding, and which evidence supports them?", "kind": "lifecycle", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "experiment-parent-pipeline-job-definition-and-correlation-binding-data", "name": "Experiment, parent, pipeline, job definition and correlation binding data", "description": "Typed data for experiment, parent, pipeline, job definition and correlation binding with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009" ] } ], "artifacts": [ { "id": "experiment-parent-pipeline-job-definition-and-correlation-binding-record", "name": "Experiment, parent, pipeline, job definition and correlation binding record", "description": "Immutable or successor-versioned training evidence for experiment, parent, pipeline, job definition and correlation binding.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for experiment-parent-pipeline-job-definition-and-correlation-binding; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009" ] } ], "inline_only_rationale": null } ] }, { "id": "objective-training-method-risk-and-authority", "name": "Objective, training method, risk and authority", "description": "Groups source-qualified training-run context for objective, training method, risk and authority.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004" ], "findings": [ { "id": "task-objective-target-hypothesis-acceptance-and-stop-plan", "name": "Task, objective, target, hypothesis, acceptance and stop plan", "description": "Records task, objective, target, hypothesis, acceptance and stop plan as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004" ], "questions": [ { "id": "task-objective-target-hypothesis-acceptance-and-stop-plan-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish task, objective, target, hypothesis, acceptance and stop plan?", "kind": "requirement", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "task-objective-target-hypothesis-acceptance-and-stop-plan-q02", "text": "Who may declare, execute, observe, review, correct or rely on task, objective, target, hypothesis, acceptance and stop plan, under which authority and limits?", "kind": "ownership", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "task-objective-target-hypothesis-acceptance-and-stop-plan-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to task, objective, target, hypothesis, acceptance and stop plan, and which evidence supports them?", "kind": "quality", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "task-objective-target-hypothesis-acceptance-and-stop-plan-data", "name": "Task, objective, target, hypothesis, acceptance and stop plan data", "description": "Typed data for task, objective, target, hypothesis, acceptance and stop plan with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004" ] } ], "artifacts": [ { "id": "task-objective-target-hypothesis-acceptance-and-stop-plan-record", "name": "Task, objective, target, hypothesis, acceptance and stop plan record", "description": "Immutable or successor-versioned training evidence for task, objective, target, hypothesis, acceptance and stop plan.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for task-objective-target-hypothesis-acceptance-and-stop-plan; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004" ] } ], "inline_only_rationale": null }, { "id": "pretraining-finetuning-adaptation-method-risk-profile-and-accountable-authority", "name": "Pretraining, fine-tuning, adaptation method, risk profile and accountable authority", "description": "Records pretraining, fine-tuning, adaptation method, risk profile and accountable authority as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004" ], "questions": [ { "id": "pretraining-finetuning-adaptation-method-risk-profile-and-accountable-authority-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish pretraining, fine-tuning, adaptation method, risk profile and accountable authority?", "kind": "classification", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "pretraining-finetuning-adaptation-method-risk-profile-and-accountable-authority-q02", "text": "Who may declare, execute, observe, review, correct or rely on pretraining, fine-tuning, adaptation method, risk profile and accountable authority, under which authority and limits?", "kind": "measurement", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "pretraining-finetuning-adaptation-method-risk-profile-and-accountable-authority-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to pretraining, fine-tuning, adaptation method, risk profile and accountable authority, and which evidence supports them?", "kind": "security", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "pretraining-finetuning-adaptation-method-risk-profile-and-accountable-authority-data", "name": "Pretraining, fine-tuning, adaptation method, risk profile and accountable authority data", "description": "Typed data for pretraining, fine-tuning, adaptation method, risk profile and accountable authority with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004" ] } ], "artifacts": [ { "id": "pretraining-finetuning-adaptation-method-risk-profile-and-accountable-authority-record", "name": "Pretraining, fine-tuning, adaptation method, risk profile and accountable authority record", "description": "Immutable or successor-versioned training evidence for pretraining, fine-tuning, adaptation method, risk profile and accountable authority.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for pretraining-finetuning-adaptation-method-risk-profile-and-accountable-authority; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004" ] } ], "inline_only_rationale": null } ] } ] }, { "id": "model-data-code-configuration-and-environment-bindings", "name": "Model, data, code, configuration and environment bindings", "description": "Groups governed training-run context for model, data, code, configuration and environment bindings.", "rationale": "Immutable or version-qualified inputs must be resolvable without copying external masters into the run.", "source_refs": [ "SRC-002", "SRC-003", "SRC-004", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-017" ], "layers": [ { "id": "base-model-tokenizer-dataset-and-split-bindings", "name": "Base model, tokenizer, dataset and split bindings", "description": "Groups source-qualified training-run context for base model, tokenizer, dataset and split bindings.", "source_refs": [ "SRC-002", "SRC-004", "SRC-006", "SRC-008", "SRC-009" ], "findings": [ { "id": "base-model-architecture-tokenizer-initialization-freeze-and-adapter-bindings", "name": "Base model, architecture, tokenizer, initialization, freeze and adapter bindings", "description": "Records base model, architecture, tokenizer, initialization, freeze and adapter bindings as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-002", "SRC-004", "SRC-006", "SRC-008", "SRC-009" ], "questions": [ { "id": "base-model-architecture-tokenizer-initialization-freeze-and-adapter-bindings-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish base model, architecture, tokenizer, initialization, freeze and adapter bindings?", "kind": "composition", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "base-model-architecture-tokenizer-initialization-freeze-and-adapter-bindings-q02", "text": "Who may declare, execute, observe, review, correct or rely on base model, architecture, tokenizer, initialization, freeze and adapter bindings, under which authority and limits?", "kind": "exception", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "base-model-architecture-tokenizer-initialization-freeze-and-adapter-bindings-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to base model, architecture, tokenizer, initialization, freeze and adapter bindings, and which evidence supports them?", "kind": "retention", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "base-model-architecture-tokenizer-initialization-freeze-and-adapter-bindings-data", "name": "Base model, architecture, tokenizer, initialization, freeze and adapter bindings data", "description": "Typed data for base model, architecture, tokenizer, initialization, freeze and adapter bindings with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-002", "SRC-004", "SRC-006", "SRC-008", "SRC-009" ] } ], "artifacts": [ { "id": "base-model-architecture-tokenizer-initialization-freeze-and-adapter-bindings-record", "name": "Base model, architecture, tokenizer, initialization, freeze and adapter bindings record", "description": "Immutable or successor-versioned training evidence for base model, architecture, tokenizer, initialization, freeze and adapter bindings.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for base-model-architecture-tokenizer-initialization-freeze-and-adapter-bindings; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-002", "SRC-004", "SRC-006", "SRC-008", "SRC-009" ] } ], "inline_only_rationale": null }, { "id": "dataset-role-snapshot-mixture-split-transform-sampling-permission-and-quality", "name": "Dataset role, snapshot, mixture, split, transform, sampling, permission and quality", "description": "Records dataset role, snapshot, mixture, split, transform, sampling, permission and quality as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-002", "SRC-004", "SRC-006", "SRC-008", "SRC-009" ], "questions": [ { "id": "dataset-role-snapshot-mixture-split-transform-sampling-permission-and-quality-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish dataset role, snapshot, mixture, split, transform, sampling, permission and quality?", "kind": "privacy", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "dataset-role-snapshot-mixture-split-transform-sampling-permission-and-quality-q02", "text": "Who may declare, execute, observe, review, correct or rely on dataset role, snapshot, mixture, split, transform, sampling, permission and quality, under which authority and limits?", "kind": "provenance", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "dataset-role-snapshot-mixture-split-transform-sampling-permission-and-quality-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to dataset role, snapshot, mixture, split, transform, sampling, permission and quality, and which evidence supports them?", "kind": "interoperability", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "dataset-role-snapshot-mixture-split-transform-sampling-permission-and-quality-data", "name": "Dataset role, snapshot, mixture, split, transform, sampling, permission and quality data", "description": "Typed data for dataset role, snapshot, mixture, split, transform, sampling, permission and quality with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-002", "SRC-004", "SRC-006", "SRC-008", "SRC-009" ] } ], "artifacts": [ { "id": "dataset-role-snapshot-mixture-split-transform-sampling-permission-and-quality-record", "name": "Dataset role, snapshot, mixture, split, transform, sampling, permission and quality record", "description": "Immutable or successor-versioned training evidence for dataset role, snapshot, mixture, split, transform, sampling, permission and quality.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for dataset-role-snapshot-mixture-split-transform-sampling-permission-and-quality; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-002", "SRC-004", "SRC-006", "SRC-008", "SRC-009" ] } ], "inline_only_rationale": null } ] }, { "id": "code-parameters-dependencies-and-runtime-environment", "name": "Code, parameters, dependencies and runtime environment", "description": "Groups source-qualified training-run context for code, parameters, dependencies and runtime environment.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-017" ], "findings": [ { "id": "source-code-revision-entrypoint-configuration-hyperparameters-and-seeds", "name": "Source code revision, entrypoint, configuration, hyperparameters and seeds", "description": "Records source code revision, entrypoint, configuration, hyperparameters and seeds as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-017" ], "questions": [ { "id": "source-code-revision-entrypoint-configuration-hyperparameters-and-seeds-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish source code revision, entrypoint, configuration, hyperparameters and seeds?", "kind": "provenance", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "source-code-revision-entrypoint-configuration-hyperparameters-and-seeds-q02", "text": "Who may declare, execute, observe, review, correct or rely on source code revision, entrypoint, configuration, hyperparameters and seeds, under which authority and limits?", "kind": "process", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "source-code-revision-entrypoint-configuration-hyperparameters-and-seeds-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to source code revision, entrypoint, configuration, hyperparameters and seeds, and which evidence supports them?", "kind": "decision", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "source-code-revision-entrypoint-configuration-hyperparameters-and-seeds-data", "name": "Source code revision, entrypoint, configuration, hyperparameters and seeds data", "description": "Typed data for source code revision, entrypoint, configuration, hyperparameters and seeds with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-017" ] } ], "artifacts": [ { "id": "source-code-revision-entrypoint-configuration-hyperparameters-and-seeds-record", "name": "Source code revision, entrypoint, configuration, hyperparameters and seeds record", "description": "Immutable or successor-versioned training evidence for source code revision, entrypoint, configuration, hyperparameters and seeds.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for source-code-revision-entrypoint-configuration-hyperparameters-and-seeds; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-017" ] } ], "inline_only_rationale": null }, { "id": "packages-images-framework-compiler-driver-hardware-and-environment", "name": "Packages, images, framework, compiler, driver, hardware and environment", "description": "Records packages, images, framework, compiler, driver, hardware and environment as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-017" ], "questions": [ { "id": "packages-images-framework-compiler-driver-hardware-and-environment-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish packages, images, framework, compiler, driver, hardware and environment?", "kind": "interoperability", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "packages-images-framework-compiler-driver-hardware-and-environment-q02", "text": "Who may declare, execute, observe, review, correct or rely on packages, images, framework, compiler, driver, hardware and environment, under which authority and limits?", "kind": "validation", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "packages-images-framework-compiler-driver-hardware-and-environment-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to packages, images, framework, compiler, driver, hardware and environment, and which evidence supports them?", "kind": "state", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "packages-images-framework-compiler-driver-hardware-and-environment-data", "name": "Packages, images, framework, compiler, driver, hardware and environment data", "description": "Typed data for packages, images, framework, compiler, driver, hardware and environment with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-017" ] } ], "artifacts": [ { "id": "packages-images-framework-compiler-driver-hardware-and-environment-record", "name": "Packages, images, framework, compiler, driver, hardware and environment record", "description": "Immutable or successor-versioned training evidence for packages, images, framework, compiler, driver, hardware and environment.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for packages-images-framework-compiler-driver-hardware-and-environment; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-017" ] } ], "inline_only_rationale": null } ] } ] }, { "id": "orchestration-distributed-execution-resources-and-progress", "name": "Orchestration, distributed execution, resources and progress", "description": "Groups governed training-run context for orchestration, distributed execution, resources and progress.", "rationale": "Desired topology, observed allocation and execution events remain distinct.", "source_refs": [ "SRC-002", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-013", "SRC-014", "SRC-015" ], "layers": [ { "id": "topology-workers-stages-attempts-and-state", "name": "Topology, workers, stages, attempts and state", "description": "Groups source-qualified training-run context for topology, workers, stages, attempts and state.", "source_refs": [ "SRC-007", "SRC-009", "SRC-010", "SRC-013" ], "findings": [ { "id": "cluster-topology-workers-ranks-parallelism-and-communication-strategy", "name": "Cluster topology, workers, ranks, parallelism and communication strategy", "description": "Records cluster topology, workers, ranks, parallelism and communication strategy as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-007", "SRC-009", "SRC-010", "SRC-013" ], "questions": [ { "id": "cluster-topology-workers-ranks-parallelism-and-communication-strategy-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish cluster topology, workers, ranks, parallelism and communication strategy?", "kind": "composition", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "cluster-topology-workers-ranks-parallelism-and-communication-strategy-q02", "text": "Who may declare, execute, observe, review, correct or rely on cluster topology, workers, ranks, parallelism and communication strategy, under which authority and limits?", "kind": "privacy", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "cluster-topology-workers-ranks-parallelism-and-communication-strategy-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to cluster topology, workers, ranks, parallelism and communication strategy, and which evidence supports them?", "kind": "identity", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "cluster-topology-workers-ranks-parallelism-and-communication-strategy-data", "name": "Cluster topology, workers, ranks, parallelism and communication strategy data", "description": "Typed data for cluster topology, workers, ranks, parallelism and communication strategy with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "0..n", "required": false, "source_refs": [ "SRC-007", "SRC-009", "SRC-010", "SRC-013" ] } ], "artifacts": [ { "id": "cluster-topology-workers-ranks-parallelism-and-communication-strategy-record", "name": "Cluster topology, workers, ranks, parallelism and communication strategy record", "description": "Immutable or successor-versioned training evidence for cluster topology, workers, ranks, parallelism and communication strategy.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for cluster-topology-workers-ranks-parallelism-and-communication-strategy; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-007", "SRC-009", "SRC-010", "SRC-013" ] } ], "inline_only_rationale": null }, { "id": "stage-task-attempt-state-transition-retry-resume-and-idempotency", "name": "Stage, task, attempt, state transition, retry, resume and idempotency", "description": "Records stage, task, attempt, state transition, retry, resume and idempotency as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-007", "SRC-009", "SRC-010", "SRC-013" ], "questions": [ { "id": "stage-task-attempt-state-transition-retry-resume-and-idempotency-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish stage, task, attempt, state transition, retry, resume and idempotency?", "kind": "state", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "stage-task-attempt-state-transition-retry-resume-and-idempotency-q02", "text": "Who may declare, execute, observe, review, correct or rely on stage, task, attempt, state transition, retry, resume and idempotency, under which authority and limits?", "kind": "lifecycle", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "stage-task-attempt-state-transition-retry-resume-and-idempotency-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to stage, task, attempt, state transition, retry, resume and idempotency, and which evidence supports them?", "kind": "classification", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "stage-task-attempt-state-transition-retry-resume-and-idempotency-data", "name": "Stage, task, attempt, state transition, retry, resume and idempotency data", "description": "Typed data for stage, task, attempt, state transition, retry, resume and idempotency with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-007", "SRC-009", "SRC-010", "SRC-013" ] } ], "artifacts": [ { "id": "stage-task-attempt-state-transition-retry-resume-and-idempotency-record", "name": "Stage, task, attempt, state transition, retry, resume and idempotency record", "description": "Immutable or successor-versioned training evidence for stage, task, attempt, state transition, retry, resume and idempotency.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for stage-task-attempt-state-transition-retry-resume-and-idempotency; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-007", "SRC-009", "SRC-010", "SRC-013" ] } ], "inline_only_rationale": null } ] }, { "id": "progress-optimizer-resources-cost-and-environment", "name": "Progress, optimizer, resources, cost and environment", "description": "Groups source-qualified training-run context for progress, optimizer, resources, cost and environment.", "source_refs": [ "SRC-002", "SRC-008", "SRC-010", "SRC-013", "SRC-014", "SRC-015" ], "findings": [ { "id": "steps-epochs-batches-samples-tokens-optimizer-scheduler-and-precision", "name": "Steps, epochs, batches, samples, tokens, optimizer, scheduler and precision", "description": "Records steps, epochs, batches, samples, tokens, optimizer, scheduler and precision as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-002", "SRC-008", "SRC-010", "SRC-013", "SRC-014", "SRC-015" ], "questions": [ { "id": "steps-epochs-batches-samples-tokens-optimizer-scheduler-and-precision-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish steps, epochs, batches, samples, tokens, optimizer, scheduler and precision?", "kind": "measurement", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "steps-epochs-batches-samples-tokens-optimizer-scheduler-and-precision-q02", "text": "Who may declare, execute, observe, review, correct or rely on steps, epochs, batches, samples, tokens, optimizer, scheduler and precision, under which authority and limits?", "kind": "quality", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "steps-epochs-batches-samples-tokens-optimizer-scheduler-and-precision-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to steps, epochs, batches, samples, tokens, optimizer, scheduler and precision, and which evidence supports them?", "kind": "relationship", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "steps-epochs-batches-samples-tokens-optimizer-scheduler-and-precision-data", "name": "Steps, epochs, batches, samples, tokens, optimizer, scheduler and precision data", "description": "Typed data for steps, epochs, batches, samples, tokens, optimizer, scheduler and precision with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-002", "SRC-008", "SRC-010", "SRC-013", "SRC-014", "SRC-015" ] } ], "artifacts": [ { "id": "steps-epochs-batches-samples-tokens-optimizer-scheduler-and-precision-record", "name": "Steps, epochs, batches, samples, tokens, optimizer, scheduler and precision record", "description": "Immutable or successor-versioned training evidence for steps, epochs, batches, samples, tokens, optimizer, scheduler and precision.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for steps-epochs-batches-samples-tokens-optimizer-scheduler-and-precision; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-002", "SRC-008", "SRC-010", "SRC-013", "SRC-014", "SRC-015" ] } ], "inline_only_rationale": null }, { "id": "requested-allocated-observed-compute-storage-network-cost-energy-and-emissions", "name": "Requested, allocated and observed compute, storage, network, cost, energy and emissions", "description": "Records requested, allocated and observed compute, storage, network, cost, energy and emissions as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-002", "SRC-008", "SRC-010", "SRC-013", "SRC-014", "SRC-015" ], "questions": [ { "id": "requested-allocated-observed-compute-storage-network-cost-energy-and-emissions-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish requested, allocated and observed compute, storage, network, cost, energy and emissions?", "kind": "measurement", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "requested-allocated-observed-compute-storage-network-cost-energy-and-emissions-q02", "text": "Who may declare, execute, observe, review, correct or rely on requested, allocated and observed compute, storage, network, cost, energy and emissions, under which authority and limits?", "kind": "security", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "requested-allocated-observed-compute-storage-network-cost-energy-and-emissions-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to requested, allocated and observed compute, storage, network, cost, energy and emissions, and which evidence supports them?", "kind": "authority", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "requested-allocated-observed-compute-storage-network-cost-energy-and-emissions-data", "name": "Requested, allocated and observed compute, storage, network, cost, energy and emissions data", "description": "Typed data for requested, allocated and observed compute, storage, network, cost, energy and emissions with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "0..n", "required": false, "source_refs": [ "SRC-002", "SRC-008", "SRC-010", "SRC-013", "SRC-014", "SRC-015" ] } ], "artifacts": [ { "id": "requested-allocated-observed-compute-storage-network-cost-energy-and-emissions-record", "name": "Requested, allocated and observed compute, storage, network, cost, energy and emissions record", "description": "Immutable or successor-versioned training evidence for requested, allocated and observed compute, storage, network, cost, energy and emissions.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for requested-allocated-observed-compute-storage-network-cost-energy-and-emissions; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-002", "SRC-008", "SRC-010", "SRC-013", "SRC-014", "SRC-015" ] } ], "inline_only_rationale": null } ] } ] }, { "id": "checkpoints-metrics-validation-quality-and-safety", "name": "Checkpoints, metrics, validation, quality and safety", "description": "Groups governed training-run context for checkpoints, metrics, validation, quality and safety.", "rationale": "Progress evidence and evaluation evidence must not be collapsed into a release decision.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-014", "SRC-017" ], "layers": [ { "id": "checkpoints-progress-metrics-and-selection", "name": "Checkpoints, progress metrics and selection", "description": "Groups source-qualified training-run context for checkpoints, progress metrics and selection.", "source_refs": [ "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-017" ], "findings": [ { "id": "checkpoint-identity-step-digest-completeness-reason-retention-and-resume", "name": "Checkpoint identity, step, digest, completeness, reason, retention and resume", "description": "Records checkpoint identity, step, digest, completeness, reason, retention and resume as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-017" ], "questions": [ { "id": "checkpoint-identity-step-digest-completeness-reason-retention-and-resume-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish checkpoint identity, step, digest, completeness, reason, retention and resume?", "kind": "evidence", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "checkpoint-identity-step-digest-completeness-reason-retention-and-resume-q02", "text": "Who may declare, execute, observe, review, correct or rely on checkpoint identity, step, digest, completeness, reason, retention and resume, under which authority and limits?", "kind": "retention", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "checkpoint-identity-step-digest-completeness-reason-retention-and-resume-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to checkpoint identity, step, digest, completeness, reason, retention and resume, and which evidence supports them?", "kind": "requirement", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "checkpoint-identity-step-digest-completeness-reason-retention-and-resume-data", "name": "Checkpoint identity, step, digest, completeness, reason, retention and resume data", "description": "Typed data for checkpoint identity, step, digest, completeness, reason, retention and resume with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-017" ] } ], "artifacts": [ { "id": "checkpoint-identity-step-digest-completeness-reason-retention-and-resume-record", "name": "Checkpoint identity, step, digest, completeness, reason, retention and resume record", "description": "Immutable or successor-versioned training evidence for checkpoint identity, step, digest, completeness, reason, retention and resume.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for checkpoint-identity-step-digest-completeness-reason-retention-and-resume; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-017" ] } ], "inline_only_rationale": null }, { "id": "loss-metric-series-effective-parameters-selection-rule-and-observed-best", "name": "Loss and metric series, effective parameters, selection rule and observed best", "description": "Records loss and metric series, effective parameters, selection rule and observed best as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-017" ], "questions": [ { "id": "loss-metric-series-effective-parameters-selection-rule-and-observed-best-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish loss and metric series, effective parameters, selection rule and observed best?", "kind": "quality", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "loss-metric-series-effective-parameters-selection-rule-and-observed-best-q02", "text": "Who may declare, execute, observe, review, correct or rely on loss and metric series, effective parameters, selection rule and observed best, under which authority and limits?", "kind": "interoperability", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "loss-metric-series-effective-parameters-selection-rule-and-observed-best-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to loss and metric series, effective parameters, selection rule and observed best, and which evidence supports them?", "kind": "constraint", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "loss-metric-series-effective-parameters-selection-rule-and-observed-best-data", "name": "Loss and metric series, effective parameters, selection rule and observed best data", "description": "Typed data for loss and metric series, effective parameters, selection rule and observed best with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-017" ] } ], "artifacts": [ { "id": "loss-metric-series-effective-parameters-selection-rule-and-observed-best-record", "name": "Loss and metric series, effective parameters, selection rule and observed best record", "description": "Immutable or successor-versioned training evidence for loss and metric series, effective parameters, selection rule and observed best.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for loss-metric-series-effective-parameters-selection-rule-and-observed-best; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-017" ] } ], "inline_only_rationale": null } ] }, { "id": "validation-data-quality-privacy-and-safety-evidence", "name": "Validation, data quality, privacy and safety evidence", "description": "Groups source-qualified training-run context for validation, data quality, privacy and safety evidence.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-014" ], "findings": [ { "id": "validation-split-evaluation-reference-leakage-contamination-and-generalization", "name": "Validation split, evaluation reference, leakage, contamination and generalization", "description": "Records validation split, evaluation reference, leakage, contamination and generalization as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-014" ], "questions": [ { "id": "validation-split-evaluation-reference-leakage-contamination-and-generalization-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish validation split, evaluation reference, leakage, contamination and generalization?", "kind": "validation", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "validation-split-evaluation-reference-leakage-contamination-and-generalization-q02", "text": "Who may declare, execute, observe, review, correct or rely on validation split, evaluation reference, leakage, contamination and generalization, under which authority and limits?", "kind": "decision", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "validation-split-evaluation-reference-leakage-contamination-and-generalization-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to validation split, evaluation reference, leakage, contamination and generalization, and which evidence supports them?", "kind": "event", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "validation-split-evaluation-reference-leakage-contamination-and-generalization-data", "name": "Validation split, evaluation reference, leakage, contamination and generalization data", "description": "Typed data for validation split, evaluation reference, leakage, contamination and generalization with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-014" ] } ], "artifacts": [ { "id": "validation-split-evaluation-reference-leakage-contamination-and-generalization-record", "name": "Validation split, evaluation reference, leakage, contamination and generalization record", "description": "Immutable or successor-versioned training evidence for validation split, evaluation reference, leakage, contamination and generalization.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for validation-split-evaluation-reference-leakage-contamination-and-generalization; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-014" ] } ], "inline_only_rationale": null }, { "id": "data-quality-bias-privacy-security-safety-red-team-and-incident-references", "name": "Data quality, bias, privacy, security, safety, red-team and incident references", "description": "Records data quality, bias, privacy, security, safety, red-team and incident references as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-014" ], "questions": [ { "id": "data-quality-bias-privacy-security-safety-red-team-and-incident-references-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish data quality, bias, privacy, security, safety, red-team and incident references?", "kind": "security", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "data-quality-bias-privacy-security-safety-red-team-and-incident-references-q02", "text": "Who may declare, execute, observe, review, correct or rely on data quality, bias, privacy, security, safety, red-team and incident references, under which authority and limits?", "kind": "state", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "data-quality-bias-privacy-security-safety-red-team-and-incident-references-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to data quality, bias, privacy, security, safety, red-team and incident references, and which evidence supports them?", "kind": "temporal", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "data-quality-bias-privacy-security-safety-red-team-and-incident-references-data", "name": "Data quality, bias, privacy, security, safety, red-team and incident references data", "description": "Typed data for data quality, bias, privacy, security, safety, red-team and incident references with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-014" ] } ], "artifacts": [ { "id": "data-quality-bias-privacy-security-safety-red-team-and-incident-references-record", "name": "Data quality, bias, privacy, security, safety, red-team and incident references record", "description": "Immutable or successor-versioned training evidence for data quality, bias, privacy, security, safety, red-team and incident references.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for data-quality-bias-privacy-security-safety-red-team-and-incident-references; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-014" ] } ], "inline_only_rationale": null } ] } ] }, { "id": "outputs-lineage-reproducibility-and-outcome", "name": "Outputs, lineage, reproducibility and outcome", "description": "Groups governed training-run context for outputs, lineage, reproducibility and outcome.", "rationale": "Produced candidates are external artifacts linked by derivation, not embedded model masters.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-014", "SRC-017" ], "layers": [ { "id": "candidate-final-artifacts-and-derivation", "name": "Candidate and final artifacts and derivation", "description": "Groups source-qualified training-run context for candidate and final artifacts and derivation.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012" ], "findings": [ { "id": "candidate-output-final-selection-packaging-format-digest-and-signature", "name": "Candidate output, final selection, packaging, format, digest and signature", "description": "Records candidate output, final selection, packaging, format, digest and signature as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012" ], "questions": [ { "id": "candidate-output-final-selection-packaging-format-digest-and-signature-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish candidate output, final selection, packaging, format, digest and signature?", "kind": "evidence", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "candidate-output-final-selection-packaging-format-digest-and-signature-q02", "text": "Who may declare, execute, observe, review, correct or rely on candidate output, final selection, packaging, format, digest and signature, under which authority and limits?", "kind": "identity", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "candidate-output-final-selection-packaging-format-digest-and-signature-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to candidate output, final selection, packaging, format, digest and signature, and which evidence supports them?", "kind": "composition", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "candidate-output-final-selection-packaging-format-digest-and-signature-data", "name": "Candidate output, final selection, packaging, format, digest and signature data", "description": "Typed data for candidate output, final selection, packaging, format, digest and signature with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012" ] } ], "artifacts": [ { "id": "candidate-output-final-selection-packaging-format-digest-and-signature-record", "name": "Candidate output, final selection, packaging, format, digest and signature record", "description": "Immutable or successor-versioned training evidence for candidate output, final selection, packaging, format, digest and signature.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for candidate-output-final-selection-packaging-format-digest-and-signature; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012" ] } ], "inline_only_rationale": null }, { "id": "base-data-code-configuration-checkpoint-builder-and-artifact-derivation", "name": "Base, data, code, configuration, checkpoint, builder and artifact derivation", "description": "Records base, data, code, configuration, checkpoint, builder and artifact derivation as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012" ], "questions": [ { "id": "base-data-code-configuration-checkpoint-builder-and-artifact-derivation-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish base, data, code, configuration, checkpoint, builder and artifact derivation?", "kind": "provenance", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "base-data-code-configuration-checkpoint-builder-and-artifact-derivation-q02", "text": "Who may declare, execute, observe, review, correct or rely on base, data, code, configuration, checkpoint, builder and artifact derivation, under which authority and limits?", "kind": "classification", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "base-data-code-configuration-checkpoint-builder-and-artifact-derivation-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to base, data, code, configuration, checkpoint, builder and artifact derivation, and which evidence supports them?", "kind": "evidence", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "base-data-code-configuration-checkpoint-builder-and-artifact-derivation-data", "name": "Base, data, code, configuration, checkpoint, builder and artifact derivation data", "description": "Typed data for base, data, code, configuration, checkpoint, builder and artifact derivation with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012" ] } ], "artifacts": [ { "id": "base-data-code-configuration-checkpoint-builder-and-artifact-derivation-record", "name": "Base, data, code, configuration, checkpoint, builder and artifact derivation record", "description": "Immutable or successor-versioned training evidence for base, data, code, configuration, checkpoint, builder and artifact derivation.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for base-data-code-configuration-checkpoint-builder-and-artifact-derivation; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012" ] } ], "inline_only_rationale": null } ] }, { "id": "reproducibility-nondeterminism-comparison-and-acceptance", "name": "Reproducibility, nondeterminism, comparison and acceptance", "description": "Groups source-qualified training-run context for reproducibility, nondeterminism, comparison and acceptance.", "source_refs": [ "SRC-001", "SRC-002", "SRC-004", "SRC-014", "SRC-017" ], "findings": [ { "id": "replay-recipe-randomness-determinism-nondeterminism-and-environment-equivalence", "name": "Replay recipe, randomness, determinism, nondeterminism and environment equivalence", "description": "Records replay recipe, randomness, determinism, nondeterminism and environment equivalence as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-002", "SRC-004", "SRC-014", "SRC-017" ], "questions": [ { "id": "replay-recipe-randomness-determinism-nondeterminism-and-environment-equivalence-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish replay recipe, randomness, determinism, nondeterminism and environment equivalence?", "kind": "validation", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "replay-recipe-randomness-determinism-nondeterminism-and-environment-equivalence-q02", "text": "Who may declare, execute, observe, review, correct or rely on replay recipe, randomness, determinism, nondeterminism and environment equivalence, under which authority and limits?", "kind": "relationship", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "replay-recipe-randomness-determinism-nondeterminism-and-environment-equivalence-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to replay recipe, randomness, determinism, nondeterminism and environment equivalence, and which evidence supports them?", "kind": "ownership", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "replay-recipe-randomness-determinism-nondeterminism-and-environment-equivalence-data", "name": "Replay recipe, randomness, determinism, nondeterminism and environment equivalence data", "description": "Typed data for replay recipe, randomness, determinism, nondeterminism and environment equivalence with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-001", "SRC-002", "SRC-004", "SRC-014", "SRC-017" ] } ], "artifacts": [ { "id": "replay-recipe-randomness-determinism-nondeterminism-and-environment-equivalence-record", "name": "Replay recipe, randomness, determinism, nondeterminism and environment equivalence record", "description": "Immutable or successor-versioned training evidence for replay recipe, randomness, determinism, nondeterminism and environment equivalence.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for replay-recipe-randomness-determinism-nondeterminism-and-environment-equivalence; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-002", "SRC-004", "SRC-014", "SRC-017" ] } ], "inline_only_rationale": null }, { "id": "baseline-comparison-outcome-quality-safety-compliance-publication-and-deployment-decision", "name": "Baseline comparison, outcome, quality, safety, compliance, publication and deployment decision", "description": "Records baseline comparison, outcome, quality, safety, compliance, publication and deployment decision as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-002", "SRC-004", "SRC-014", "SRC-017" ], "questions": [ { "id": "baseline-comparison-outcome-quality-safety-compliance-publication-and-deployment-decision-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish baseline comparison, outcome, quality, safety, compliance, publication and deployment decision?", "kind": "decision", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "baseline-comparison-outcome-quality-safety-compliance-publication-and-deployment-decision-q02", "text": "Who may declare, execute, observe, review, correct or rely on baseline comparison, outcome, quality, safety, compliance, publication and deployment decision, under which authority and limits?", "kind": "authority", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "baseline-comparison-outcome-quality-safety-compliance-publication-and-deployment-decision-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to baseline comparison, outcome, quality, safety, compliance, publication and deployment decision, and which evidence supports them?", "kind": "measurement", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "baseline-comparison-outcome-quality-safety-compliance-publication-and-deployment-decision-data", "name": "Baseline comparison, outcome, quality, safety, compliance, publication and deployment decision data", "description": "Typed data for baseline comparison, outcome, quality, safety, compliance, publication and deployment decision with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-001", "SRC-002", "SRC-004", "SRC-014", "SRC-017" ] } ], "artifacts": [ { "id": "baseline-comparison-outcome-quality-safety-compliance-publication-and-deployment-decision-record", "name": "Baseline comparison, outcome, quality, safety, compliance, publication and deployment decision record", "description": "Immutable or successor-versioned training evidence for baseline comparison, outcome, quality, safety, compliance, publication and deployment decision.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for baseline-comparison-outcome-quality-safety-compliance-publication-and-deployment-decision; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-002", "SRC-004", "SRC-014", "SRC-017" ] } ], "inline_only_rationale": null } ] } ] }, { "id": "failure-governance-access-retention-correction-and-projections", "name": "Failure, governance, access, retention, correction and projections", "description": "Groups governed training-run context for failure, governance, access, retention, correction and projections.", "rationale": "Failure and governance evidence must survive cleanup and remain successor-correctable.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-016" ], "layers": [ { "id": "failure-cancel-recovery-cleanup-and-records", "name": "Failure, cancellation, recovery, cleanup and records", "description": "Groups source-qualified training-run context for failure, cancellation, recovery, cleanup and records.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-007", "SRC-010" ], "findings": [ { "id": "warning-error-failure-early-stop-cancellation-root-cause-and-impact", "name": "Warning, error, failure, early stop, cancellation, root cause and impact", "description": "Records warning, error, failure, early stop, cancellation, root cause and impact as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-007", "SRC-010" ], "questions": [ { "id": "warning-error-failure-early-stop-cancellation-root-cause-and-impact-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish warning, error, failure, early stop, cancellation, root cause and impact?", "kind": "exception", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "warning-error-failure-early-stop-cancellation-root-cause-and-impact-q02", "text": "Who may declare, execute, observe, review, correct or rely on warning, error, failure, early stop, cancellation, root cause and impact, under which authority and limits?", "kind": "requirement", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "warning-error-failure-early-stop-cancellation-root-cause-and-impact-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to warning, error, failure, early stop, cancellation, root cause and impact, and which evidence supports them?", "kind": "exception", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "warning-error-failure-early-stop-cancellation-root-cause-and-impact-data", "name": "Warning, error, failure, early stop, cancellation, root cause and impact data", "description": "Typed data for warning, error, failure, early stop, cancellation, root cause and impact with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-007", "SRC-010" ] } ], "artifacts": [ { "id": "warning-error-failure-early-stop-cancellation-root-cause-and-impact-record", "name": "Warning, error, failure, early stop, cancellation, root cause and impact record", "description": "Immutable or successor-versioned training evidence for warning, error, failure, early stop, cancellation, root cause and impact.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for warning-error-failure-early-stop-cancellation-root-cause-and-impact; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-007", "SRC-010" ] } ], "inline_only_rationale": null }, { "id": "recovery-rollback-cleanup-retention-legal-hold-disposition-and-proof", "name": "Recovery, rollback, cleanup, retention, legal hold, disposition and proof", "description": "Records recovery, rollback, cleanup, retention, legal hold, disposition and proof as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-007", "SRC-010" ], "questions": [ { "id": "recovery-rollback-cleanup-retention-legal-hold-disposition-and-proof-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish recovery, rollback, cleanup, retention, legal hold, disposition and proof?", "kind": "retention", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "recovery-rollback-cleanup-retention-legal-hold-disposition-and-proof-q02", "text": "Who may declare, execute, observe, review, correct or rely on recovery, rollback, cleanup, retention, legal hold, disposition and proof, under which authority and limits?", "kind": "constraint", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "recovery-rollback-cleanup-retention-legal-hold-disposition-and-proof-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to recovery, rollback, cleanup, retention, legal hold, disposition and proof, and which evidence supports them?", "kind": "provenance", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "recovery-rollback-cleanup-retention-legal-hold-disposition-and-proof-data", "name": "Recovery, rollback, cleanup, retention, legal hold, disposition and proof data", "description": "Typed data for recovery, rollback, cleanup, retention, legal hold, disposition and proof with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-007", "SRC-010" ] } ], "artifacts": [ { "id": "recovery-rollback-cleanup-retention-legal-hold-disposition-and-proof-record", "name": "Recovery, rollback, cleanup, retention, legal hold, disposition and proof record", "description": "Immutable or successor-versioned training evidence for recovery, rollback, cleanup, retention, legal hold, disposition and proof.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for recovery-rollback-cleanup-retention-legal-hold-disposition-and-proof; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-007", "SRC-010" ] } ], "inline_only_rationale": null } ] }, { "id": "access-correction-audit-and-interoperability", "name": "Access, correction, audit and interoperability", "description": "Groups source-qualified training-run context for access, correction, audit and interoperability.", "source_refs": [ "SRC-001", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-016" ], "findings": [ { "id": "identity-secret-data-rights-role-purpose-access-audit-correction-and-current-head", "name": "Identity, secret, data rights, role, purpose, access, audit, correction and current head", "description": "Records identity, secret, data rights, role, purpose, access, audit, correction and current head as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-016" ], "questions": [ { "id": "identity-secret-data-rights-role-purpose-access-audit-correction-and-current-head-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish identity, secret, data rights, role, purpose, access, audit, correction and current head?", "kind": "access", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "identity-secret-data-rights-role-purpose-access-audit-correction-and-current-head-q02", "text": "Who may declare, execute, observe, review, correct or rely on identity, secret, data rights, role, purpose, access, audit, correction and current head, under which authority and limits?", "kind": "event", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "identity-secret-data-rights-role-purpose-access-audit-correction-and-current-head-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to identity, secret, data rights, role, purpose, access, audit, correction and current head, and which evidence supports them?", "kind": "process", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "identity-secret-data-rights-role-purpose-access-audit-correction-and-current-head-data", "name": "Identity, secret, data rights, role, purpose, access, audit, correction and current head data", "description": "Typed data for identity, secret, data rights, role, purpose, access, audit, correction and current head with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "1", "required": true, "source_refs": [ "SRC-001", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-016" ] } ], "artifacts": [ { "id": "identity-secret-data-rights-role-purpose-access-audit-correction-and-current-head-record", "name": "Identity, secret, data rights, role, purpose, access, audit, correction and current head record", "description": "Immutable or successor-versioned training evidence for identity, secret, data rights, role, purpose, access, audit, correction and current head.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for identity-secret-data-rights-role-purpose-access-audit-correction-and-current-head; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-016" ] } ], "inline_only_rationale": null }, { "id": "mlflow-mlmd-openlineage-prov-kubeflow-slsa-oci-otel-mlperf-sci-projections", "name": "MLflow, MLMD, OpenLineage, PROV, Kubeflow, SLSA, OCI, OpenTelemetry, MLPerf and SCI projections", "description": "Records mlflow, mlmd, openlineage, prov, kubeflow, slsa, oci, opentelemetry, mlperf and sci projections as source-qualified training-run context while dataset, source code, base model, trained model artifact, evaluation, registry, deployment, secrets, policy and infrastructure masters remain external.", "source_refs": [ "SRC-001", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-016" ], "questions": [ { "id": "mlflow-mlmd-openlineage-prov-kubeflow-slsa-oci-otel-mlperf-sci-projections-q01", "text": "What stable identity, version-qualified values, scope and explicit unknowns establish mlflow, mlmd, openlineage, prov, kubeflow, slsa, oci, opentelemetry, mlperf and sci projections?", "kind": "interoperability", "answer_data": [ "identifiers and run scope", "version-qualified values and units", "unknown and not-applicable states" ] }, { "id": "mlflow-mlmd-openlineage-prov-kubeflow-slsa-oci-otel-mlperf-sci-projections-q02", "text": "Who may declare, execute, observe, review, correct or rely on mlflow, mlmd, openlineage, prov, kubeflow, slsa, oci, opentelemetry, mlperf and sci projections, under which authority and limits?", "kind": "temporal", "answer_data": [ "human, agent, service and owner roles", "authority, policy, purpose and limits", "review, exception and escalation path" ] }, { "id": "mlflow-mlmd-openlineage-prov-kubeflow-slsa-oci-otel-mlperf-sci-projections-q03", "text": "Which planned, event, effective, recorded, ingested and knowledge times apply to mlflow, mlmd, openlineage, prov, kubeflow, slsa, oci, opentelemetry, mlperf and sci projections, and which evidence supports them?", "kind": "validation", "answer_data": [ "distinct run and knowledge times", "evidence, provenance and uncertainty", "successor correction and retention" ] } ], "data_elements": [ { "id": "mlflow-mlmd-openlineage-prov-kubeflow-slsa-oci-otel-mlperf-sci-projections-data", "name": "MLflow, MLMD, OpenLineage, PROV, Kubeflow, SLSA, OCI, OpenTelemetry, MLPerf and SCI projections data", "description": "Typed data for mlflow, mlmd, openlineage, prov, kubeflow, slsa, oci, opentelemetry, mlperf and sci projections with run scope, source, authority, state, unit, time, evidence and provenance.", "value_kind": "collection", "cardinality": "0..n", "required": false, "source_refs": [ "SRC-001", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-016" ] } ], "artifacts": [ { "id": "mlflow-mlmd-openlineage-prov-kubeflow-slsa-oci-otel-mlperf-sci-projections-record", "name": "MLflow, MLMD, OpenLineage, PROV, Kubeflow, SLSA, OCI, OpenTelemetry, MLPerf and SCI projections record", "description": "Immutable or successor-versioned training evidence for mlflow, mlmd, openlineage, prov, kubeflow, slsa, oci, opentelemetry, mlperf and sci projections.", "media_or_form": [ "logical model-training assertion", "configuration, event, metric, checkpoint, lineage, governance or projection record" ], "serial": true, "identity_strategy": "Run ID plus independent assertion, event or artifact ID for mlflow-mlmd-openlineage-prov-kubeflow-slsa-oci-otel-mlperf-sci-projections; model name, dataset name, timestamp, metric value, checkpoint step and file path never identify a record alone.", "source_refs": [ "SRC-001", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-016" ] } ], "inline_only_rationale": null } ] } ] } ] }, "functions": [ { "id": "register-training-run", "name": "Register a model training or fine-tuning run", "description": "Governed operation to register a model training or fine-tuning run without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "run identity", "experiment and job definition", "owner and authority" ], "outputs": [ "stable run aggregate" ], "preconditions": [ "namespace, objective, method, inputs and accountability are resolvable" ], "effects": [ "a new immutable-head run is registered without copying external masters" ], "source_refs": [ "SRC-001", "SRC-006", "SRC-007", "SRC-008", "SRC-009" ] }, { "id": "resolve-immutable-inputs", "name": "Resolve immutable run inputs", "description": "Governed operation to resolve immutable run inputs without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "run", "base model", "datasets", "code", "configuration" ], "outputs": [ "validated version-qualified bindings" ], "preconditions": [ "identities, versions, digests, rights and availability pass" ], "effects": [ "input bindings are recorded without mutating source masters" ], "source_refs": [ "SRC-002", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-011" ] }, { "id": "declare-method-objective-and-authority", "name": "Declare method, objective and authority", "description": "Governed operation to declare method, objective and authority without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "run", "training method", "objective", "risk and approval profile" ], "outputs": [ "governed execution plan" ], "preconditions": [ "acceptance, stop, privacy, security, cost and resource limits pass" ], "effects": [ "the planned method and accountable approvals become auditable" ], "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004" ] }, { "id": "dispatch-and-bind-execution", "name": "Record dispatch and bind execution", "description": "Governed operation to record dispatch and bind execution without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "approved plan", "runtime", "topology", "resource request" ], "outputs": [ "dispatch and execution binding" ], "preconditions": [ "delegated execution authority and environment constraints pass" ], "effects": [ "dispatch is recorded; this specification grants no autonomous compute allocation" ], "source_refs": [ "SRC-007", "SRC-009", "SRC-010", "SRC-013" ] }, { "id": "record-progress-resources-and-telemetry", "name": "Record progress, resources and telemetry", "description": "Governed operation to record progress, resources and telemetry without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "run", "stage and attempt events", "measurements" ], "outputs": [ "ordered progress and resource history" ], "preconditions": [ "units, clocks, sampling, source and aggregation methods pass" ], "effects": [ "observations are appended without treating planned values as observed" ], "source_refs": [ "SRC-008", "SRC-009", "SRC-010", "SRC-013", "SRC-015", "SRC-016" ] }, { "id": "checkpoint-resume-and-retry", "name": "Checkpoint, resume and retry", "description": "Governed operation to checkpoint, resume and retry without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "run", "checkpoint candidate", "retry or resume request" ], "outputs": [ "validated checkpoint binding and successor attempt" ], "preconditions": [ "completeness, digest, compatibility, authority and idempotency pass" ], "effects": [ "checkpoint and attempt histories remain independently reconstructable" ], "source_refs": [ "SRC-006", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-017" ] }, { "id": "record-metrics-validation-and-safety-evidence", "name": "Record metrics, validation and safety evidence", "description": "Governed operation to record metrics, validation and safety evidence without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "run", "metric series", "evaluation and review references" ], "outputs": [ "source-qualified evidence bindings" ], "preconditions": [ "split role, method, units, comparator, leakage and review status pass" ], "effects": [ "evidence is recorded without making a release or deployment decision" ], "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-014" ] }, { "id": "finalize-cancel-fail-or-early-stop", "name": "Finalize, cancel, fail or early-stop a run", "description": "Governed operation to finalize, cancel, fail or early-stop a run without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "run", "terminal event", "reason and authority" ], "outputs": [ "terminal successor state" ], "preconditions": [ "state transition, outcome, evidence, cleanup and retention duties pass" ], "effects": [ "the run closes without deleting external inputs or outputs" ], "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-007", "SRC-010" ] }, { "id": "bind-and-select-produced-artifact", "name": "Bind and select a produced model artifact", "description": "Governed operation to bind and select a produced model artifact without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "terminal or candidate run", "artifact references", "selection evidence" ], "outputs": [ "typed produced and selected artifact bindings" ], "preconditions": [ "digest, provenance, packaging, evaluation and accountable selection pass" ], "effects": [ "external model artifacts are linked by derivation; registry and deployment remain separate" ], "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012", "SRC-014" ] }, { "id": "correct-project-retain-disclose-and-audit", "name": "Correct, project, retain, disclose and audit", "description": "Governed operation to correct, project, retain, disclose and audit without autonomous training, resource acquisition, secret retrieval, privacy decision, release, deployment or destructive cleanup.", "inputs": [ "run", "target profile", "access and records policy" ], "outputs": [ "successor, projection, disclosure, tombstone or disposition event" ], "preconditions": [ "mapping versions, loss, privacy, hold, authority and idempotency pass" ], "effects": [ "run context stays reconstructable and explicit about current head and projection loss" ], "source_refs": [ "SRC-001", "SRC-003", "SRC-004", "SRC-005", "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-016" ] } ], "composition": [ { "target": "WM-DAT-001 Dataset", "relation": "REFERENCE", "purpose": "Bind candidate training, validation, evaluation and auxiliary dataset snapshots without owning their content, rights or lifecycle.", "required": true, "source_refs": [ "SRC-002", "SRC-004", "SRC-005", "SRC-006", "SRC-008", "SRC-009" ] }, { "target": "WM-SFT-004 produced model artifact", "relation": "REFERENCE", "purpose": "Represent the candidate PRODUCES ledger edge by a non-owning output reference and derivation record, without granting registry, release, deployment or cascade authority.", "required": false, "source_refs": [ "SRC-003", "SRC-006", "SRC-008", "SRC-009", "SRC-011", "SRC-012" ] }, { "target": "Source code, base model, tokenizer, evaluation, registry, deployment, infrastructure, secret, policy, provenance, audit and records models", "relation": "REFERENCE", "purpose": "Resolve authoritative inputs, controls, evidence and lifecycle records without absorbing their ownership.", "required": false, "source_refs": [ "SRC-001", "SRC-002", "SRC-003", "SRC-004", "SRC-006" ] }, { "target": "MLflow, MLMD, OpenLineage, PROV, Kubeflow, SLSA, OCI, OpenTelemetry, MLPerf, SCI and PyTorch", "relation": "ALIGN", "purpose": "Project version-pinned execution, lineage, packaging, telemetry, benchmark, environmental and reproducibility views with information-loss declarations.", "required": false, "source_refs": [ "SRC-006", "SRC-007", "SRC-008", "SRC-009", "SRC-010", "SRC-011", "SRC-012", "SRC-013", "SRC-014", "SRC-015", "SRC-017" ] } ], "serviceLayers": { "dimension": { "owner_package_requirements": [ "Dimension owner, accountable AI owner and training mandate", "Authoritative dataset, source code, base model, trained model, evaluation, registry, deployment, infrastructure, secret, policy, provenance, audit and record registries", "Approved task, training method, jurisdiction, data-rights, privacy, security, safety, quality, cost, energy, retention and interoperability profiles", "Role, delegation, approval, incident, release, deployment and agent-operation policies" ], "namespace_guidance": "Mint run, stage, task, attempt, event, metric-series, checkpoint, candidate, correction, disclosure and projection IDs; preserve external model, dataset, code, artifact and infrastructure identifiers.", "registry_links": [ "https://ver.cy/models/", "https://ver.cy/model-agent-protocol.md" ] }, "canon_and_patch": { "canonicalization_rules": [ "Canonicalize one run by authoritative execution-system identifier and owning namespace; never by model name, dataset name, timestamp, metric or checkpoint path alone.", "Keep experiment definition, run execution, stage, attempt, checkpoint, dataset, source code, base model, produced model, evaluation, registry entry and deployment independently identifiable." ], "patch_rules": [ "Extensions declare task, method, framework, hardware, jurisdiction, rights, privacy, security, safety, cost, energy and interoperability effects.", "Released input, state, metric, checkpoint, outcome and lineage assertions are immutable; corrections create linked successors.", "Never silently change base model, dataset, code, configuration, method, evaluation, output, authority, privacy, security, retention or provenance." ], "compatibility_rules": [ "Ignore additive fields only when run identity, input and output bindings, method, state, source, authority, units, time, privacy and provenance survive.", "Every projection pins standard, implementation, schema, semantic convention, profile and mapping versions and declares information loss." ] }, "artifact_rules": { "identity_priority": [ "Authoritative master-system identifier for each run, attempt, checkpoint, metric series, event or output binding, qualified by issuer, namespace and record kind.", "Governed globally resolvable run IRI.", "Dimension UUID or ULID when neither preceding identifier exists." ], "timestamp_rule": "Use RFC 3339 timestamps with seconds and explicit offset or Z; distinguish planned, submitted, queued, started, checkpointed, observed, stopped, completed, recorded, ingested and knowledge times whenever they differ.", "serial_naming_rule": "Use {run-id}--{assertion-event-or-artifact-id}--{artifact-kind}--{revision-id}.", "integrity_rule": "Store digest, media type, record kind, run and attempt scope, input and output versions, actor, event and knowledge times, units, privacy marking and provenance." }, "policies": [ "The run does not own Dataset, Source Code, Base Model, Tokenizer, Trained Model Artifact, Evaluation, Registry, Deployment, Infrastructure, Secret, Policy, Provenance, Access Audit or Records masters.", "Desired configuration, requested resources, actual allocation, observed execution, measured result, interpretation and release decision are independently sourced assertions.", "A seed does not establish reproducibility across releases, platforms or hardware; claims require bounded environment and nondeterminism evidence.", "Agents cannot allocate unrestricted compute, expose secrets or protected data, waive rights, publish or deploy a model, or destroy data and artifacts outside explicit delegated authority." ], "crud": { "read": [ "Resolve purpose, current head, inputs, configuration, topology, attempts, progress, resources, checkpoints, metrics, outputs, lineage, outcome, access, retention and projection loss under the permitted view." ], "create": [ "Bind stable run identity, experiment or job definition, owner, objective, method, external inputs, source, authority, initial state and submission time before execution context." ], "update": [ "Append successor plan, state, resource, progress, metric, checkpoint, failure, output, correction and disclosure assertions with reason, authority, expected revision, event time and knowledge time." ], "delete": [ "Apply data-rights, privacy, security, legal-hold and adopting-Dimension records policy; retire or tombstone only the run assertion without cascading to external data, code, model or artifact masters, and let authoritative systems execute physical disposition." ] }, "roles": [ { "name": "AI system owner and accountable deployer", "responsibilities": [ "Own purpose, risk acceptance, release boundaries and accountable use of resulting artifacts." ] }, { "name": "Model or ML engineer", "responsibilities": [ "Define method and configuration, execute within delegation and preserve reproducible evidence." ] }, { "name": "Data owner and data steward", "responsibilities": [ "Authorize dataset versions, roles, rights, privacy, quality and permitted transformations." ] }, { "name": "Platform or infrastructure operator", "responsibilities": [ "Provide approved runtime, resource, telemetry, isolation, secret and incident controls." ] }, { "name": "Independent evaluator, safety and security reviewer", "responsibilities": [ "Review evaluation, abuse, privacy, security, safety and red-team evidence without becoming the run owner." ] }, { "name": "Model registry and release steward", "responsibilities": [ "Validate artifact identity, provenance, approval and promotion into separate registry and deployment systems." ] }, { "name": "Privacy, legal and records steward", "responsibilities": [ "Own lawful processing, intellectual-property, disclosure, correction, hold, retention and disposition profiles." ] } ], "access": { "default_rule": "Deny training data content, personal data, secrets, proprietary code, weights, checkpoints, security findings and restricted metrics unless a purpose-bound policy permits the minimum necessary view.", "scopes": [ "bundle", "layer", "finding", "artifact" ], "exceptions": [ "Declared incident response, audit, legal, subject-rights, security, safety or emergency access must cite authority, scope, purpose and time limit where applicable and must be logged." ], "audit_requirements": [ "Log actor, agent, role, purpose, run and attempt scope, operation, authority, policy, RFC 3339 time, affected fields, source revision and outcome without unnecessary secret or protected-data duplication." ] }, "agents_bootstrap": { "filename": "AGENTS.md", "required_fields": [ "Name", "Type", "Specification URL", "Storage type URL", "Interface URL", "Processes URL" ], "read_order": [ "Read Dimension AI, data, compute, privacy, security, safety, cost, retention and agent policies.", "Read this run and linked dataset, source code, base model, trained model, evaluation, registry, deployment, infrastructure, provenance and records models before mutation." ] } }, "coverage": { "claim": "WM-AI-006 covers one governed model training or fine-tuning execution aggregate from objective and immutable inputs through distributed progress, resources, checkpoints, evidence, outputs, lineage, terminal outcome, correction, retention and projection. Task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent external review remain deferred.", "confidence": "medium", "checklist": [ { "dimension": "identity", "status": "covered", "notes": "Identity is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "classification and direct properties", "status": "covered", "notes": "Classification and direct properties is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "recognition and observation", "status": "covered", "notes": "Recognition and observation is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "capabilities and possible actions", "status": "covered", "notes": "Capabilities and possible actions is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "composition", "status": "covered", "notes": "Composition is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "lifecycle", "status": "covered", "notes": "Lifecycle is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "relationships", "status": "covered", "notes": "Relationships is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "temporal", "status": "covered", "notes": "Temporal is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "spatial", "status": "covered", "notes": "Spatial is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "provenance", "status": "covered", "notes": "Provenance is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "ownership and stewardship", "status": "covered", "notes": "Ownership and stewardship is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "validation and quality", "status": "covered", "notes": "Validation and quality is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "access and privacy", "status": "covered", "notes": "Access and privacy is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "retention and deletion", "status": "covered", "notes": "Retention and deletion is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "interoperability", "status": "covered", "notes": "Interoperability is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." }, { "dimension": "authority and ethics", "status": "covered", "notes": "Authority and ethics is explicit; task, method, framework, hardware, data-rights, jurisdiction, evaluation, release-pinned mappings and independent review remain held where applicable." } ], "known_omissions": [ "Claude and Grok each timed out on one bounded attempt; no independent external result was admitted.", "The relation-ledger edges WM-AI-006 REFERENCE WM-DAT-001 and WM-AI-006 PRODUCES WM-SFT-004 are candidates and grant no target ownership, mutation, release or cascade authority.", "Pretraining, supervised and preference fine-tuning, continual learning, distillation, adapter tuning and other methods require explicit profiles.", "NIST AI RMF 1.0 is under revision; this result pins the inspected 1.0 publication and does not predict the revision." ], "conflicts": [ "Experiment or job definition, run, stage, task, attempt and checkpoint are not interchangeable identities.", "Base model, trained candidate, selected model artifact, registry entry and deployment are separate lifecycle objects.", "Requested resources, actual allocation, observed use, invoice cost, energy and carbon estimates use different sources and measurement methods." ], "regional_assumptions": [ "Training data rights, privacy, intellectual property, security, safety, export, environmental reporting, records and high-risk AI obligations depend on jurisdiction, industry and use case.", "The EU AI Act and GDPR are European Union profiles; NIST publications are voluntary United States public-authority guidance unless adopted by policy or contract.", "MLflow, MLMD, OpenLineage, Kubeflow, SLSA, OCI, OpenTelemetry, MLPerf, SCI and PyTorch are versioned profiles, not universal lossless schemas." ], "adversarial_checks": [ "Reject a run without stable identity, owner, objective, method, immutable input references, source, authority, state, time and current head.", "Reject a training artifact whose base, data, code, configuration, builder, checkpoint and digest lineage cannot be reconstructed.", "Reject reproducibility claims based only on a random seed or a successful rerun on a different release, platform or hardware.", "Reject autonomous resource acquisition, protected-data access, secret disclosure, model publication, deployment or destructive cleanup without delegated authority.", "Reject benchmark, quality, safety, cost, energy or carbon claims without method, units, boundary, source, uncertainty and version-qualified evidence." ] }, "researchAdjudication": { "providerMode": "single-provider-waiver", "activeProviders": [ "codex" ], "waivedProviders": [ "claude", "grok" ], "providerPolicy": { "contract_version": "1.0.0", "mode": "single-provider-waiver", "effective_at": "2026-09-06T00:00:00Z", "scope": "Canonical single-stream subject-model research after the six-workstream consolidation", "active_providers": [ "codex" ], "waived_providers": [ { "provider": "claude", "authorized_by": "repository owner", "authorized_at": "2026-09-06T00:00:00Z", "reason": "Claude produced no result on prior 1800-second and 900-second attempts and again timed out on bounded 600-second Sonnet and 300-second Haiku passes. The owner prioritized completion over provider availability." }, { "provider": "grok", "authorized_by": "repository owner", "authorized_at": "2026-09-06T00:00:00Z", "reason": "The repository owner authorized completion without Grok when Grok is unavailable, slow or schema-invalid. Grok may still be attempted as a bounded supplemental reviewer, but its failure never blocks a valid Claude plus no-tools result." } ], "review_rule": "Codex may complete source-grounded fallback research after bounded Claude and Grok attempts fail. It requires a separate no-tools adversarial audit and remains reviewable-draft with a visible absence-of-external-review hold.", "supplemental_provider_attempts": [ { "provider": "claude", "required": false, "maximum_attempts": 1, "failure_policy": "record-and-continue", "admission_rule": "Use only a locally schema-valid result whose sources and boundaries survive adjudication." }, { "provider": "grok", "required": false, "maximum_attempts": 1, "failure_policy": "record-and-continue", "admission_rule": "Use only a locally schema-valid result whose sources and boundaries survive adjudication." } ] }, "boundaryDecision": { "entry_kind": "aggregate", "status": "accepted", "rationale": "A training run composes independently identifiable input bindings, stages, tasks, attempts, events, metrics, checkpoints, output bindings and governance assertions over one execution identity. Aggregate is more accurate than experiment definition, event, artifact or trained model. The frozen record_plane world-model and catalogue entry_kind standalone-mm are separate axes." }, "decisions": [ { "concept": "Run versus experiment definition and trained model", "disposition": "accepted", "rationale": "The aggregate owns one execution history. Reusable experiment or job definitions, base models, produced artifacts, registry entries and deployments remain external masters." }, { "concept": "Dataset reference", "disposition": "accepted-candidate", "rationale": "WM-AI-006 REFERENCE WM-DAT-001 is a candidate edge. The run may bind immutable dataset roles, splits, transforms and rights, but cannot own or mutate dataset content or lifecycle." }, { "concept": "Produced model relation", "disposition": "accepted-candidate-as-reference", "rationale": "WM-AI-006 PRODUCES WM-SFT-004 remains candidate. The current schema represents it as a non-owning output reference plus derivation evidence, with no registry, release, deployment or cascade authority." }, { "concept": "Method profiles", "disposition": "accepted-with-profile-hold", "rationale": "Pretraining, supervised and preference fine-tuning, continual learning, distillation, adapter tuning and other methods share a run core but need separate task, data, objective and risk profiles." }, { "concept": "Desired, allocated and observed execution", "disposition": "accepted", "rationale": "Configuration and requested resources are plans; scheduler allocation and telemetry are observations. They retain independent source, time, units and provenance." }, { "concept": "Reproducibility", "disposition": "accepted-with-bounded-claim", "rationale": "Seeds and deterministic settings are evidence, not proof. Replay claims pin framework, release, platform, hardware, dependencies, environment and known nondeterminism." }, { "concept": "Evaluation, publication and deployment", "disposition": "accepted-external", "rationale": "The run binds evaluation and selection evidence but does not own independent evaluation conclusions, registry promotion, publication or deployment decisions." }, { "concept": "Cost, energy and carbon", "disposition": "accepted-with-measurement-hold", "rationale": "Requested capacity, observed use, invoice cost, energy and carbon intensity use different boundaries and methods. Values require source, unit, interval, functional unit and uncertainty." }, { "concept": "Privacy, intellectual property, security and jurisdiction", "disposition": "accepted-with-mandatory-hold", "rationale": "Dataset and weight access, personal data, secrets, proprietary code, export, incident, retention and release controls remain purpose-bound adopting-Dimension policies." }, { "concept": "Machine-readable projections", "disposition": "accepted-with-validation-hold", "rationale": "MLflow, MLMD, OpenLineage, PROV, Kubeflow, SLSA, OCI, OpenTelemetry, MLPerf, SCI and PyTorch cover different scopes. Every mapping must pin release, profile and information loss." }, { "concept": "Single-provider waiver and no-tools audit", "disposition": "accepted-with-mandatory-hold", "rationale": "One Claude Sonnet and one Grok 4.6 research attempt each timed out after 120 seconds. Codex separately audited the frozen validated result and comparison locally without tools or new research facts. Confidence remains medium and assurance remains reviewable-draft." } ], "publicationHolds": [ "Absence-of-external-review hold: one Claude Sonnet and one Grok 4.6 research attempt for WM-AI-006 each timed out after 120 seconds. No external research result was admitted.", "Relation hold: WM-AI-006 REFERENCE WM-DAT-001 and WM-AI-006 PRODUCES WM-SFT-004 remain candidate edges and grant no target ownership, mutation, release, deployment or cascade behavior.", "Method-profile hold: pretraining, supervised and preference fine-tuning, continual learning, distillation, adapter tuning and other methods require explicit profiles.", "Rights and jurisdiction hold: data rights, privacy, intellectual property, security, safety, export, incident, retention and high-risk AI duties require adopting-Dimension and jurisdiction policies.", "Evaluation and release hold: benchmark success or an observed best checkpoint does not itself authorize model selection, publication, registry promotion or deployment.", "Measurement hold: cost, energy and carbon assertions require source, unit, interval, system boundary, functional unit, method and uncertainty.", "Interoperability hold: all MLflow, MLMD, OpenLineage, PROV, Kubeflow, SLSA, OCI, OpenTelemetry, MLPerf, SCI and PyTorch projections require release-pinned mappings and loss validation.", "Source-version hold: NIST AI RMF 1.0 is under revision; this specification pins the inspected 1.0 publication.", "Independent external review was explicitly waived by the repository owner; this codex-only result remains a reviewable draft." ], "deferredResearch": [ "Approve or reject candidate dataset-reference and produced-model relations and register code, base-model, evaluation, registry, deployment, infrastructure, secret, provenance and records relations.", "Create task and method profiles for pretraining, supervised and preference fine-tuning, continual learning, distillation, adapter tuning and other training methods.", "Validate jurisdiction and organization-specific data-rights, privacy, intellectual-property, export, security, safety, cost, energy, incident, retention and release policies.", "Test release-pinned MLflow, MLMD, OpenLineage, PROV, Kubeflow, SLSA, OCI, OpenTelemetry, MLPerf, SCI and PyTorch mappings with conformance, round-trip and information-loss evidence.", "Refresh the NIST AI RMF mapping after a new normative revision and obtain supplemental independent external review before canonical promotion." ] }, "statistics": { "sources": 17, "bundles": 6, "layers": 12, "findings": 24, "questions": 72, "artifacts": 24, "functions": 10 } }