{"actors":["Digital preservation librarians who review and prioritize at-risk objects","Repository engineers who maintain storage, format identification, fixity checking, and rendering services","Collection managers who allocate preservation-review capacity","Researchers and other patrons whose ability to retrieve and render deposited objects is affected"],"affected_objective":"Prioritize limited preservation-review capacity while preserving access and avoiding unnecessary intervention in stable digital objects.","arm":"COMMON_P1","authority_safety":{"authorized_first_step":"Run a retrospective, read-only comparison using previously recorded repository observations; derived scores and event labels remain outside production systems.","decision_authority":"The designated collection manager retains prioritization authority; preservation librarians adjudicate object-level cases, and repository engineers authorize technical remediation.","excluded_actions":["No automatic deletion, migration, quarantine, access restriction, or rights-status change","No live reprioritization of the preservation queue during the first evidence step","No inference that a missing observation is a preservation failure","No replacement or alteration of authoritative metadata or audit logs"],"halt_rollback":"Stop if object identifiers cannot be reliably joined, event timestamps are ambiguous, sensitive-use metadata would be exposed, or derived classifications are mistaken for authoritative records. Remove the derived analysis dataset and retain the existing workflow unchanged."},"baseline":"A periodic review assigns each digital object to a small number of preservation-risk tiers using the latest available metadata. Objects within a tier are treated as equivalent until the next review, while exceptional failures may enter a separate incident queue without a documented relationship to the tier model.","candidate_id":"discrete_continuous_model_selection__library_information_science__COMMON_P1","causal_chain":["Tier-only representation converts gradual changes in renderability, dependency support, replication coverage, and time since validation into coarse categories.","Coarse tiers make objects near a cutoff receive different priorities despite similar observable conditions.","Periodic tier updates can also smooth over discrete events such as a confirmed fixity mismatch or failed render that changes the required response immediately.","An event-only representation would preserve those jumps but discard gradual accumulation that can support advance scheduling.","A hybrid object-version model retains a weekly continuous preservation-risk trajectory while recording consequential failures, recoveries, migrations, and dependency losses as timestamped events.","Explicit event boundaries and sampling rules allow reviewers to distinguish accumulated risk from a state-changing incident.","Comparing hybrid priorities with later access outcomes and expert adjudication reveals whether either smoothing or segmentation changes preservation decisions.","Only a validated representation is eligible for a separately authorized operational pilot."],"cell_id":"discrete_continuous_model_selection__library_information_science","consequence":"Preservation effort can be allocated inconsistently: gradual deterioration may remain hidden inside a tier, abrupt failures may wait for the next review, and arbitrary tier boundaries may separate objects with nearly identical evidence.","diversity_from_prior_proposals":"Not evaluated against prior proposals because runtime isolation prohibits inspecting them; this candidate is independently grounded in digital-preservation review rather than cross-cell comparison.","experiment_id":"eoa_inverse_innovation_exp13_second_slot_policy60_20260806","intervention":"Create a read-only hybrid preservation model at the object-version level. Sample continuous indicators weekly, including days since successful fixity verification, days since successful rendering, supported-renderer count, replication deficit, and unresolved dependency age. Separately record discrete events: a fixity mismatch confirmed by a repeat read, loss of the last validated renderer, failed retrieval, completed migration, verified restoration, or rights-status change. A confirmed event opens one event episode for that object version; repeated observations belong to the same episode until restoration, replacement, or documented acceptance closes it. Event episodes override score-only ordering by requiring librarian review, not automatic remediation. Outside open episodes, the continuous trajectory supports relative scheduling without bins. Recalculate assumptions when observation cadence, instrumentation, collection policy, or object composition changes.","mechanism_mapping":[{"counterfactual_removal":"Without the hybrid mechanism, either gradual accumulation disappears into incident labels or consequential jumps disappear into a smooth score.","mechanism_slug":"hybrid_discrete_continuous_model","role":"Combines continuous preservation-risk trajectories with discrete failures, recoveries, and repository interventions."},{"counterfactual_removal":"Without an explicit cadence, fast transitions may occur between observations and apparent smoothness may be an artifact of sparse checking.","mechanism_slug":"sampling_interval_choice","role":"Sets weekly indicator snapshots while retaining native timestamps for event-producing checks."},{"counterfactual_removal":"Without the audit, the model could appear coherent while failing to detect transitions at the timing and object granularity needed for review.","mechanism_slug":"transition_resolution_audit","role":"Tests whether event definitions and observation cadence preserve decision-relevant changes."},{"counterfactual_removal":"Without comparison to the existing tier representation, there is no direct test of whether discretization or smoothing changes prioritization errors.","mechanism_slug":"state_machine_vs_flow_model","role":"Compares tier states, continuous trajectories, and their hybrid against the same adjudicated cases."}],"nearest_rivals":["Periodic static preservation-risk tiers with manual exceptions","An incident-only queue driven by failed fixity or retrieval checks","A single continuous preservation-risk score without explicit event episodes","More frequent manual audits using the existing tier definitions"],"negative_tests":{"intervention_falsifier":"The intervention is falsified if, on held-out objects, the hybrid representation does not reduce either cutoff-sensitive priority disagreements or missed consequential events relative to the simpler tier-only and event-only models, or if reviewers cannot apply its boundary rules consistently.","problem_falsifier":"The inferred problem is falsified if preservation decisions are unchanged across tier-only, continuous-only, event-only, and hybrid representations, with no reviewed errors attributable to smoothing, cutoff placement, or observation cadence.","risks":["Continuous indicators may imply precision unsupported by repository measurements.","Objects with sparse monitoring may appear safer than heavily monitored objects.","Event definitions may double-count repeated symptoms or merge distinct failures.","A priority trajectory could be misread as an authoritative preservation or appraisal judgment.","More detailed monitoring may expose sensitive use, rights, or collection-status information.","The hybrid model may add review burden without changing decisions."],"strongest_counterevidence":"If later access failures and expert priority judgments are explained adequately by a small set of stable tiers, while timestamped events add no decision-relevant information and continuous indicators are too sparse or noisy, the added hybrid structure is unwarranted."},"next_evidence_step":"Select at most 200 object versions from one repository collection covering a fixed prior 12-month window. Using only existing fixity, rendering, retrieval, dependency, migration, and review records, have two preservation librarians independently adjudicate event episodes and pairwise review priority for a stratified sample enriched around existing tier boundaries. Compare tier-only, continuous-only, event-only, and hybrid representations on missed adjudicated events, boundary-sensitive priority reversals, reviewer agreement, and missing-data sensitivity. Do not change live queues or repository objects.","observable_state":"For each repository object version: timestamps and outcomes of fixity checks, retrieval attempts, and rendering tests; count of validated renderers or dependencies; replica count versus policy target; unresolved dependency age; migration and restoration timestamps; current rights-status record; existing risk tier; and subsequent documented access failure or successful use. Missingness and monitoring frequency are recorded as state rather than treated as safety.","prior_art_status":"UNSEARCHED","problem":"A digital repository uses coarse periodic risk tiers to schedule preservation review even though some preservation conditions drift gradually while others change through abrupt, consequential events. The representation can therefore create artificial priority cliffs and conceal failures between review cycles.","proposal_index":1,"remaining_contrastive_claim":"For preservation scheduling, the hybrid representation is preferable to tier-only, continuous-only, or event-only rivals only if reviewed decision errors occur both near category boundaries and around timestamped state-changing events, and if the explicit hybrid rules distinguish those cases reproducibly.","revision_record":{"claim_changes":["Initial version makes only a conditional contrastive claim and asserts no novelty, prevalence, demand, or effect size."],"conceptual_changes":["Initial candidate maps discrete–continuous model selection to digital-preservation review using continuous risk trajectories within discrete event episodes."],"evidence_changes":["No external or repository evidence was inspected; prior art remains unsearched."],"operational_changes":["Initial design limits the first step to a read-only retrospective comparison on at most 200 object versions from one collection and one 12-month window."],"parent_version":null,"progress_targets_addressed":["Concrete domain problem and decision need","Observable state and consequence","Structural and mechanism mappings","Causal chain and nearest-rival contrast","Authority boundaries, safeguards, falsifiers, risks, and bounded evidence step"]},"schema_version":1,"structural_mapping":[{"archetype_element":"Decision Need","domain_realization":"Schedule preservation review for digital object versions under limited staff capacity."},{"archetype_element":"Process Change Signature","domain_realization":"Renderability, dependency support, verification age, and replication coverage can drift gradually, while fixity failures, retrieval failures, migrations, and restorations occur as discrete events."},{"archetype_element":"Cost of False Smoothness","domain_realization":"Averaging or interpolating through a confirmed failure can delay review of an object whose preservation state changed abruptly."},{"archetype_element":"Cost of False Discreteness","domain_realization":"Risk tiers create cutoff-sensitive priority differences and hide meaningful variation within each tier."},{"archetype_element":"Granularity Choice","domain_realization":"Use object version as the unit, weekly observations for gradual indicators, and native timestamps for consequential events."},{"archetype_element":"Step Boundary","domain_realization":"An event episode opens on a confirmed state-changing observation and closes on verified restoration, replacement, or documented acceptance; repeated symptoms inside the episode are not new events."},{"archetype_element":"Continuity Assumption","domain_realization":"Continuous indicators may be interpolated only between valid weekly observations when no state-changing event or monitoring gap is recorded."},{"archetype_element":"Measurement Resolution","domain_realization":"Weekly derived snapshots retain underlying check timestamps and explicitly encode missing observations and cadence variation."},{"archetype_element":"Hybrid Boundary Rule","domain_realization":"An open event episode routes the object to librarian review; otherwise its continuous trajectory supports relative scheduling without conversion into risk bins."},{"archetype_element":"Transition Validation","domain_realization":"Compare inferred transitions and priorities with later access outcomes and independent librarian adjudication."},{"archetype_element":"Approximation Error Check","domain_realization":"Measure decision disagreements attributable to tier cutoffs, missed events, sparse observations, and interpolation assumptions."}],"title":"Hybrid Event–Trajectory Model for Digital-Preservation Review","version":0}