{"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":"task_state_contract_for_substitutable_interaction_modalities","proposal_index":1,"version":0,"title":"Task-State Contract for Substitutable Maintenance-Reporting Interfaces","problem":"A field-maintenance organization offers the same incident-reporting workflow through desktop, touch-tablet, voice, and screen-reader-oriented interfaces. Each interface independently encodes navigation, validation, confirmation, retry, and submission behavior in its widget or dialogue structure. Consequently, changing a widget tree or adding a modality can alter which actions are reachable, when errors appear, whether dependent answers are invalidated, or whether a retry creates a duplicate report, even though every surface is nominally the same workflow.","actors":["Field technicians, including technicians using assistive technology","Interaction and accessibility designers","Desktop, tablet, and voice-interface developers","Maintenance-report service owner","Quality and usability evaluators"],"observable_state":"For an identical scripted reporting task, modality-specific traces can be reduced to an abstract session state containing collected facts, unresolved validation issues, currently available semantic actions, explicit-confirmation status, and submission outcome. The problem is observable when two interfaces presented with equivalent semantic action sequences reach different abstract states, expose different required actions, produce different error classes, or create different submission outcomes.","consequence":"Users can omit required information, become trapped on one modality, receive inconsistent recovery guidance, or submit duplicate or unintended reports; interface teams must also coordinate broadly whenever presentation internals change because the intended cross-modality behavior is not independently testable.","affected_objective":"Preserve task correctness, recoverability, and equivalent user agency across interaction modalities while allowing each modality's presentation and input representation to evolve independently.","intervention":"Define a representation-independent Maintenance Report Session contract before modifying any production interface. Model the abstract session as report facts, dependency status, validation issues, confirmation state, and commit status; expose semantic operations such as start, set fact, remove fact, attach evidence, request review, confirm submission, retry submission, undo an uncommitted change, and cancel. Specify preconditions, postconditions, error classes, side-effect limits, and sequence laws, including: submission is unavailable while required issues remain; changing a parent answer invalidates dependent answers; observation and navigation do not mutate report facts; cancellation never commits; and repeated submission with the same intent token commits at most one report. Each visual, touch, voice, or assistive-technology adapter may use different widgets, dialogue turns, focus rules, and local caches, but must map its observable interaction trace to this abstract state and pass the same black-box conformance oracle. Presentation-specific information may be exposed only through named, non-semantic extension points that cannot change the shared task transitions.","structural_mapping":[{"archetype_element":"Abstract behavioral surface","domain_realization":"The Maintenance Report Session's semantic actions, abstract task state, validation outcomes, confirmation state, and commit outcome, independent of screens, widgets, focus order, or spoken prompts."},{"archetype_element":"Hidden representation","domain_realization":"Each modality's component tree, dialogue graph, focus-management strategy, local cache, navigation layout, and input-event encoding."},{"archetype_element":"Behavioral laws and invariants","domain_realization":"No commit with unresolved required issues; dependent facts are invalidated when their premise changes; read-only actions do not mutate facts; cancellation does not commit; and a submission intent is committed at most once."},{"archetype_element":"Error and side-effect semantics","domain_realization":"The contract distinguishes invalid user action, incomplete report, unavailable attachment, and transient submission failure, and specifies whether state remains unchanged and whether retry is safe."},{"archetype_element":"Abstraction mapping","domain_realization":"A modality adapter maps widget events, gestures, keystrokes, or dialogue turns into semantic operations and maps its internal state back to the contract's abstract session state."},{"archetype_element":"Conformance oracle","domain_realization":"One parameterized black-box suite drives semantic task sequences through every adapter and compares only contract-level states, action availability, error classes, and commit records."},{"archetype_element":"Substitutability rule","domain_realization":"An interface adapter is eligible to represent the workflow only if it passes the shared suite, exposes every required semantic action, introduces no unauthorized task transition, and confines modality-specific behavior to declared extension points."},{"archetype_element":"Leakage control","domain_realization":"An audit checks whether widget identifiers, screen order, exact error wording, focus side effects, dialogue turn counts, timing, or decoded tokens have become undeclared dependencies."}],"mechanism_mapping":[{"mechanism_slug":"abstract_data_type_specification","role":"Defines the abstract report-session values and operations, plus an invariant for each modality implementation and a mapping from its concrete UI state to the shared task state.","counterfactual_removal":"Without the abstract state and mapping, teams could compare operation names but could not determine whether different screens or dialogue states mean the same task state."},{"mechanism_slug":"design_by_contract_clause","role":"Assigns preconditions, postconditions, unchanged-state guarantees, and error classes to actions such as review, confirm, retry, undo, and cancel.","counterfactual_removal":"Without these clauses, edge cases would remain modality-local conventions and equivalent signatures could still yield incompatible recovery or submission behavior."},{"mechanism_slug":"opaque_type_or_module_boundary","role":"Prevents modality code from mutating raw report-session storage or submission flags except through semantic operations.","counterfactual_removal":"Without the boundary, a surface could bypass validation or confirmation by writing internal fields directly, making its representation an unofficial workflow API."},{"mechanism_slug":"black_box_contract_test_suite","role":"Runs the same contract-derived task sequences against every modality adapter and gates substitutability on contract-level observations.","counterfactual_removal":"Without a shared oracle, each modality could pass its own tests while still disagreeing with the others on reachable actions, errors, or state transitions."},{"mechanism_slug":"property_based_conformance_test","role":"Generates action sequences involving edits, dependency changes, cancellation, failure, and retry to test invariants beyond hand-authored paths.","counterfactual_removal":"Without generated sequences, conformance evidence would be limited to anticipated journeys and could miss state corruption produced by unusual but legal action orderings."},{"mechanism_slug":"representation_leakage_probe","role":"Identifies observable but unpromised details—such as screen order, exact prompt wording, timing, focus changes, or token structure—that clients, tests, or automation might treat as stable semantics.","counterfactual_removal":"Without the probe, adapters might pass promised-behavior tests while accidental presentation details silently harden into dependencies."},{"mechanism_slug":"abstraction_barrier_code_review","role":"Requires reviewers to classify every new cross-boundary dependency as either shared task semantics, a declared modality extension, or prohibited reach-through.","counterfactual_removal":"Without human review, mechanically legal but semantically questionable dependencies could expand the contract or weaken the boundary without an explicit decision."}],"causal_chain":["Modality-specific widget and dialogue structures currently carry task semantics implicitly.","A shared abstract session model separates task meaning from presentation and input representation.","Contract clauses fix allowed operations, state transitions, invariants, errors, and side-effect limits at that abstract level.","Opaque access forces each modality adapter to translate through the semantic operations instead of mutating workflow internals.","The shared black-box and property-based oracle exposes adapters that omit actions, add transitions, mishandle retries, or reach different abstract outcomes.","Leakage review prevents convenient presentation details from becoming substitute semantic contracts.","Adapters that satisfy the same acceptance rule can change their internal interaction representation without requiring clients or the report service to depend on that representation."],"baseline":"Maintain separate modality specifications and end-to-end tests, using the incumbent desktop sequence and backend request shape as informal references. Accessibility review then evaluates each surface individually, while cross-modality differences are reconciled case by case.","nearest_rivals":["A shared design system that standardizes components, labels, and visual states but does not define modality-independent task transitions or retry semantics.","A responsive or cross-platform UI framework that reuses implementation code but can still propagate presentation-specific assumptions and does not establish behavioral substitutability.","Backend API standardization that fixes request and response shapes while leaving user-reachable actions, validation timing, confirmation, undo, and recovery behavior unspecified.","Per-modality usability and accessibility evaluation that can reveal interaction failures but does not itself provide a reusable acceptance oracle for replacement implementations."],"remaining_contrastive_claim":"The proposal's distinguishing commitment is to make abstract user-task state and its transition laws—not shared widgets, source code, screens, or backend payloads—the acceptance boundary for every interaction representation. It remains useful only if that contract permits materially different modality designs while preserving the same task-level obligations.","authority_safety":{"decision_authority":"The maintenance-report service owner and interaction-system owner may jointly authorize a sandbox contract and prototype evaluation; accessibility specialists and field representatives review whether the proposed abstract operations omit necessary user agency.","authorized_first_step":"Model one existing report journey in an isolated prototype, write contract-derived traces before implementing adapters, and run them against two deliberately different non-production interfaces using synthetic reports.","excluded_actions":["Replacing or routing any production interface","Writing prototype submissions to the live maintenance system","Collecting employee or incident data","Recruiting or recording participants without the organization's consent and research-review procedures","Treating conformance as accessibility certification or as evidence of usability","Removing modality-specific accommodations merely because they are absent from the shared minimum contract"],"halt_rollback":"Stop if the model cannot represent a required accommodation without exposing presentation internals, if a prototype can reach live records, or if a seeded invariant violation passes the oracle. Disable the sandbox adapters, discard synthetic session data, and revert the prototype boundary while revising or rejecting the contract."},"negative_tests":{"strongest_counterevidence":"The strongest counterevidence would be that safe and usable voice, touch, and assistive-technology interactions require fundamentally different task states or user commitments, so mapping them to one contract either removes necessary agency or freezes presentation details into a lowest-common-denominator workflow.","problem_falsifier":"The inferred problem is falsified if matched traces show that the existing modalities already preserve the same task states, reachable semantic actions, error classes, and submission outcomes, and observed failures instead arise from device reliability, content comprehension, training, or backend defects unrelated to representation coupling.","intervention_falsifier":"The intervention is falsified if independently built adapters pass the contract suite yet disagree on a declared task-level obligation, or if the suite can distinguish implementations only by encoding widget order, exact wording, focus sequence, or other representation details.","risks":["An underspecified contract may certify divergent or unsafe behavior.","An over-specified contract may freeze current screen logic and obstruct legitimate modality-specific accommodation.","A shared minimum may erase actions needed by users of one modality.","Generated tests may cover legal sequences poorly or encode assumptions taken from the incumbent interface.","Opaque boundaries may impede diagnosis unless limited, privacy-safe introspection is explicitly designed.","Teams may mistake behavioral conformance for evidence that an interface is understandable, efficient, accessible, or acceptable to users."]},"next_evidence_step":"Conduct a bounded sandbox feasibility test on one report type: independently state the abstract model and twelve representative task traces, implement one simple form adapter and one dialogue-style adapter over synthetic data, seed adapters that violate dependency invalidation and retry idempotence, and ask an accessibility reviewer to identify missing semantic actions. Continue only if the oracle rejects the seeded violations, both valid adapters map to the same abstract outcomes without shared presentation structure, and no required accommodation must bypass the contract.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Not applicable: proposal_index 1 has no earlier sealed proposal in this session.","revision_record":{"parent_version":null,"progress_targets_addressed":["Initial complete proposal"],"conceptual_changes":["Established the user-task session, rather than a widget set or backend endpoint, as the abstract component.","Defined cross-modality substitutability at the level of user-reachable semantic actions and task-state laws."],"operational_changes":["Specified a sandbox-only first step, joint authority, explicit exclusions, and rollback conditions.","Added concrete invariants, adapter acceptance criteria, leakage channels, and seeded negative controls."],"evidence_changes":["Set prior-art status to unsearched.","Bound initial evidence to synthetic traces, two non-production adapters, mutation detection, and accessibility review."],"claim_changes":["Made no claim of novelty, prevalence, demand, or effect size.","Limited the contrastive claim to the proposal's task-state acceptance boundary and stated the condition under which it would remain useful."]}}