{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp05_complete_proposal_portfolio20_20260803","cell_id":"layer_decay_and_expiration_management__cognitive_science","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"cand_dependency_gated_scaffold_sunset_05","proposal_index":5,"version":0,"title":"Dependency-Gated Scaffold Sunset for Procedural Skill Training","problem":"An interactive tutor for a multistep fault-diagnosis task adds highlighting overlays, next-step prompts, checklists, worked examples, and error-specific hints as a learner progresses. Supports introduced at different stages can remain simultaneously active after their original purpose changes. Persistent scaffolds may crowd the task display, continue supplying steps that the learner is expected to retrieve, and make supported performance difficult to distinguish from independent competence. Abrupt removal, however, can expose unresolved dependencies, violate an accommodation, or erase the support configuration needed to interpret the learner's training history.","actors":["Learner practicing the procedural task","Instructor responsible for formative decisions","Instructional designer defining each scaffold's purpose","Tutor-system operator configuring support states","Accessibility or accommodation authority","Evaluator reconstructing the conditions under which performance occurred"],"observable_state":"For one training module, an inventory identifies every visible or requestable scaffold, when and why it was introduced, which task steps and later materials refer to it, its last validation, and whether it is active, review-due, on-demand, hidden-recoverable, preserved, or retired. Observable warning states include prompts that describe an earlier interface, multiple supports covering the same step, scaffolds with no withdrawal criterion, later instructions that reference a supposedly removable aid, and assessments whose recorded results omit the active support configuration. Matched supported and scaffold-reduced probes can show whether a candidate aid remains functionally load-bearing.","consequence":"Accumulated supports can consume attention and permit correct task completion through continued cue following rather than the unaided retrieval or strategy use the training claims to assess. Premature removal can instead produce avoidable failure, conceal an instructional dependency, or remove an authorized accommodation.","affected_objective":"Develop and measure independent procedural competence while keeping necessary support available, recoverable, and accurately represented in the training record.","intervention":"Give each instructional scaffold a lifecycle contract when introduced: identity, pedagogical purpose, covered task steps, prerequisite relationships, minimum exposure condition, review deadline, withdrawal evidence, accommodation status, and disposition path. Review expiry never removes a scaffold automatically. A dashboard detects supports whose content, interface references, or pedagogical role have become stale, while an age-weighted value score ranks candidates for review. Before withdrawal, the system checks references from later instructions, assessments, task steps, and accommodation plans and runs a bounded scaffold-reduced probe where appropriate. Approved scaffolds move gradually from always visible to learner-requestable, then hidden but recoverable, rather than disappearing in one step. A quarantine window permits immediate reinstatement if predefined rollback conditions occur. Historical support bundles are archived with performance records, tombstones identify successors, and sampled restore tests verify that an earlier training configuration can be reconstructed. Initial deployment is a shadow lifecycle review in a non-safety-critical module.","structural_mapping":[{"archetype_element":"Sequential deposits","domain_realization":"Highlights, prompts, checklists, worked examples, and corrective hints added during successive stages of procedural instruction."},{"archetype_element":"Changed active usefulness","domain_realization":"A scaffold useful during initial acquisition may become redundant, misleading after an interface change, or counterproductive to testing independent retrieval."},{"archetype_element":"Old layers remaining active by default","domain_realization":"Earlier supports remain visible as new supports are added because introduction is specified but withdrawal is not."},{"archetype_element":"Resource and interpretive pressure","domain_realization":"Multiple visible aids compete for learner attention and obscure which knowledge or strategy generated successful performance."},{"archetype_element":"Age and expiration rules","domain_realization":"Each scaffold receives a review lease and withdrawal criteria tied to training stage, validation age, task change, and demonstrated dependencies."},{"archetype_element":"Dependency check","domain_realization":"Later instructions, assessments, task steps, accommodations, and observed scaffold-reduced performance are checked before an aid is withdrawn."},{"archetype_element":"Differentiated disposition paths","domain_realization":"A support can remain visible, be renewed, become requestable, become hidden-recoverable, be revised, be preserved by exception, or be retired."},{"archetype_element":"Preservation exceptions","domain_realization":"Accessibility accommodations, safety-relevant guidance, disputed mastery judgments, and designated reconstruction needs override ordinary fading."},{"archetype_element":"Reversible cleanup","domain_realization":"Temporary withdrawal and quarantine allow immediate restoration when rollback criteria are met."},{"archetype_element":"Audit and reconstruction","domain_realization":"Archived support bundles and tombstones preserve the exact instructional conditions attached to earlier performance records."}],"mechanism_mapping":[{"mechanism_slug":"time_to_live_ttl_policy","role":"Assigns each scaffold a review lease at introduction, with expiry triggering reassessment or a supervised fading probe rather than automatic removal.","counterfactual_removal":"Without a creation-time review trigger, supports can persist indefinitely because the curriculum records when to add them but not when to reconsider them."},{"mechanism_slug":"stale_layer_detection_dashboard","role":"Inventories active supports and flags outdated interface references, superseded explanations, duplicated functions, missing owners, and absent withdrawal criteria.","counterfactual_removal":"Without semantic staleness detection, a recently introduced scaffold can remain active even after a task or explanation changes."},{"mechanism_slug":"age_weighted_value_score","role":"Ranks review candidates using validation age, learner access, unique instructional function, observed dependency, reconstruction value, display cost, and accommodation status.","counterfactual_removal":"Without ranking, designers face an unbounded scaffold inventory without a defensible review order; the score itself cannot authorize withdrawal."},{"mechanism_slug":"dependency_safe_delete_check","role":"Traces references from later materials, assessments, task steps, and accommodation plans and blocks retirement when any live dependency remains unresolved.","counterfactual_removal":"Without the gate, removing a seemingly redundant prompt could make a later instruction unintelligible or violate a required support dependency."},{"mechanism_slug":"lifecycle_storage_tiering_policy","role":"Implements a fading ladder from always visible to learner-requestable to hidden-recoverable, preserving the scaffold while reducing its default presence.","counterfactual_removal":"Without tiered access, designers must choose between leaving a support fully active and removing it entirely."},{"mechanism_slug":"retention_schedule","role":"Defines lifecycle rules for temporary acquisition aids, enduring reference aids, safety guidance, accessibility supports, and support bundles attached to assessment records.","counterfactual_removal":"Without class-level rules and exceptions, individual instructors can fade or preserve equivalent supports inconsistently."},{"mechanism_slug":"soft_delete_quarantine_window","role":"Places a withdrawn scaffold in a recoverable state and restores it if predefined error, help-request, or dependency conditions occur.","counterfactual_removal":"Without quarantine, a mistaken mastery judgment turns scaffold withdrawal into an unnecessarily abrupt and difficult-to-reverse change."},{"mechanism_slug":"tombstone_or_deletion_marker","role":"Records that a scaffold was retired, its rationale, affected task steps, and any replacement so historical references do not resolve to an unexplained absence.","counterfactual_removal":"Without a tombstone, later reviewers cannot distinguish deliberate fading from missing instructional content."},{"mechanism_slug":"archive_restore_test","role":"Reconstructs sampled historical training screens using archived support bundles and verifies that the supports attached to an earlier performance record remain renderable.","counterfactual_removal":"Without restoration tests, recorded scaffold configurations may become unusable after tutor interfaces or content formats change."}],"causal_chain":["Instructional designers add supports in response to new task stages, anticipated difficulties, and observed errors.","The supports persist as later aids are deposited because each has an introduction rule but no governed retirement path.","Learners perform within an increasingly layered cue environment, and successful completion can remain dependent on prompts that have outlived their intended acquisition stage.","Persistent aids consume attention and make it difficult to attribute performance to independent knowledge versus available cues.","Lifecycle contracts make every scaffold's purpose, age, dependencies, exception status, and withdrawal conditions inspectable.","Staleness detection and value ranking select bounded review candidates without changing learner access.","Reference checks and supervised scaffold-reduced probes identify supports that remain functionally or institutionally load-bearing.","Approved candidates descend through visible, requestable, and hidden-recoverable states, with quarantine enabling rollback.","Archived configurations preserve interpretability of earlier performance while the active training surface contains only explicitly justified supports."],"baseline":"Designers add hints and prompts as the module evolves, use fixed curriculum stages or instructor judgment to hide some aids, and retain screenshots or configuration files inconsistently. The system may adapt hint frequency, but individual supports lack explicit review leases, dependency gates, accommodation overrides, graduated retirement states, rollback windows, and tested historical reconstruction.","nearest_rivals":["A fixed schedule that removes all prompts after a specified training stage","Performance-threshold fading based on recent task accuracy","An adaptive hint system that varies help according to current errors","A minimalist redesign that removes redundant interface elements globally","Instructor-controlled toggles for manually showing or hiding supports"],"remaining_contrastive_claim":"The proposal's testable differentiator is lifecycle governance for each deposited scaffold: withdrawal depends on explicit pedagogical and institutional dependencies, proceeds through reversible access tiers, and preserves reconstructable support configurations. Fixed fading, adaptive hinting, or manual toggles can change current assistance, but do not by themselves manage stale scaffold accumulation, exceptions, rollback, and historical interpretability as one lifecycle.","authority_safety":{"decision_authority":"The instructional designer and responsible instructor approve ordinary lifecycle rules and withdrawal decisions. The accessibility authority controls accommodation exceptions, and the relevant safety authority controls any safety-critical guidance. Automated scores and probe results provide review evidence but have no independent removal authority.","authorized_first_step":"Inventory scaffolds in one non-safety-critical training module, assign provisional lifecycle states, trace documented dependencies, and simulate alternative visibility states in a sandbox without changing the production curriculum or learner records.","excluded_actions":["Automatic removal based only on age, usage frequency, or recent accuracy","Withdrawal of an authorized accessibility accommodation","Removal of safety-critical instructions or warnings","Changing summative assessment conditions without explicit approval and disclosure","Interpreting scaffold dependence as a diagnosis or stable learner deficit","Hard deletion of support configurations linked to recorded performance","Using a hidden scaffold state to deny a learner an authorized help path"],"halt_rollback":"Halt if a required accommodation or safety instruction is classified for retirement, a later task references a hidden scaffold, the sandbox cannot reproduce the original interface, or a supervised probe crosses its predefined stop condition. Restore the complete prior scaffold configuration and discard only the derived shadow states; source content and learner records remain unchanged."},"negative_tests":{"strongest_counterevidence":"If persistent performance limitations arise from missing instruction or inadequate practice rather than accumulated supports, fading scaffolds will expose the deficit without resolving it. A well-calibrated adaptive hint system may already provide assistance only when needed while preserving an uncluttered default display.","problem_falsifier":"The selected module contains no superseded, duplicated, unowned, or indefinitely active scaffolds; every support has a current purpose and withdrawal rule; supported and appropriately scaffold-reduced probes show no hidden dependency; and historical performance records already identify reconstructable support conditions.","intervention_falsifier":"Compared with a fixed-stage fading rule or the existing adaptive-hint baseline, lifecycle management does not clarify whether performance is independent, repeatedly restores supposedly retireable aids, causes missed task steps or unauthorized support loss, or requires review effort disproportionate to the bounded scaffold set.","risks":["Premature fading can convert an instructional design error into apparent learner failure.","Scaffold-reduced probes can underestimate competence if they also alter familiar task context.","Learners may avoid requesting a warm-tier aid because requesting help feels costly or stigmatizing.","Recent performance can create false confidence about mastery under different task conditions.","Age-weighted scoring can undervalue old but foundational reference aids.","Incomplete dependency maps can break later instructions or assessments.","Frequent state transitions can increase interface inconsistency and learner confusion.","Accommodation exceptions can be mishandled or exposed through lifecycle metadata.","Archived support bundles may contain sensitive learner-linked configuration information.","Preservation exceptions can accumulate until the active curriculum is again difficult to simplify."]},"next_evidence_step":"Choose one existing non-safety-critical procedural training module containing a bounded set of visible or requestable scaffolds. Build a read-only inventory, have the instructor and accessibility authority verify purposes and exceptions, and trace references from later steps and assessments. In an authorized formative sandbox, use matched task forms to compare the current support configuration, a fixed-stage fading rival, and the proposed visible-to-requestable-to-hidden lifecycle for a small reviewed subset. Record task-step completion, help requests, stop-condition activations, restoration events, learner explanations of strategy, reviewer disagreements, and whether an archived prior configuration can be rendered. Do not alter summative scores, production settings, accommodations, or source records.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 governs laboratory protocol artifacts so superseded files are not assembled into a current experiment; proposal 5 governs instructional supports presented during skill acquisition, and its causal concern is persistent cue dependence and obscured independent competence rather than artifact-version selection. Proposal 2 governs the historical evidence contributing to an adaptive estimate of a user's state; proposal 5 does not expire observations or change their inferential weight, but changes the lifecycle of prompts and aids available to the learner. Proposal 3 governs operational handoff cues that represent unresolved conditions across shifts; proposal 5 governs pedagogical scaffolds whose intended endpoint is graduated withdrawal after dependencies are resolved, not continued representation of current work. Proposal 4 governs executable production rules inside a computational cognitive model; proposal 5 leaves model code and theoretical mechanisms untouched and acts on the learner-facing instructional environment. The scaffold lifecycle is independently adoptable by a training designer without adopting any protocol registry, evidence-lease estimator, handoff board, or cognitive-model maintenance harness.","revision_record":{"parent_version":null,"progress_targets_addressed":["Initial complete specification for proposal index 5","Explicit material differentiation from sealed proposals 1 through 4","Causal preservation of accumulation, expiration, dependency checking, tiering, exceptions, rollback, and reconstruction","Bounded evidence and authority safeguards for learner-facing intervention"],"conceptual_changes":["Initial version; no parent proposal","Targets accumulated pedagogical scaffolds and the transition from supported to independent performance"],"operational_changes":["Initial version; defines scaffold lifecycle contracts, dependency-gated fading, visible-to-requestable-to-hidden tiers, quarantine restoration, and historical support bundles"],"evidence_changes":["Initial version; specifies a sandbox comparison with fixed-stage fading and existing support conditions plus restoration testing"],"claim_changes":["Initial version; limits the contrastive claim to governed, reversible retirement of individual instructional scaffolds","Makes no claim of novelty, prevalence, demand, or effect size"]}}