{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp05_complete_proposal_portfolio20_20260803","cell_id":"computability_boundary_mapping__veterinary_medicine","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"cbm-vet-bounded-behavior-assessment-synthesis-005","proposal_index":5,"version":0,"title":"Bounded Synthesis of Discriminating Animal-Behavior Assessments","problem":"A veterinary behavior or wildlife-rehabilitation team wants a universal planner that accepts any two executable animal-response models and always determines whether some finite adaptive assessment can distinguish them, then synthesizes such an assessment when one exists. The assessment may choose the next observation or low-impact stimulus from earlier responses, while the models may contain arbitrary computation, unbounded state, or nonterminating branches. Failed heuristic search cannot prove that no distinguishing assessment exists, and a successful plan on selected simulations cannot prove discrimination for every state and allowed outcome. Any assertion that unrestricted synthesis is undecidable must also match the exact model encoding, quantifiers, response semantics, and guarantee through a checked reduction.","actors":["Veterinary behaviorist defining the assessment question","Ethologist encoding candidate response models","Wildlife-rehabilitation veterinarian responsible for animal welfare","Computational modeler implementing the synthesizer","Independent reviewer checking the reduction and bounded algorithm","Facility staff who could later conduct an approved assessment"],"observable_state":"The proposed planner accepts executable response models without fixed state or runtime bounds; candidate assessments may branch indefinitely; randomized or heuristic searches stop without a protocol and report the models as indistinguishable; generated protocols are tested on selected initial states; and reports omit the action alphabet, observation partition, horizon, response bound, or model assumptions under which discrimination was evaluated.","consequence":"A search timeout may be mistaken for evidence that two behavioral explanations cannot be distinguished, while a plan that works only on sampled simulations may be treated as guaranteed. Either error can misdirect assessment design, expose an animal to procedures without a justified information objective, or attach unwarranted certainty to a rehabilitation or behavioral interpretation.","affected_objective":"Design finite, reviewable animal-behavior assessments with exact model-relative discrimination guarantees inside declared bounds, without representing failure inside or outside those bounds as a general scientific conclusion.","intervention":"Separate unrestricted adaptive-protocol synthesis from a bounded, decidable service. First formalize the unrestricted instance: two encoded response programs, allowed actions, observable outputs, admissible initial states and nondeterminism, and the requirement that one finite adaptive protocol force distinguishable observations on every covered execution. Attempt a source-to-target many-one reduction, such as encoding a source computation into whether one response model can ever diverge observably from a comparison model, and accept an impossibility conclusion only after independent checking of totality, computability, answer preservation, and scope. For operation, require a finite promise: finite model-state sets, finite action and observation alphabets, a response-step limit, a maximum protocol depth, and explicit allowed outcome transitions. Enumerate every adaptive action-observation decision tree within that fence and check it against every admitted state and outcome path. Return DISCRIMINATING-PROTOCOL-WITHIN-BOUND with a replayable coverage certificate when a tree succeeds, or NONE-WITHIN-BOUND after complete exhaustion. Return OUT-OF-MODEL for inputs that violate the finite contract and UNKNOWN if an implementation resource cap interrupts otherwise complete enumeration. Send any proposed live use through veterinary welfare review; the synthesizer itself authorizes no animal procedure. Record models, bounds, property, algorithm, certificate, result, complexity assessment, and recheck triggers in a versioned boundary decision record.","structural_mapping":[{"archetype_element":"Problem-Class Specification","domain_realization":"Given two animal-response models, determine whether an adaptive sequence of permitted actions and observations is guaranteed to distinguish them under the stated initial-state and outcome quantifiers."},{"archetype_element":"Instance Representation Contract","domain_realization":"Defines model states, action and observation alphabets, transition tables, initial-state sets, nondeterministic outcomes, response-step limit, protocol depth, and the exact distinguishability predicate."},{"archetype_element":"Quantifier and Scope Map","domain_realization":"Separates existence of one protocol from its required success across every covered pair of initial states and every allowed outcome path, and separates bounded synthesis from the unrestricted executable-model claim."},{"archetype_element":"Reduction Preservation Contract","domain_realization":"Requires any source computation to be translated computably into a model pair whose possession of a distinguishing protocol preserves the source answer in both directions."},{"archetype_element":"Promise-Problem Restriction","domain_realization":"The exact operational guarantee applies only when both models and the protocol space satisfy mechanically checked finiteness and response-bound promises."},{"archetype_element":"Termination Condition","domain_realization":"Enumeration terminates after visiting the finite set of protocol trees and covered executions determined by the declared alphabets, state sets, and depth."},{"archetype_element":"Constructive Procedure Witness","domain_realization":"The bounded synthesizer enumerates every admissible adaptive tree and produces either a successful tree with coverage evidence or an exhaustive negative certificate for the declared fence."},{"archetype_element":"Complexity Follow-On Gate","domain_realization":"State, outcome, and decision-tree growth are assessed separately after bounded decidability is established, before larger bounds are authorized."},{"archetype_element":"Fallback Solution Contract","domain_realization":"Interrupted searches yield UNKNOWN, contract violations yield OUT-OF-MODEL, and exhaustive negatives retain the label NONE-WITHIN-BOUND."},{"archetype_element":"Computability Guarantee Record","domain_realization":"A supersedable record binds each synthesis result to model versions, encodings, bounds, proof status, certificates, resource limits, and welfare-review status."}],"mechanism_mapping":[{"mechanism_slug":"many_one_reduction_proof","role":"Tests the unrestricted synthesis claim through a total computable map from a known source problem to a pair of executable response models, with existence of a distinguishing protocol preserving the source answer.","counterfactual_removal":"Without an exhibited answer-preserving map, planner timeouts or intuitive resemblance to program reachability could be misreported as an unrestricted impossibility proof."},{"mechanism_slug":"reduction_direction_checklist","role":"Checks that the reduction runs from the independently established source into the synthesis target and audits totality, computability, answer preservation, encoding, and quantifiers.","counterfactual_removal":"Without this gate, a reversed map or a proof covering only one outcome direction could wrongly define the computability boundary."},{"mechanism_slug":"promise_problem_restriction","role":"Makes the operational solver accountable only for model pairs satisfying enforceable finite-state, finite-alphabet, bounded-response, and bounded-depth promises.","counterfactual_removal":"Without checked promises, arbitrary or nonterminating model code could enter the bounded solver while its outputs retained an exact guarantee."},{"mechanism_slug":"bounded_domain_exhaustive_search","role":"Enumerates every adaptive protocol tree and every covered execution inside the finite fence, providing a complete two-sided answer within that fence.","counterfactual_removal":"Without exhaustive coverage, failure to find a protocol would remain merely heuristic and could not support NONE-WITHIN-BOUND."},{"mechanism_slug":"constructive_algorithm_and_correctness_proof","role":"Supplies the bounded synthesis algorithm with a termination argument based on finite enumeration and a correctness argument relating its certificates to the declared discrimination property.","counterfactual_removal":"Without a proved constructive procedure, finiteness alone would not show that the implementation enumerates all legitimate protocols and outcome paths."},{"mechanism_slug":"computational_complexity_analysis","role":"Measures how candidate trees and model-product states grow with depth, alphabet size, and nondeterminism after in-principle decidability has been established.","counterfactual_removal":"Without the complexity gate, teams could confuse a finite but unusable search space with an operationally feasible assessment-planning service."},{"mechanism_slug":"fallback_mode_router","role":"Preserves DISCRIMINATING-PROTOCOL-WITHIN-BOUND, NONE-WITHIN-BOUND, UNKNOWN, and OUT-OF-MODEL as distinct results and routes proposed protocols to welfare review.","counterfactual_removal":"Without governed routing, an interrupted search or bound-qualified negative could be converted into a general indistinguishability claim."},{"mechanism_slug":"computability_boundary_decision_record","role":"Records the unrestricted status, finite promise, synthesized tree, coverage certificate, complexity result, welfare constraints, and changes requiring re-analysis.","counterfactual_removal":"Without the record, altered models, observations, allowed actions, or bounds could silently inherit an obsolete discrimination guarantee."}],"causal_chain":["Arbitrary executable response models and unbounded adaptive protocols create a universal synthesis requirement whose computation model and quantifiers are initially hidden.","A formal model-pair and discrimination contract makes it possible to test the unrestricted requirement with a directionally correct reduction rather than inference from failed searches.","The operational service withdraws the unrestricted claim and admits only instances whose state, action, observation, outcome, response, and horizon spaces are mechanically finite.","Finiteness makes the complete adaptive protocol-tree space and the covered model-product executions enumerable.","Exhaustive evaluation can therefore construct a protocol that succeeds on every covered path or certify that none exists inside the exact fence.","Coverage certificates make successful plans replayable, while exhaustive-negative certificates retain their bounds and interrupted searches remain UNKNOWN.","Complexity analysis prevents the finite classification from being mistaken for practical feasibility as protocol depth or model size grows.","Veterinary welfare review evaluates whether an informationally valid protocol is ethically and clinically acceptable before any interaction with an animal."],"baseline":"A behavior specialist proposes several assessment sequences, simulates or mentally evaluates them against selected model states, and revises them heuristically. If no sequence succeeds before the planning budget expires, the models may be described as practically indistinguishable. This process can generate useful ideas but provides neither exhaustive coverage nor a valid negative result.","nearest_rivals":["Expert-authored behavioral assessment: incorporates biological and welfare judgment but does not exhaustively establish discrimination across every encoded state and outcome.","Random or heuristic protocol search: may find short protocols quickly but cannot certify nonexistence when search ends without success.","Bayesian active learning: selects observations expected to reduce uncertainty under priors and likelihoods, but expected information gain is different from guaranteed discrimination on every allowed path.","Passive observation extension: gathers more behavior without synthesizing contingent actions and offers no finite guarantee that the selected model pair will separate.","Verification of a fixed assessment protocol: can test one proposed tree but does not synthesize or exhaust all trees within the declared bound.","Unrestricted theorem-prover search: may produce a proof or plan for some models, but timeout or an open goal does not provide the bounded solver's complete negative certificate."],"remaining_contrastive_claim":"The proposal guarantees only that complete finite enumeration finds a discriminating adaptive assessment if one exists inside the declared model, action, observation, response, and depth bounds. It does not claim that the encoded hypotheses are biologically correct, that discrimination is possible outside the fence, that a synthesized protocol is acceptable for live use, or that decidability implies feasible search cost.","authority_safety":{"decision_authority":"The veterinary behaviorist defines the interpretation target, the computational reviewer approves the formal guarantee, and the veterinarian responsible for animal welfare has sole authority to approve, modify, or reject any live assessment.","authorized_first_step":"Implement and evaluate the bounded synthesizer solely on synthetic finite-state response models and simulated observations, with no live animals, clinical records, or facility procedures.","excluded_actions":["Presenting a synthesized protocol to animal-care staff as an authorized procedure","Exposing, restraining, dosing, startling, handling, or relocating an animal based on prototype output","Using the result alone for diagnosis, treatment, placement, release, breeding, or euthanasia decisions","Reporting NONE-WITHIN-BOUND as general observational equivalence or biological indistinguishability","Reporting UNKNOWN or OUT-OF-MODEL as a negative result","Changing model states, outcome rules, actions, observations, or bounds while retaining the previous certificate"],"halt_rollback":"Stop the prototype and invalidate affected results if an independently enumerated protocol is omitted, a returned coverage certificate fails replay, an allowed model path is absent from checking, a bound disappears from a result label, or output reaches live-animal planning without welfare authorization. Revert to simulation-only use until correction and independent review are complete."},"negative_tests":{"strongest_counterevidence":"A checkable constructive solver that always terminates and correctly decides and synthesizes distinguishing protocols for the full declared class of arbitrary executable response models would defeat the proposed unrestricted boundary and move the project to complexity assessment.","problem_falsifier":"The inferred computability problem is absent if the actual response models, action and observation alphabets, outcome sets, response time, and protocol depth are already finite and enforced; the task would then be a bounded synthesis and complexity problem from the outset.","intervention_falsifier":"The intervention fails if the enumerator omits an admissible adaptive tree or outcome path, the model-product checker implements a weaker discrimination predicate than the operational claim, or finite-contract membership can be bypassed.","risks":["Decision-tree enumeration may become infeasible at modest depth despite remaining finite.","The encoded observation partition may conceal behaviorally important distinctions or create artificial ones.","A protocol guaranteed in the model may fail because real animal responses fall outside the transition contract.","Binary discrimination between two models may ignore additional plausible explanations.","Information-maximizing actions may still be stressful, unsafe, or clinically inappropriate.","Users may remove bound labels when communicating negative results.","A complete search over a flawed formalization can produce confident but scientifically irrelevant certificates."]},"next_evidence_step":"Define two pairs of synthetic finite-state response models using no more than four states per model, three permitted actions, three observations, two nondeterministic outcomes per transition, and protocol depth three. Construct one pair with a known one-step distinguishing protocol and one pair with no distinguishing protocol inside the bound. Exhaustively enumerate all adaptive trees, independently brute-force their covered state-outcome paths, replay positive certificates, and verify the negative enumeration count. Add one deliberately out-of-model executable simulator and one forced resource interruption to test OUT-OF-MODEL and UNKNOWN routing. A second reviewer should inspect the reduction obligations and the bounded algorithm's termination and completeness arguments. This evidence may assess only the synthetic implementation and guarantee labels, not biological validity, animal-welfare acceptability, or practical effect.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 verifies whether a programmable veterinary care policy avoids prohibited actions by restricting policies to a finite-state language and model-checking a conservative patient abstraction. Proposal 5 instead synthesizes an adaptive assessment that forces two response models to produce distinguishable observations; it constructs prospective information-gathering protocols rather than certifying treatment-policy safety. Proposal 2 retrospectively searches open-ended animal-behavior simulators for one execution reproducing an observed trace, using dovetailing and a one-sided witness-or-UNKNOWN contract. Proposal 5 deliberately accepts only finite promised models and exhaustively gives a two-sided answer about prospective adaptive protocol existence within a horizon; it neither fits parameters nor matches a past trace. Proposal 3 determines whether legacy and replacement laboratory pipelines terminate and produce equivalent reports, using proof certificates and counterexamples to govern migration. Proposal 5 seeks action-conditioned observational separation rather than program equivalence, and its output is a bounded decision tree subject to welfare review rather than a software-release verdict. Proposal 4 admits recurring herd-surveillance workflows through producer-supplied termination certificates followed by complexity review. Proposal 5 checks no submitted termination proof and schedules no workflow; its causal engine is finite exhaustive synthesis across protocol trees and paired model executions. Its decision object, adopter, output artifact, intervention, safeguards, and causal path are therefore independently distinct from all four earlier proposals.","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}