{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp06_four_proposal_generalization60_20260803","cell_id":"representation_independent_interface_contract__human_computer_interaction","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"redesign_stable_usability_evidence_contract","proposal_index":2,"version":0,"title":"Redesign-Stable Contract for Usability Evidence","problem":"A human-computer interaction team measures use of a customer-support console through telemetry tied to DOM selectors, component names, screen routes, and current widget events. When the console is redesigned, equivalent user intentions produce different records, while renamed or reorganized controls appear as behavioral change. Researchers must either abandon longitudinal comparisons or manually reinterpret each release, and analytics code becomes coupled to presentation internals.","actors":["Customer-support agents whose interactions generate evidence","Usability researchers","Interaction designers","Frontend and instrumentation engineers","Product analysts","Privacy and data-governance reviewers"],"observable_state":"For a synthetic task performed on the incumbent console and a redesigned prototype, the raw event streams differ because their widget trees and navigation differ. The problem is observable when those streams cannot be mapped deterministically to the same contract-level evidence state—task started, semantic action attempted, outcome class, recovery transition, milestone reached, and terminal status—or when a presentation-only change alters a supposedly longitudinal measure.","consequence":"The organization cannot tell whether an apparent change in completion, recovery, or abandonment reflects user behavior, a changed task, or merely changed instrumentation. Historical dashboards require release-specific patches, and collecting extra raw interface detail to preserve interpretability can increase privacy exposure.","affected_objective":"Maintain interpretable, privacy-bounded usability evidence across interface redesigns without freezing the interface's screens, controls, routes, or implementation technology.","intervention":"Create an abstract Usability Evidence Session contract that defines evidence by task meaning rather than interface events. Its public operations include begin session, declare task version, record semantic action attempt, record outcome class, record recovery transition, mark milestone, and close as completed, abandoned, interrupted, or invalid. The contract specifies state transitions, required associations, error semantics, permitted metadata, and side-effect limits. It excludes raw entered content, screenshots, DOM paths, exact pointer coordinates, and stable personal identifiers. A release-specific adapter translates local clicks, keystrokes, commands, or navigation events into the abstract evidence operations; its component tree, routing structure, event names, and buffering strategy remain hidden. Every adapter must pass the same black-box and metamorphic oracle before its records may be joined to the longitudinal evidence series. Changes to task meaning require an explicit contract version rather than a silent adapter rewrite.","structural_mapping":[{"archetype_element":"Abstract behavioral surface","domain_realization":"The operations and states of an Usability Evidence Session: session start, task-version declaration, semantic attempts, outcome classes, recovery transitions, milestones, and terminal status."},{"archetype_element":"Hidden representation","domain_realization":"DOM selectors, component identifiers, route names, widget events, gesture coordinates, instrumentation buffers, and release-specific event pipelines."},{"archetype_element":"Behavioral laws and invariants","domain_realization":"Every outcome refers to a preceding semantic attempt; each session has one declared task version and at most one terminal state; events after closure are rejected; milestone order must respect the declared task model; and excluded content cannot enter permitted metadata."},{"archetype_element":"Error and side-effect semantics","domain_realization":"Unknown semantic actions, outcomes without attempts, duplicate closure, and disallowed metadata produce defined rejection classes without mutating the accepted evidence state."},{"archetype_element":"Abstraction mapping","domain_realization":"Each interface release supplies an adapter mapping its concrete event traces to abstract task-evidence transitions, with unmapped events classified as presentation-only, intentionally ignored, or contract gaps."},{"archetype_element":"Conformance oracle","domain_realization":"A shared suite feeds synthetic interface traces to each adapter and checks only the resulting abstract evidence state, rejection classes, and privacy-bounded output."},{"archetype_element":"Substitutability rule","domain_realization":"A new release adapter may contribute to the same longitudinal series only if it conforms for unchanged task semantics; a semantic task change must declare a new contract version and cannot be disguised as adapter substitution."},{"archetype_element":"Representation leakage control","domain_realization":"A leakage audit rejects records or downstream queries that depend on widget identifiers, screen order, exact labels, coordinate regions, raw text, or other unpromised details of the current interface."}],"mechanism_mapping":[{"mechanism_slug":"abstract_data_type_specification","role":"Models a usability evidence session as abstract values and transitions, then requires each release-specific trace representation to map to that model while preserving its invariants.","counterfactual_removal":"Without the abstract model and mapping, event-name consistency could be mistaken for semantic consistency, and researchers could not determine whether two release traces represent the same evidence."},{"mechanism_slug":"interface_definition_language","role":"Expresses the permitted evidence operations, typed outcome classes, task-version field, and privacy-bounded metadata as a machine-readable boundary from which instrumentation bindings can be generated.","counterfactual_removal":"Without a compiled boundary, release teams could add local fields or event variants that silently leak implementation structure or sensitive data into the shared evidence stream."},{"mechanism_slug":"design_by_contract_clause","role":"Defines preconditions and postconditions for attempt, outcome, milestone, and closure operations, including unchanged-state behavior after malformed or prohibited input.","counterfactual_removal":"Without these clauses, adapters could satisfy the same schema while differing on orphan outcomes, duplicate closure, invalid milestone order, or rejected metadata."},{"mechanism_slug":"black_box_contract_test_suite","role":"Applies identical synthetic journeys to every adapter and gates admission to the longitudinal series on equivalent abstract evidence and correct rejection behavior.","counterfactual_removal":"Without a reusable oracle, each release would validate its instrumentation against itself, allowing release-specific interpretations to enter supposedly comparable records."},{"mechanism_slug":"metamorphic_behavior_test","role":"Checks relations such as relabeling a control, changing layout, or replacing a click with an equivalent keyboard command must leave the abstract evidence unchanged when task meaning is held constant.","counterfactual_removal":"Without representation-changing tests, the suite could cover known journeys yet fail to detect that harmless presentation changes alter the recorded usability evidence."},{"mechanism_slug":"representation_leakage_probe","role":"Compares every emitted field and downstream dependency with the promised evidence contract, identifying both presentation coupling and prohibited raw-content leakage.","counterfactual_removal":"Without the probe, unpromised identifiers or sensitive interface details could become useful analytic dependencies even while the formal operations continued to conform."},{"mechanism_slug":"semantic_versioning_and_deprecation_gate","role":"Separates compatible adapter changes from changes to task meaning and requires a new version plus migration record when evidence semantics change.","counterfactual_removal":"Without version governance, teams could preserve a field name while redefining its meaning, making historical records appear comparable when they are not."},{"mechanism_slug":"abstraction_barrier_code_review","role":"Requires reviewers to classify each instrumentation dependency as contract-level task evidence, release-local representation, or prohibited collection.","counterfactual_removal":"Without intent-sensitive review, mechanically valid additions could expand measurement scope or bind analysis to the current interface without an explicit decision."}],"causal_chain":["Current telemetry treats presentation events as if they were direct observations of task behavior.","An abstract evidence-session model separates the meaning of an observed task transition from the interface event used to infer it.","Release adapters translate concrete interface traces through that model while the schema blocks direct access to presentation internals and excluded content.","Contract clauses and shared conformance tests expose orphan events, inconsistent transitions, silent semantic reinterpretation, and privacy-boundary violations.","Metamorphic tests vary presentation while holding task meaning constant, revealing adapters whose records change with representation.","Versioning prevents a genuine task change from masquerading as an interchangeable instrumentation implementation.","Accepted adapters can feed one evidence model without requiring the interface design itself to remain fixed."],"baseline":"Instrument each release with locally named UI events, repair dashboards after redesigns, and use screenshots, session replay, or analyst judgment to interpret discontinuities. Historical queries depend on release-specific mappings that are maintained separately from the interface specification.","nearest_rivals":["A shared analytics event-naming convention, which can standardize labels without specifying task-state laws, invalid transitions, or semantic equivalence across redesigns.","Session replay, which preserves rich presentation traces for later interpretation but remains tied to the recorded interface and can collect more user information than the proposed evidence boundary permits.","A data-warehouse semantic layer, which can normalize stored columns after collection but does not constrain what interface adapters observe or prove that differently generated records have the same task meaning.","Repeated controlled usability studies using fixed benchmark tasks, which can provide direct evidence at selected releases but do not establish a substitutability rule for continuous instrumentation."],"remaining_contrastive_claim":"The proposal makes a privacy-bounded task-evidence state machine—not event names, UI recordings, dashboard columns, or a fixed benchmark interface—the contract governing whether instrumentation from a redesigned interface may be treated as a substitute. It does not establish that any resulting metric is a valid measure of an unobserved psychological construct; it only makes the provenance and behavioral meaning declared by the contract testable.","authority_safety":{"decision_authority":"The usability-research owner and instrumentation owner may jointly authorize the sandbox model; the privacy or data-governance authority must approve the permitted metadata before any production consideration.","authorized_first_step":"Use synthetic traces from one incumbent workflow and one non-production redesign to draft the evidence contract, build isolated adapters, and test representation changes without enabling new collection from workers.","excluded_actions":["Deploying the adapter or adding production telemetry","Collecting screenshots, raw text, pointer trails, or stable personal identifiers","Joining sandbox records to employee performance data","Reclassifying an existing task change as compatible solely to preserve a dashboard series","Using contract conformance as proof that a metric measures confusion, workload, satisfaction, or usability","Retaining synthetic or replay-derived data beyond the bounded evaluation"],"halt_rollback":"Stop if semantic mapping requires prohibited content, if reviewers cannot distinguish presentation-only events from task evidence, if the sandbox connects to production analytics, or if a seeded leak or transition violation passes the oracle. Disable the adapters, delete the sandbox records, and retain only the reviewed contract questions needed to decide whether revision is justified."},"negative_tests":{"strongest_counterevidence":"The strongest counterevidence would show that the redesign changes the task's meaning so extensively that no representation-independent evidence state preserves what researchers need, or that constructs of interest depend essentially on presentation-specific phenomena that the abstract contract removes.","problem_falsifier":"The inferred problem is falsified if existing instrumentation already maps releases to an explicit task-state model with stable transition semantics and privacy limits, and observed longitudinal discontinuities correspond to declared task changes rather than presentation coupling.","intervention_falsifier":"The intervention is falsified if independently written adapters pass the suite while mapping the same contract-declared synthetic journey to different evidence states, or if making the adapters agree requires encoding DOM structure, screen sequence, exact labels, or prohibited user content in the contract.","risks":["The abstract categories may conceal interaction difficulties that are inherently presentation-specific.","Analysts may infer comparability beyond the exact task semantics covered by the contract.","Adapter authors may tune mappings to preserve preferred trends rather than represent declared behavior.","Overly broad semantic actions may erase meaningful recovery paths, while overly narrow actions may freeze the current design.","Privacy exclusions may reduce diagnostic detail needed to interpret an anomalous record.","A green conformance suite may be mistaken for validation of a usability construct or metric.","Version pressure may encourage teams to classify genuine task changes as compatible."]},"next_evidence_step":"Run a bounded offline mutation study using one task with synthetic data: specify ten contract-level journeys before inspecting adapter output, implement independent adapters for an incumbent interface trace and a structurally different prototype trace, apply presentation-only transformations such as relabeling, layout reordering, and equivalent keyboard activation, and seed orphan outcomes, duplicate closure, semantic relabeling, and prohibited metadata. Proceed only if valid transformations preserve the abstract evidence, every seeded violation is rejected, and reviewers can identify genuine task changes that require a version boundary.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 governs live maintenance-report task execution across visual, touch, voice, and assistive interfaces so users retain equivalent actions, validation, recovery, and submission behavior. This proposal instead governs the evidence produced about interaction across successive redesigns so researchers can distinguish task semantics from telemetry representation. It is independently adoptable for a single interface and does not replace, extend, or require Proposal 1's cross-modality workflow contract; its actors, protected object, intervention boundary, causal path, consequences, and acceptance evidence are measurement-centered rather than execution-centered.","revision_record":{"parent_version":null,"progress_targets_addressed":["Initial complete proposal at proposal_index 2","Material differentiation from sealed proposal 1"],"conceptual_changes":["Established the usability-evidence session as the abstract component.","Separated representation-stable behavioral provenance from claims about latent usability constructs.","Made task-semantic version changes distinct from substitutable instrumentation changes."],"operational_changes":["Specified a machine-readable, privacy-bounded event boundary and release-specific adapters.","Added metamorphic presentation transformations, seeded privacy violations, joint authority, exclusions, and rollback conditions."],"evidence_changes":["Bound the first evidence step to synthetic offline traces and two independent adapters.","Included negative controls for transition, semantic-version, and prohibited-metadata failures.","Kept prior-art status unsearched."],"claim_changes":["Made no claim of novelty, prevalence, demand, metric validity, or effect size.","Limited the claim to testable preservation of contract-declared evidence semantics across interface representations."]}}