{"schema_version":1,"assessment_id":"eoa_inverse_innovation_exp03_opportunity320_20260801","source_experiment_id":"eoa_inverse_innovation_exp03_full320_20260801","cell_id":"computability_boundary_mapping__speech_language_pathology","archetype_slug":"computability_boundary_mapping","domain_slug":"speech_language_pathology","title":"Enforceable Verification Boundaries for Adaptive AAC Dialogue Policies","opportunity_summary":"Evaluate whether an AAC policy language actually exceeds total-exact verification boundaries, then enforce proof-backed restricted fragments and explicit UNKNOWN, out-of-scope, and tool-failure states. The opportunity could prevent unresolved verification or timeouts from becoming clinical clearance, but the sealed candidate does not establish that a deployed unrestricted policy class or universal verification claim exists.","adopter_authorizer":"Clinical AAC governance lead jointly with the affected user's authorized care team and software safety owner, while preserving the AAC user's control over personal communication choices.","scores":{"meaningful_impact":{"score":4,"rationale":"If the stated failure occurs, erroneous clearance, unnecessary withholding, or loss of timely communication could materially affect safety and communicative agency. The score is limited because the sealed candidate supplies no evidence about occurrence or prevalence."},"stakeholder_pull":{"score":2,"rationale":"AAC users, clinicians, caregivers, governance staff, and software owners have plausible interests in trustworthy verification labels, but the candidate does not show that any deployed service has the claimed unrestricted verifier, observed demand, or willingness to adopt this workflow."},"incremental_advantage":{"score":4,"rationale":"Relative to retries, informal timeout interpretation, and larger timeouts, enforced policy classes plus exact, abstract, UNKNOWN, and human-review routing directly improve the meaning and governance of results. Advantage remains conditional on the target language and modeled properties matching real runtime behavior."},"distinctiveness_plausibility":{"score":2,"rationale":"The mechanism composition is coherent, but prior-art status is explicitly UNSEARCHED and no comparison with AAC assurance, dialogue-policy verification, clinical governance, or safety-case practices is present."},"technical_implementability":{"score":3,"rationale":"A bounded sandbox using synthetic policies, one formal property, and one restricted fragment appears implementable, while semantics extraction, checked reductions, exact-fragment enforcement, sound abstraction, and runtime integration may require substantial specialized work."},"adoption_authority_feasibility":{"score":3,"rationale":"Relevant clinical, care-team, user, and software-safety authority is specifically identified, with rollback controls. Joint authority and the user's communication rights create necessary coordination and constrain unilateral adoption."},"evidence_readiness":{"score":3,"rationale":"The candidate provides observable states, separate problem and intervention falsifiers, negative tests, and a safe retrospective pilot. Readiness is limited by absent deployed-language specifications, corpus definitions, expected verdicts, and evidence that the problem exists."},"safety_net_benefit":{"score":5,"rationale":"Explicit UNKNOWN, out-of-scope, and tool-failure states, prohibition on timeout-based clearance, continued human review, halt criteria, and rollback directly provide a strong safety net even when exact verification is unavailable."},"scalability":{"score":3,"rationale":"The classification-and-routing pattern could be reused across policies, but each language version, runtime semantics, property, fragment, abstraction, and clinical-use model may require separate formal validation and governance."}},"score_confidence":"MODERATE","costs":{"first_evidence":{"band_2026_usd":"10K_TO_50K","scope":"Obtain and inspect one versioned policy-language/runtime specification; formalize one safety property and bounded history model; construct a small synthetic and de-identified corpus; compare the boundary workflow with current timeout/result labeling in a retrospective sandbox.","confidence":"LOW","assumptions":["A cooperative AAC service or software owner supplies usable specifications and de-identified structures.","The first study covers one language version, one property, and one candidate fragment.","No live device or clinical decision is controlled.","Formal-methods and clinical-review labor dominate the resource requirement."]},"initial_deployment_startup":{"band_2026_usd":"50K_TO_250K","scope":"Engineer fragment-membership enforcement, result-state handling, audit records, reviewer guidance, safety review, and integration into one AAC policy release workflow after successful evidence.","confidence":"LOW","assumptions":["One existing verifier and one governance workflow are integrated.","No major verifier or AAC runtime rewrite is required.","Security, privacy, accessibility, and clinical governance review are included.","The exact fragment proves usable for a material subset of policies."]},"operational_launch":{"band_2026_usd":"250K_TO_1M","scope":"Validate representative policies and runtime behavior, train reviewers, implement monitoring and rollback, complete organizational approvals, and launch controlled use at one service or product organization.","confidence":"LOW","assumptions":["Launch remains limited to one organization or product line.","Multiple policy authors and clinical reviewers require coordination.","Independent checking of exactness and abstraction claims is included.","Live-policy blocking remains subject to approved governance and user-access safeguards."]},"annual_recurring":{"band_2026_usd":"50K_TO_250K","scope":"Maintain semantics and proof artifacts, reclassify guarantees after language or runtime changes, monitor UNKNOWN consumption and reviewer comprehension, audit releases, and provide specialist support.","confidence":"LOW","assumptions":["One organization or product line is supported.","Language and runtime changes occur periodically but not continuously.","Human review remains necessary for unresolved and out-of-scope policies.","No exact volume of policies, sites, or incidents is assumed."]}},"research_burden":"HIGH","earliest_credible_horizon":"3_TO_12_MONTHS","pipeline_gates":{"recognizable_externally_supportable_problem":{"status":"UNCERTAIN","reason":"The candidate specifies observable timeouts, ambiguous result states, and serious consequences, but provides no external evidence that an unrestricted programmable AAC policy class, universal total-exact claim, or associated operational failure exists."},"identifiable_adopter_or_authorizer":{"status":"YES","reason":"The clinical AAC governance lead, affected user's authorized care team, software safety owner, and user's retained communication authority are explicitly identified."},"distinct_testable_incremental_claim":{"status":"YES","reason":"The proposal can be compared with retrying timeouts or using a larger timeout by measuring classification correctness, stated-bound termination, and reviewer discrimination of exact, UNKNOWN, out-of-scope, and tool-failure labels."},"bounded_next_evidence_step":{"status":"YES","reason":"A retrospective sandbox limited to one property, one bounded history model, one candidate fragment, and synthetic or de-identified policy structures is explicitly authorized and has defined falsifiers."},"no_unresolved_safety_or_authority_stop":{"status":"YES","reason":"The candidate excludes live-policy blocking based solely on UNKNOWN, automatic intent replacement, unsupported impossibility claims, and timeout laundering, while specifying joint authority, halt criteria, and rollback to human release review."},"implementation_cost_scope_and_range":{"status":"UNCERTAIN","reason":"A bounded pilot and deployment components can be described, but actual resource requirements depend on unavailable language specifications, verifier architecture, policy volume, compliance obligations, and partner access."}},"blocking_evidence":["Whether any deployed AAC service accepts an unrestricted programmable policy class and makes the stated universal total-exact verification claim.","A versioned syntax, semantics, runtime, and interaction-horizon specification establishing whether accepted policies are finite, bounded, or sufficiently expressive for a valid reduction.","Operational definitions for one clinically material property, the bounded history model, corpus construction, expected verdicts, and exact termination bound.","Evidence that a candidate fragment includes common communication needs and that membership can be mechanically enforced.","Evidence that reviewers and downstream systems reliably distinguish UNKNOWN, out-of-scope, alarm, exact verdict, and tool failure without treating unresolved states as clearance.","A comparison with relevant prior approaches before any distinctiveness or originality claim."],"next_evidence_step":"With one AAC software or service partner, preregister a retrospective sandbox study covering one versioned policy language, one clinically reviewed safety property, one bounded history model, and one mechanically enforceable candidate fragment. Compare the proposed result taxonomy with the existing timeout workflow on a labeled synthetic and de-identified corpus; stop or reject the approach if an in-fragment verdict is wrong, exact analysis exceeds its stated bound, the real policy class is already finite and bounded, or reviewers cannot reliably distinguish UNKNOWN from clearance.","research_questions":["Does a deployed unrestricted programmable AAC policy class and universal total-exact verification claim actually exist?","Are accepted policies and interaction histories already finite, mechanically bounded, and exhaustively enumerable?","Do the policy syntax and runtime semantics support the computation required by any proposed impossibility reduction?","Which clinically material safety property can be modeled without omitting consequential runtime behavior?","Can a useful restricted fragment be decided exactly while preserving common communication needs and user agency?","Can fragment membership and guarantee reclassification be enforced across language and runtime updates?","How often do sound abstractions or bounded analyses return UNKNOWN, and what access delays result?","Can reviewers and downstream systems reliably interpret all result labels without laundering uncertainty into approval?","What materially similar assurance, verification, and governance mechanisms already exist?","What partner coordination, integration, compliance, and recurring maintenance resources are required?"] ,"recommendation":"VALIDATE_PROBLEM_FIRST","uncertainty_constraints":["Closed-book assessment provides no evidence of problem prevalence, stakeholder demand, realized impact, market size, or deployment volume.","The unrestricted impossibility branch is conditional and must not be stated as a domain fact without a checked semantics-preserving proof.","Natural conversational openness alone does not establish computational undecidability.","A successful bounded formal check would support only the modeled property, language version, and runtime behavior, not overall AAC safety or communicative usability.","Cost bands are resource-equivalent planning ranges based on assumed scope, not quotations or point estimates.","Joint clinical, software-safety, care-team, and user authority may constrain adoption timing and acceptable automation."],"closed_book_prior_art_boundary":"Prior-art status is UNSEARCHED. This assessment makes no claim that the boundary workflow, restricted-fragment design, result taxonomy, or governance composition is novel or uncommon; distinctiveness requires external comparison with AAC software assurance, dialogue-policy verification, clinical decision-support governance, formal verification, and safety-case practices."}