{"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-behavior-simulator-witness-search-002","proposal_index":2,"version":0,"title":"Witness-Only Matching for Open-Ended Animal-Behavior Simulators","problem":"A computational ethology laboratory wants a universal fitting service that accepts any executable animal-behavior simulator and a finite observed behavior trace, then always terminates with YES or NO on whether some encoded parameter set, initial state, and random seed makes the simulator halt and reproduce that trace. Because admitted simulators may contain unrestricted loops, recursion, and generated submodels, some candidate runs may never terminate. Serial optimization, finite testing, and elapsed-time limits cannot establish that no matching execution exists, while an impossibility claim is justified only if a checked reduction matches the exact simulator representation and existential guarantee.","actors":["Veterinary behaviorists interpreting observed behavior","Ethologists designing behavioral models and experiments","Computational modelers supplying executable simulators","Research software engineers operating the fitting service","Independent computability and proof reviewer","Animal-welfare veterinarian overseeing any later intervention derived from research"],"observable_state":"The service accepts simulator programs without a proved runtime bound; candidate parameters and seeds form an open-ended countable space; one nonterminating candidate can block later candidates in a serial search; exhausted search budgets are reported as no fit; and a successful trace match may be displayed without the parameter encoding and execution needed to reproduce it.","consequence":"A timeout can be mistaken for evidence that an observed behavior is incompatible with a model family, altering model selection or experimental planning without a valid negative result. Conversely, an unchecked match can be mistaken for causal or clinical validation even though it establishes only that one encoded execution reproduces the finite trace.","affected_objective":"Support reproducible comparison of executable animal-behavior models while preserving the distinction among a witnessed match, absence within a finite search box, unresolved existence, and scientific validation of the model.","intervention":"Replace the promised universal Boolean fitter with a one-sided witness protocol. Specify the instance as a finite target trace, an executable simulator, a decidable encoding of candidate parameters, initial states and seeds, and an exact matching predicate. Test the unrestricted decision claim by mapping a source computation to a simulator that emits the target trace exactly when the source halts, and require independent checking of the reduction before classifying the full problem. For operational search, enumerate all well-formed candidate encodings and dovetail their simulator executions so a looping candidate cannot starve later ones. Return FOUND-WITNESS only with a replayable encoding and finite execution certificate that an independent checker validates. At a declared step or time budget, return UNKNOWN rather than NO. Optionally, when users declare a finite candidate set and per-run step bound, exhaust it and return NO-WITHIN-BOUND, with the bounds attached to the result. Record the output contract, fairness schedule, computation model, assumptions, and recheck triggers in a versioned computability-boundary record. No result is labeled model correctness, causal explanation, diagnosis, or treatment guidance.","structural_mapping":[{"archetype_element":"Problem-Class Specification","domain_realization":"Pairs an arbitrary encoded partial simulator with a finite observed behavior trace and asks whether at least one encoded parameter-state-seed tuple halts with exactly that trace."},{"archetype_element":"Instance Representation Contract","domain_realization":"Defines parseable encodings for the simulator, finite target trace, candidate parameters, initial conditions, seeds, output alphabet, and exact trace-equivalence rule."},{"archetype_element":"Quantifier and Scope Map","domain_realization":"Makes explicit that the unrestricted claim quantifies existentially over an unbounded candidate space but requires one terminating Boolean procedure for every admitted simulator and target trace."},{"archetype_element":"One-Sided Recognition Contract","domain_realization":"A finite successful execution certifies existence of a match; failure to find such an execution supplies no unrestricted negative certificate."},{"archetype_element":"Unknown and Nontermination Policy","domain_realization":"Looping candidates remain active within a fair dovetailing schedule, while an operational resource limit produces UNKNOWN and never an unrestricted NO."},{"archetype_element":"Termination Condition","domain_realization":"The service always stops at a declared operational budget, returning a checked witness if found and UNKNOWN otherwise; a separate finite-box mode terminates by exhaustive finiteness."},{"archetype_element":"Fallback Solution Contract","domain_realization":"The unrestricted Boolean request is weakened to witness production, explicit unknown, or a bound-qualified negative result without changing those labels downstream."},{"archetype_element":"Computability Guarantee Record","domain_realization":"Versions the simulator model, encodings, reduction evidence, enumeration and fairness rules, resource budget, certificate checker, result vocabulary, and re-analysis triggers."}],"mechanism_mapping":[{"mechanism_slug":"halting_problem_reduction","role":"Tests the total Boolean requirement by constructing a simulator that ignores its candidate input and emits the requested behavior trace if and only if an arbitrary source computation halts.","counterfactual_removal":"Without the reduction, nontermination observed during fitting could be mistaken for proof that no universal decider exists, or an informal analogy could be applied beyond the encoded simulator class."},{"mechanism_slug":"reduction_direction_checklist","role":"Confirms that the construction maps the known-undecidable source into the simulator-matching target and checks total construction, computability, answer preservation, representation, and quantifiers.","counterfactual_removal":"Without this gate, a reversed, partial, or representation-mismatched construction could wrongly determine the shipped guarantee."},{"mechanism_slug":"enumeration_and_dovetailing","role":"Fairly interleaves all encoded candidate executions so each receives unbounded progress in the ideal recognizer and no single looping simulation blocks a later finite witness.","counterfactual_removal":"Without fair interleaving, the service could fail to find an existing match merely because an earlier candidate never terminates."},{"mechanism_slug":"semi_decision_with_explicit_unknown","role":"Turns the ideal one-sided recognizer into an always-terminating service: it returns a checked FOUND-WITNESS on success and UNKNOWN when the declared operational budget expires.","counterfactual_removal":"Without an explicit UNKNOWN state, budget exhaustion would likely be rendered as an unsupported negative answer."},{"mechanism_slug":"bounded_domain_exhaustive_search","role":"Provides an optional, independently labeled mode that checks every candidate and execution step inside a user-declared finite box and can certify NO-WITHIN-BOUND.","counterfactual_removal":"Without this bounded mode, users needing a complete answer for a genuinely finite experiment would receive only the weaker unrestricted recognizer contract."},{"mechanism_slug":"proof_checking","role":"Independently checks both the unrestricted reduction argument and each finite execution certificate returned as a witness.","counterfactual_removal":"Without independent checking, the boundary could rest on a defective proof and successful matches could depend only on the search engine's assertion."},{"mechanism_slug":"computability_boundary_decision_record","role":"Preserves the chosen one-sided guarantee, computation model, result meanings, search schedule, finite-bound semantics, residual uncertainty, and triggers for reclassification.","counterfactual_removal":"Without the record, later changes to simulator expressiveness or output labels could silently recreate the unsupported universal claim."}],"causal_chain":["Unrestricted executable simulators allow candidate executions that may never halt, while the requested service must decide match existence for every simulator and trace.","A precise simulator encoding and existential matching predicate expose the class-wide claim rather than conflating it with the outcome of one fitting run.","A checked source-to-target reduction can classify the unrestricted Boolean requirement under the declared computation model without treating timeouts as impossibility evidence.","Because a finite matching execution is a checkable witness, the positive side remains recognizable even when a total exact decision procedure is unavailable.","Fair dovetailing prevents any nonterminating candidate from starving every later candidate and therefore preserves completeness of the ideal positive recognizer.","A resource cap makes the operational service terminate, while the restricted output alphabet forces an unresolved search to remain UNKNOWN.","Finite-box exhaustion supplies a complete negative only inside attached parameter and step bounds, preventing it from inheriting the unrestricted claim.","Versioned guarantees and veterinary research oversight prevent trace reproduction from being promoted into causal, diagnostic, welfare, or treatment authority."],"baseline":"A conventional fitting pipeline tries parameter and seed combinations serially or through an optimizer, terminates individual simulations after an implementation timeout, and reports either a best match or no fit after the overall budget. This can be useful as a heuristic search but does not preserve fairness across nonterminating candidates or distinguish failure within the budget from nonexistence over the full encoded class.","nearest_rivals":["Ordinary parameter optimization: may locate useful fits quickly but offers no completeness guarantee when objectives are discontinuous, candidates are unbounded, or simulator calls do not terminate.","Approximate Bayesian or likelihood-free fitting: quantifies support under selected priors, summaries, and sampling assumptions but does not decide exact existential matching for every executable simulator.","Serial enumeration with per-run timeouts: always advances eventually but can miss a witness whose execution needs more than the arbitrary individual timeout and therefore is not a complete positive recognizer.","Finite grid search: can certify absence only for the enumerated grid and execution bound; it cannot support an unrestricted negative result.","Manual comparison by a veterinary behaviorist: adds biological interpretation but is not a declared uniform procedure for resolving every encoded simulator instance.","Restricting all simulators to a total finite-state language: could recover two-sided decidability but would be a different trade, sacrificing the proposal's ability to accept open-ended research simulators."],"remaining_contrastive_claim":"The proposal offers a complete positive recognizer in the ideal computation model and an honest terminating operational approximation whose strongest unrestricted result is a replayable existence witness. Its claim is neither that every match will be found within a practical budget nor that trace reproduction validates the biological mechanism. Bound-qualified negative answers remain confined to their finite boxes.","authority_safety":{"decision_authority":"The research principal investigator governs model-comparison use, an independent reviewer approves the computability classification and certificate contract, and an attending or welfare veterinarian retains authority over any live-animal observation, intervention, diagnosis, or treatment.","authorized_first_step":"Implement an offline prototype over synthetic target traces and toy simulators, including terminating matches, nonterminating candidates, malformed encodings, and match-free finite boxes; verify fairness, certificate replay, result labels, and the proposed reduction without involving animals or clinical records.","excluded_actions":["Changing an animal's environment, handling, medication, diet, or treatment based on a fitter result","Presenting FOUND-WITNESS as proof that the simulator is biologically correct or causally explanatory","Presenting UNKNOWN as no match or model rejection","Generalizing NO-WITHIN-BOUND beyond its attached candidate and execution limits","Silently dropping candidates from the fair schedule","Executing untrusted simulator code outside an isolated resource-controlled environment"],"halt_rollback":"Stop the prototype and invalidate its output records if a claimed witness cannot be independently replayed, a well-formed candidate is starved by the scheduler, UNKNOWN is converted to a negative verdict, bounds disappear from a finite-box result, or simulator isolation fails. Revert to research-only manual analysis until the defect is corrected and independently reviewed."},"negative_tests":{"strongest_counterevidence":"A checkable constructive procedure that always terminates and correctly decides exact match existence for every simulator and trace in the full declared unrestricted encoding would defeat the proposed one-sided classification and move the project to complexity and implementation assessment.","problem_falsifier":"The computability problem is absent if every admitted simulator is already proved total, the candidate parameter-state-seed set is finite and effectively enumerable, outputs are bounded, and the requested result concerns only that finite class; exhaustive search would then decide the question in principle.","intervention_falsifier":"The intervention fails if candidate encodings cannot be enumerated effectively, the schedule is not fair, matching certificates are not independently decidable, or downstream users treat UNKNOWN as evidence against the model.","risks":["Dovetailing may allocate too little practical computation to deep but scientifically relevant candidates.","A search budget may yield mostly UNKNOWN results while remaining formally honest.","Exact trace equality may reward incidental reproduction and omit biologically important variation.","A valid witness can be overinterpreted as model identification despite multiple incompatible simulators producing the same finite trace.","Finite-box negative results may lose their bounds when copied into publications or model registries.","Running arbitrary simulator code can consume resources or escape isolation unless execution controls are independently tested.","The reduction may formalize a target different from the scientific question actually intended."]},"next_evidence_step":"Construct a sealed offline suite of at most twenty toy simulators over a finite behavior alphabet: include immediate matches, matches reachable only after long execution, several permanently looping candidate branches, malformed instances, and finite boxes with exhaustively known absence. Predeclare a dovetail schedule and global step budget. Test that every returned FOUND-WITNESS replays exactly, no early loop starves a later witnessed match in unbounded schedule simulation, budget exhaustion returns only UNKNOWN, and finite-box absence retains its bounds. Separately have a second reviewer check the proposed halting-to-matching construction step by step. This evidence can assess protocol correctness only for the synthetic suite and proof statement; it cannot establish biological validity, practical yield, or effects on animals.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 concerns pre-deployment safety and termination verification for programmable veterinary care policies and intervenes by excluding unrestricted programs, enforcing a finite-state clinical-policy fragment, conservatively abstracting patient transitions, and model-checking for prohibited actions. This proposal concerns existential trace matching for open-ended animal-behavior research simulators. It deliberately retains unrestricted partial simulators, weakens the demanded two-sided decision to fair witness recognition with explicit UNKNOWN, and uses replay certificates plus optional finite-box exhaustion. Its actors, decision object, output contract, failure consequence, intervention, and causal path are distinct, and a laboratory can adopt it without adopting the clinical policy verifier.","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}