{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp05_complete_proposal_portfolio20_20260803","cell_id":"layer_decay_and_expiration_management__physics","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"physics_dependency_aware_monte_carlo_configuration_lifecycle","proposal_index":2,"version":0,"title":"Dependency-Aware Lifecycle for Markov-Chain Physics Configurations","problem":"A computational-physics campaign can generate a sequential stack of field configurations from one or more Markov chains. Early thermalization states, statistically redundant configurations, superseded parameter runs, production samples, restart checkpoints, and anomalous states can remain together in active storage after their scientific and operational roles have diverged. If their status is not explicit, invalid or superseded configurations may enter estimators as though they belonged to the current ensemble. Keeping every configuration active consumes constrained storage and obscures the authoritative sample, while deleting old states by trajectory number or age can remove restart anchors, rare configurations, provenance needed to reconstruct an estimator, or raw states required for later observables.","actors":["computational physicist defining the ensemble","simulation-production lead operating the Markov chains","statistical analyst estimating thermalization and autocorrelation","physics analyst consuming configurations and derived observables","research data steward","high-performance-computing storage operator","reproducibility reviewer"],"observable_state":"For each generated configuration, observable metadata include ensemble and chain identity, trajectory index, generation time, action and physical parameters, code build, random-number provenance, predecessor or restart checkpoint, thermalization classification, validation results, measured observables, estimated correlation context, analysis manifests that reference it, regeneration cost, last access, storage tier, preservation holds, and lifecycle state. The problematic state is an accumulation in which configuration files remain equally discoverable despite differing validity and dependency status, while storage operators cannot determine which files may be demoted or removed without affecting estimators, restarts, or reconstruction.","consequence":"A stale or misclassified configuration admitted to an active ensemble can change the sample used for a physics estimate without an explicit scientific decision. Conversely, irreversible removal of a referenced configuration, checkpoint, or provenance-bearing state can prevent an analysis from being reproduced, a chain from being restarted, or a newly defined observable from being evaluated on the original field state. Indefinite active retention also makes authoritative ensemble boundaries harder to inspect and maintain.","affected_objective":"Keep the configuration set used for physics inference explicit and statistically qualified while bounding active storage and preserving sufficient raw states, checkpoints, provenance, and exceptions for restart, reconstruction, and authorized reanalysis.","intervention":"Install a configuration lifecycle controller around one simulation ensemble. Register every field configuration and checkpoint in deposition order and assign an explicit state: thermalization, active production sample, questioned, superseded, warm compressed, cold archive, quarantined for removal, or disposed. Statistical review determines whether a configuration is eligible for the active sample; a separate age-weighted operational-value score ranks inactive configurations for storage review using last access, uniqueness, regeneration cost, provenance value, and restart value, with hard vetoes for unresolved scientific status and dependencies. A retention matrix assigns different paths to production samples, burn-in states, checkpoints, anomalous configurations, and superseded runs. Before irreversible removal, trace inbound references from analysis manifests, ensemble definitions, restart chains, publications in preparation, and preservation holds. Move cleared configurations through hot, warm, and cold tiers; place removal candidates in recoverable quarantine; periodically restore sampled archives end to end; and leave a tombstone containing identity, disposition, retained observables, provenance, and any successor ensemble version.","structural_mapping":[{"archetype_element":"Sequential deposits accumulate after their roles change","domain_realization":"Each Markov trajectory deposits another field configuration, but thermalization, correlation, supersession, completed analyses, and restart progress change what that configuration is useful for."},{"archetype_element":"Layer inventory and identity map","domain_realization":"A registry resolves each configuration and checkpoint to its ensemble, chain, trajectory, physical parameters, code build, provenance, and storage object."},{"archetype_element":"Deposition-order and age index","domain_realization":"Trajectory order records causal generation sequence, while generation time and time since validation or access support operational aging decisions without substituting for statistical qualification."},{"archetype_element":"Decay and value rule","domain_realization":"Operational retention value can decay with time and inactivity, but uniqueness, restart value, regeneration cost, anomalous physics content, and live dependencies can preserve or raise a configuration's standing."},{"archetype_element":"Expiration trigger with differentiated disposition","domain_realization":"Completion of thermalization review, ensemble supersession, elapsed class-review periods, or storage-tier thresholds trigger reevaluation; expiration can mean loss of active authority, compression, archival, quarantine, or eventual removal rather than automatic destruction."},{"archetype_element":"Dependency and reconstruction check","domain_realization":"A configuration cannot leave recoverable custody while an analysis manifest, ensemble estimator, restart chain, provenance audit, or approved reanalysis depends on its raw state."},{"archetype_element":"Preservation exception register","domain_realization":"Named configurations may be held because they anchor a restart, exhibit an unresolved anomaly, support an estimator reconstruction, or represent a parameter region selected for later observables; each hold has an owner and review date."},{"archetype_element":"Reversible cleanup pathway","domain_realization":"Files cleared for removal first become invisible to ordinary analysis and enter recoverable quarantine, allowing dependency mistakes to be corrected before bytes are destroyed."},{"archetype_element":"Deletion evidence and successor marker","domain_realization":"A tombstone survives removal and records that the configuration existed, why it was removed, which metadata and derived observables remain, and whether another ensemble version supersedes it."},{"archetype_element":"Review and revalidation loop","domain_realization":"Recurring statistical-status review, exception renewal, archive-restore drills, and storage review keep both the active ensemble and retained historical stack inspectable."}],"mechanism_mapping":[{"mechanism_slug":"stale_layer_detection_dashboard","role":"Maintains the identity-resolved configuration inventory and surfaces files whose active label conflicts with current ensemble definitions, validation results, physical parameters, or software provenance.","counterfactual_removal":"Without the inventory and stale-context view, lifecycle decisions remain split across directories, scheduler logs, notebooks, and analysis manifests, allowing superseded or invalid configurations to appear current."},{"mechanism_slug":"age_weighted_value_score","role":"Ranks statistically inactive configurations for retention review by applying an age or inactivity decay to operational value while allowing provenance, anomaly, restart, regeneration, and dependency factors to override the decay. It cannot determine statistical validity or authorize deletion.","counterfactual_removal":"Without a comparable ranking input, inactive configurations must be reviewed ad hoc or removed by trajectory age alone, obscuring configurations that are old and unused but expensive or impossible to reconstruct."},{"mechanism_slug":"retention_schedule","role":"Maps configuration classes to minimum retention, review cadence, eligible storage tiers, disposition paths, and named preservation exceptions.","counterfactual_removal":"Without class-specific rules, burn-in states, production samples, checkpoints, and anomalous configurations receive inconsistent treatment, and temporary holds can become indefinite."},{"mechanism_slug":"dependency_safe_delete_check","role":"Blocks irreversible removal until live inbound references from analyses, ensemble manifests, restart chains, and preservation holds have been resolved.","counterfactual_removal":"Without this gate, a configuration judged redundant for one estimator could be removed despite remaining load-bearing for another observable, a restart, or a reconstruction."},{"mechanism_slug":"lifecycle_storage_tiering_policy","role":"Moves retained configurations from active storage to compressed warm storage and then to cold archive as access cools, without equating demotion with deletion.","counterfactual_removal":"Without tiering, the system must choose between keeping every retained raw configuration on the active filesystem and destroying states whose value is infrequent but legitimate."},{"mechanism_slug":"soft_delete_quarantine_window","role":"Makes removal two-stage by hiding a cleared configuration from ordinary analysis while retaining a recoverable copy for a grace period sized by regeneration and dependency impact.","counterfactual_removal":"Without quarantine, an incomplete reference graph or incorrect classification becomes irreversible at the first deletion action."},{"mechanism_slug":"archive_restore_test","role":"Samples archived configurations across ensemble, format, and age classes and exercises retrieval, decompression, parsing, integrity validation, and reconnection to a controlled analysis environment.","counterfactual_removal":"Without an end-to-end restore drill, archive metadata may report success even when stored configurations can no longer be parsed or connected to an executable reconstruction path."},{"mechanism_slug":"tombstone_or_deletion_marker","role":"Preserves the removed configuration's identity, trajectory position, disposition rationale, retained metadata, derived-observable locations, and successor resolution.","counterfactual_removal":"Without a tombstone, a missing trajectory is ambiguous between intentional removal, failed generation, corruption, or misplaced storage, and historical references cannot resolve consistently."}],"causal_chain":["A simulation deposits sequential configurations whose scientific validity and operational value later diverge.","The registry binds every deposited state to its trajectory, ensemble definition, provenance, dependencies, and current lifecycle status.","Statistical review separates membership in the authoritative physics sample from the separate question of how an inactive configuration should be retained.","The operational-value score and retention matrix rank inactive layers for review while vetoes protect anomalous, unique, restart-critical, or referenced states.","Dependency tracing prevents configurations required by analyses or reconstruction paths from entering irreversible disposition.","Tiering and compression move infrequently used but retained states out of active storage without erasing their identity or claimed recoverability.","Quarantine supplies a rollback interval for classification or dependency errors, after which approved files may be removed.","Restore drills test whether archived raw states remain usable, and tombstones preserve lineage after physical deletion.","Periodic revalidation of active labels, archives, and exceptions keeps the authoritative ensemble distinct from the bounded historical stack."],"baseline":"The baseline is directory- and campaign-level management: configurations are written sequentially, analysts maintain separate lists of thermalization cuts and selected trajectories, storage is reclaimed through fixed-stride thinning, deletion of old runs, or manual migration, and checkpoints are retained according to operator judgment. Statistical status, storage status, analysis dependencies, and preservation exceptions are not governed through one inspectable lifecycle.","nearest_rivals":["Retain every raw configuration indefinitely: maximizes immediate reanalysis options but leaves invalid, superseded, and authoritative states insufficiently distinguished and keeps all files in the retention burden.","Fixed-stride thinning: retains every nth trajectory and removes the rest, but a uniform stride does not represent dependency, anomaly, restart, provenance, or changing correlation structure.","Delete all burn-in and superseded configurations after validation: simplifies the ensemble but can erase evidence needed to reassess thermalization, diagnose chain behavior, or reconstruct how the accepted boundary was chosen.","Compress every configuration without lifecycle classification: reduces file size while preserving the same ambiguity about authority, dependencies, exceptions, and eventual disposition.","Retain only random seeds and periodic checkpoints and regenerate other states when needed: reduces stored raw states but depends on executable reproducibility, stable software and hardware behavior, and a viable restart chain.","Manage retention only at whole-ensemble granularity: provides a clear project boundary but cannot preserve selected checkpoints or anomalous states while expiring redundant members of the same ensemble."],"remaining_contrastive_claim":"After accounting for these rivals, the proposal's specific claim is that statistical authority and retention value are separate lifecycle dimensions for each sequential configuration: a state may expire from the active ensemble yet remain archived for reconstruction, or remain statistically valid while moving to a colder tier. Explicit dependencies, reversible disposition, restore testing, and tombstones govern transitions between those states.","authority_safety":{"decision_authority":"The ensemble's scientific lead controls statistical membership; the simulation-production lead controls restart designations; the data steward and storage operator authorize tier transitions under existing policy; irreversible removal requires a cleared dependency verdict and joint approval from the scientific lead and data steward. The score has no independent authority.","authorized_first_step":"Apply the proposed registry, lifecycle labels, score, and dependency trace retrospectively in shadow mode to one completed, non-publication-critical ensemble. Use existing metadata and read-only analysis manifests, and perform restore tests only into an isolated scratch environment from selected archive copies. Do not migrate, hide, rewrite, or delete source configurations.","excluded_actions":["changing the configuration membership of an active or published estimator without scientific review","automatically deleting or demoting files from the age-weighted score","treating estimated statistical redundancy as proof that a raw state has no reconstruction value","removing a checkpoint before validating all restart descendants and inbound references","overwriting original configuration metadata or analysis manifests","using a restored archive in production before validation","destroying quarantined configurations during the bounded first evidence step"],"halt_rollback":"Stop the pilot if configuration identities cannot be reconciled with trajectory records, restore activity risks changing source archives, the dependency trace misses a known analysis, or lifecycle labels would silently alter an estimator. Delete only isolated scratch restorations under ordinary scratch policy, retain all source files and original manifests unchanged, and mark every pilot classification non-authoritative."},"negative_tests":{"strongest_counterevidence":"Immutable ensemble manifests already distinguish every valid and invalid state, complete dependency metadata make reconstruction automatic, and retaining the entire raw ensemble in one managed tier imposes no relevant storage, search, maintenance, or interpretive constraint. Under those conditions, configuration-level expiration adds governance without resolving the stated tension.","problem_falsifier":"The problem is falsified for the test ensemble if every configuration's statistical status and dependencies are already explicit and consistently enforced, no stale or superseded state can enter an analysis accidentally, all raw states can remain maintainably retained, and no disposition decision is required.","intervention_falsifier":"The intervention is falsified if the shadow registry cannot reproduce authoritative ensemble membership, misses known analysis or restart dependencies, assigns unstable lifecycle states under reasonable statistical judgments, or produces no decision-relevant distinction beyond fixed-stride thinning or whole-ensemble retention. Failure to restore sampled archives through the documented path also falsifies the claimed archival safeguard until corrected.","risks":["Lifecycle selection rules could introduce unrecognized selection bias if operational disposition is confused with statistical sampling.","Thermalization and autocorrelation estimates can be uncertain, making apparently precise classifications contestable.","Analysis dependencies expressed only in notebooks, copied files, or external systems may be invisible to the deletion gate.","Compression or format migration can change numerical readability or detach provenance.","Archive-restore success on a sample does not establish recoverability of every format or configuration.","Quarantine duplicates may temporarily increase storage use and be mistaken for ordinary analysis copies.","Regeneration from seeds may fail if software, numerical libraries, hardware behavior, or random-number implementations change.","Tombstone and derived-observable retention cannot substitute for a raw configuration when an unforeseen observable is later proposed."]},"next_evidence_step":"For one completed ensemble, reconcile a bounded set of configurations spanning thermalization, production, checkpoints, anomalies, and superseded trajectories against authoritative manifests and known analyses. Have one team assign shadow lifecycle states and dependency verdicts, while an independent physics-and-data panel classifies the same set without seeing the score. Restore a stratified sample of archived copies into isolation and test parsing, metadata integrity, and execution of one already documented observable. Compare disagreements, missed dependencies, unresolved identities, restore failures, and disposition differences against fixed-stride thinning and whole-ensemble retention. Source files, ensemble manifests, and published results remain unchanged.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 governs physical debris films accumulated shot by shot on optical shields, with optical-transfer drift leading to cartridge removal, physical quarantine, cleaning, or specimen preservation. This proposal instead governs digital field configurations accumulated trajectory by trajectory in a Markov-chain simulation, where the central failure is confusion between statistical ensemble authority and retention value. Its intervention uses ensemble manifests, thermalization and correlation status, restart dependencies, storage tiering, executable archive restoration, and configuration tombstones rather than optical measurements or shield handling. Its causal path runs from sequential simulation states through statistical qualification and dependency-aware data disposition, not from deposited matter through calibration degradation and physical replacement. It can be adopted by a computational-physics campaign without adopting the debris-shield protocol, and the shield protocol can be adopted without this configuration lifecycle.","revision_record":{"parent_version":null,"progress_targets_addressed":["Created one complete proposal at index 2.","Addressed a materially different physics problem from proposal 1.","Specified an independently adoptable intervention, causal chain, authority model, safeguards, rivals, falsifiers, and bounded evidence step.","Explicitly contrasted the proposal with every earlier sealed proposal."],"conceptual_changes":["Initial formulation; no parent version.","Instantiated accumulated layers as sequential Markov-chain field configurations whose statistical authority and retention value diverge.","Separated scientific ensemble membership from storage disposition."],"operational_changes":["Initial formulation; no parent version.","Defined per-configuration lifecycle states, dependency gates, storage tiers, quarantine, archive restoration, and tombstones.","Restricted initial adoption to retrospective shadow classification and isolated restore tests."],"evidence_changes":["Initial formulation; no parent version.","Specified independent classification, dependency auditing, and archive restoration against fixed-stride and whole-ensemble baselines."],"claim_changes":["Initial formulation; no parent version.","Limited the contrastive claim to lifecycle separation of statistical authority, retention value, recoverability, and irreversible disposition.","Made no claim of novelty, prevalence, demand, or effect size."]}}