{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"layer_decay_and_expiration_management__data_science","trajectory_id":"R","attempt_index":0,"archetype_slug":"layer_decay_and_expiration_management","domain_slug":"data_science","decision":"CANDIDATE","problem_id":"stale_analytical_artifact_accumulation","causal_lever_id":"governed_analytical_artifact_lifecycle","proposal":{"problem":"Data-science platforms accumulate superseded datasets, feature snapshots, model versions, notebooks, and reports that remain discoverable without a reliable lifecycle state. Analysts may reuse obsolete artifacts as current evidence, while owners avoid cleanup because deletion could break lineage, reproducibility, audits, or rollback.","actors_substrate":["data scientists and analysts","data and ML platform operators","model-risk, privacy, compliance, and audit staff","downstream decision-makers","versioned datasets, features, models, notebooks, reports, and lineage metadata"],"observable_state":"A registry shows many old or superseded analytical artifacts still marked active, uncertain ownership and inbound dependencies, inconsistent retention rules, stale artifacts appearing in ordinary discovery, and no demonstrated restore path for archived versions.","consequence":"Obsolete analytical context competes with current evidence, storage and maintenance burdens grow, and improvised deletion can erase reproducibility or required records.","affected_objective":"Keep analytical discovery current and the artifact estate bounded while preserving required lineage, reconstruction, compliance, and rollback capability.","structural_mapping":[{"archetype_element":"sequential deposits whose active usefulness changes","domain_realization":"Successive datasets, feature snapshots, model versions, notebooks, and reports accumulate after supersession.","claim_kind":"INFERENCE"},{"archetype_element":"old layers remain active by default","domain_realization":"Unlabeled or weakly labeled artifacts remain searchable and can be reused as though current.","claim_kind":"HYPOTHESIS"},{"archetype_element":"removal threatens reconstruction and accountability","domain_realization":"Deleting inputs, code-linked outputs, or model versions may break lineage, reproducibility, audit, or rollback.","claim_kind":"INFERENCE"},{"archetype_element":"lifecycle governance over the stack","domain_realization":"A registry assigns lifecycle state, retention class, staleness status, dependencies, exceptions, and disposition to each artifact.","claim_kind":"HYPOTHESIS"},{"archetype_element":"boundedness without loss of needed memory","domain_realization":"Superseded artifacts leave active discovery through refresh, archive, compaction, quarantine, or deletion while protected versions remain recoverable.","claim_kind":"HYPOTHESIS"}],"component_map":[{"component":"layer_inventory_and_identity_map","status":"direct","domain_realization":"Canonical analytical-artifact registry with stable identities."},{"component":"deposition_order_and_age_index","status":"direct","domain_realization":"Creation time, version sequence, and supersession chronology."},{"component":"retention_policy_matrix","status":"direct","domain_realization":"Artifact-class retention and disposition table."},{"component":"decay_function_or_aging_rule","status":"adapted","domain_realization":"Class-specific review horizons; age raises review priority but never independently authorizes deletion."},{"component":"expiration_trigger_definition","status":"direct","domain_realization":"TTL, supersession, owner departure, failed validation, or inactivity triggers lifecycle review."},{"component":"layer_value_and_risk_score","status":"adapted","domain_realization":"Transparent value, sensitivity, reproducibility, use, and recreation-cost fields rather than one authoritative score."},{"component":"dependency_and_reconstruction_check","status":"direct","domain_realization":"Pre-disposition scan of lineage, jobs, reports, audits, and reconstruction paths."},{"component":"pruning_or_archival_pathway","status":"direct","domain_realization":"Refresh, demote, compact, archive, quarantine, or destroy pathway."},{"component":"preservation_exception_register","status":"direct","domain_realization":"Named reproducibility, evidentiary, historical, and safety exceptions with owners."},{"component":"deletion_audit_and_rollback_window","status":"direct","domain_realization":"Audited soft deletion followed by a class-appropriate recovery window."},{"component":"resource_budget_and_storage_tier","status":"direct","domain_realization":"Storage budget and hot, warm, cold, and archive placement."},{"component":"review_cadence_and_revalidation_loop","status":"direct","domain_realization":"Periodic revalidation of stale flags, exceptions, policies, and restore ability."},{"component":"legal_hold_or_compliance_override","status":"direct","domain_realization":"Privacy, litigation, regulatory, and records-management overrides."},{"component":"hot_warm_cold_access_state","status":"direct","domain_realization":"Access tier reflecting expected retrieval frequency and latency."},{"component":"orphan_layer_detection","status":"direct","domain_realization":"Detection of artifacts with no known live inbound dependency."},{"component":"supersession_marker","status":"direct","domain_realization":"Deleted or superseded identity resolves to status and optional successor."},{"component":"layer_compaction_candidate_flag","status":"adapted","domain_realization":"Flag for lossless packaging, deduplication, or snapshot consolidation; semantic rollups require separate validation."},{"component":"rehydration_or_restore_procedure","status":"direct","domain_realization":"Documented and exercised artifact restoration with lineage reconnection."},{"component":"deletion_impact_estimate","status":"direct","domain_realization":"Estimate of affected consumers, reconstruction cost, sensitivity, and irreversibility."},{"component":"stale_context_detector","status":"direct","domain_realization":"Signals for supersession, invalid references, changed source semantics, failed validation, or departed ownership."}],"mechanism_dispositions":[{"slug":"age_weighted_value_score","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Reject a single composite score as an authorization mechanism because judgmental weights can conceal reproducibility or compliance vetoes; retain separate inspectable fields.","counterfactual_removal":"No causal loss because class rules, staleness evidence, dependencies, and explicit overrides govern disposition."},{"slug":"archive_restore_test","disposition":"selected_load_bearing","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Sample archived artifacts across classes and ages and test retrieval, readability, fidelity, and lineage reconnection.","counterfactual_removal":"Archival would be an unverified promise, so deletion from active tiers could silently eliminate reconstruction capability."},{"slug":"cache_eviction_rule","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Fast-tier eviction optimizes latency under instantaneous capacity pressure, not semantic validity, retention, or safe disposition.","counterfactual_removal":"Lifecycle governance remains intact; storage tier transitions cover the relevant resource problem."},{"slug":"dependency_safe_delete_check","disposition":"selected_load_bearing","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Gate destructive action on scans of live lineage, scheduled jobs, reports, external registrations, and audit paths; uncertainty blocks deletion.","counterfactual_removal":"Cleanup could break downstream decisions, reproducibility, or audits, invalidating safe expiration."},{"slug":"lifecycle_storage_tiering_policy","disposition":"selected_supporting","contribution_type":"OPERATIONAL","adaptation_or_rejection":"Move infrequently used but retained artifacts to cheaper tiers; allow only validated lossless compaction by default.","counterfactual_removal":"Validity control remains, but retained history stays unnecessarily expensive and resource boundedness weakens."},{"slug":"log_rotation_and_cleanup_job","disposition":"considered_rejected","contribution_type":"NONE","adaptation_or_rejection":"Generational rotation is content-blind and fits append-only logs better than heterogeneous, dependency-rich analytical artifacts.","counterfactual_removal":"No material loss because class-specific lifecycle actions replace positional purging."},{"slug":"retention_schedule","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Map analytical-artifact classes to minimums, maximums, review points, disposition actions, exceptions, and legal holds.","counterfactual_removal":"There would be no legitimate, consistent rule for when artifacts may or must leave active status."},{"slug":"soft_delete_quarantine_window","disposition":"selected_load_bearing","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Hide eligible artifacts from normal discovery while retaining audited restoration for a risk-sized grace period; hard-delete mandates override quarantine where required.","counterfactual_removal":"Early cleanup errors become immediately irreversible, hard-gating safe automation."},{"slug":"stale_layer_detection_dashboard","disposition":"selected_load_bearing","contribution_type":"CORE_CAUSAL","adaptation_or_rejection":"Expose the canonical inventory and evidence-based staleness signals to owners without authorizing disposal.","counterfactual_removal":"Stale active artifacts remain hard to distinguish from current ones, so the main misuse pathway persists."},{"slug":"time_to_live_ttl_policy","disposition":"selected_supporting","contribution_type":"OPERATIONAL","adaptation_or_rejection":"Use TTL only for preclassified temporary artifacts; expiry means refresh or review eligibility, never unconditional destruction.","counterfactual_removal":"Governance still works, but naturally temporary artifacts require more periodic review."},{"slug":"tombstone_or_deletion_marker","disposition":"selected_supporting","contribution_type":"SAFETY_GUARDRAIL","adaptation_or_rejection":"Preserve a minimal marker recording deletion or supersession and successor resolution, with its own expiry rule.","counterfactual_removal":"References can confuse deletion with absence and stale indexes may revive or rediscover retired artifacts."}],"causal_chain":["Inventory and identity resolution make accumulated analytical artifacts inspectable.","Staleness and supersession evidence identifies artifacts that should stop presenting as current.","Retention classes, holds, and named exceptions determine permitted disposition.","Dependency and impact checks block unsafe removal and select refresh, archive, quarantine, or deletion.","Tiering and quarantine remove eligible artifacts from active discovery while preserving bounded recovery.","Restore drills and periodic revalidation verify that retained reconstruction promises remain real.","Tombstones and audit records preserve accountability after the active artifact is removed."],"baseline":"Ordinary practice is owner-by-owner cleanup: artifacts remain active until storage pressure or an incident prompts ad hoc archival or deletion, with inconsistent checks and labels.","nearest_rival":"A metadata catalog with current/superseded badges and search ranking but no retention authority, dependency-gated disposition, reversible deletion, or tested restoration.","authority_safety":{"affected_parties":["artifact owners","downstream analysts and decision-makers","data subjects represented in retained data","platform operators","audit, legal, privacy, and model-risk teams"],"decision_authority":"Artifact owners may refresh or propose demotion; the platform lifecycle service may execute only policy-authorized reversible actions; data governance, legal, privacy, or model-risk authorities approve contested, irreversible, held, or sensitive dispositions.","authorized_first_step":"Run a 30-day shadow pilot on one non-production artifact class: inventory it, label suspected staleness, simulate policy decisions, and archive then restore a stratified sample. No hard deletion is authorized.","excluded_actions":["automatic hard deletion based only on age, inactivity, or a score","removing artifacts with unresolved dependencies or ownership","overriding legal holds, minimum retention, privacy duties, or named safety exceptions","lossy compaction presented as reproducible source data","using stale flags as proof that analytical conclusions are false"],"halt_rollback":"Halt on an unresolved dependency, incorrect stale classification affecting active work, policy conflict, failed restore, missing audit record, or unauthorized sensitive-data exposure; restore quarantined artifacts and revert discovery labels from the preserved registry state."}},"negative_tests":{"strongest_counterevidence":"If obsolete artifacts are already reliably excluded from discovery, ownership and dependencies are complete, retention is consistently enforced, storage remains bounded, and sampled archives restore successfully, the proposed problem is not materially present.","analogy_break":"Analytical artifacts are not interchangeable sediment: an old dataset or model may remain uniquely necessary for reproducibility, and semantic validity does not decay monotonically with age. Therefore age can trigger review but cannot establish worthlessness or deletion safety.","failure_condition":"The approach fails if lineage is materially incomplete, lifecycle metadata is not maintained, archive restoration is unreliable, or governance exceptions become permanent unreviewed holds.","problem_falsifier":"A representative registry audit finds no meaningful accumulation of superseded-but-active artifacts, no stale-artifact reuse, and no cleanup avoidance attributable to reconstruction, audit, or compliance uncertainty.","intervention_falsifier":"In the bounded pilot, lifecycle labels and shadow dispositions do not reduce stale artifacts in ordinary search or owner triage time, or they create unacceptable false retirements, dependency breaks, restore failures, or compliance exceptions relative to the baseline.","risks":["False staleness flags can hide valid evidence or bias analysts toward newer artifacts.","Incomplete dependency graphs can authorize destructive mistakes.","Cold-tier retrieval latency can obstruct incident response or reproducibility.","Quarantine may violate hard-deletion commitments by retaining data longer than allowed.","Lifecycle metadata, tombstones, and exceptions can themselves accumulate.","Central governance can concentrate authority and slow legitimate analytical work."]},"null_rationale":null,"classification":{"candidate_kind":"MECHANISM_COMPOSITION","prior_art_status":"UNSEARCHED","evidence_maturity":"HYPOTHESIS"},"revision_change_log":{"revision_kind":"ORIGINAL","prior_problem_id":null,"prior_causal_lever_id":null,"problem_changed":false,"causal_lever_changed":false,"conceptual_changes":[],"operational_changes":[],"repairs_addressed":[]},"confidence":0.88,"generator_notes":"Closed-book structural inference from the supplied trajectory packet. Empirical prevalence and effect claims remain hypotheses; no prior-art search was performed."}