{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp05_complete_proposal_portfolio20_20260803","cell_id":"computability_boundary_mapping__film_media_production","arm":"COMPLETE_PROPOSAL_PORTFOLIO","candidate_id":"cbm_fmp_constraint_cut_feasibility_03","proposal_index":3,"version":0,"title":"Honest Feasibility Search for Constraint-Driven Cuts","problem":"An editorial-planning system accepts footage, generated elements, and executable rules describing duration, shot ordering, dialogue coverage, adjacency restrictions, placement obligations, and delivery requirements. It is asked to terminate on every brief and either synthesize a finite cut satisfying every rule or prove that no such cut exists. The unrestricted specification permits edit decision lists of arbitrary finite length and rule scripts with general computation, but candidate bounds, evaluator termination, environmental inputs, and the meaning of infeasible are unstated. Successful searches demonstrate individual witnesses, while timeouts are being treated as evidence that no acceptable cut exists.","actors":["editor","trailer or versioning producer","editorial pipeline engineer","media-asset librarian","localization coordinator","standards reviewer","delivery approver"],"observable_state":"For the same encoded brief, the planner may quickly return a conforming cut, continue searching until a resource limit, or stop after exploring a configured subset. Its interface nevertheless exposes only CUT FOUND or NO CUT. Search depth, maximum cut length, allowed edit points, evaluator status, and candidate-space coverage are absent from the decision record, and an elapsed-time limit is not distinguishable from an exhaustive infeasibility result.","consequence":"A feasible deliverable can be abandoned because its witness lies beyond an implicit search bound, while an actually infeasible bounded brief can consume repeated search effort without a termination argument. Producers may also treat machine-declared infeasibility as a judgment about creative possibility even when the system evaluated only a formal subset of cuts or failed to finish evaluating a rule.","affected_objective":"Determine when a machine-stated editorial brief has a constructible cut, return checkable witnesses when available, and reserve proven infeasibility for candidate spaces that were completely and validly examined.","intervention":"Replace the universal cut-or-infeasible solver with a scope-enforced feasibility service. First specify the encoding of assets, edit decisions, generated elements, rule scripts, evaluator outputs, and candidate cuts. Test the unrestricted total-exact requirement with a reviewed reduction: for an arbitrary program and input, define candidate cut C_n for each natural number n and a total rule evaluator that accepts C_n exactly when the source program halts within n steps. A satisfying cut exists exactly when the source computation halts, so a total solver that always distinguishes existence from nonexistence for the unrestricted class would decide halting. If that construction and its assumptions pass review, ship two explicitly different modes. In bounded mode, enforce a finite asset set, finite edit-point grid, fixed track and event limits, maximum cut length, finite parameter domains, and a total rule language; enumerate the entire resulting candidate space and return FOUND-WITNESS or PROVEN-INFEASIBLE-WITHIN-BOUND with a coverage certificate. In open search mode, fairly enumerate finite cut encodings and dovetail their rule evaluations; return FOUND-WITNESS when a checkable accepting cut appears, but at the operational resource limit return UNKNOWN rather than NO CUT. Route briefs by mechanically checked scope, record every bound and assumption, and leave aesthetic acceptance and release authority with the designated editorial decision-maker.","structural_mapping":[{"archetype_element":"Problem-Class Specification","domain_realization":"Define the unrestricted decision problem as whether at least one finite encoded cut satisfies every machine-stated rule, rather than whether the planner found one during a particular run."},{"archetype_element":"Instance Representation Contract","domain_realization":"Version asset identifiers, available ranges, edit-point granularity, track structure, generated-element encoding, rule syntax, parameter domains, and the canonical encoding of a candidate cut."},{"archetype_element":"Computation Model Contract","domain_realization":"State whether rule evaluators are total bounded predicates or arbitrary executable scripts and identify metadata services, generators, reviewers, and other external information sources."},{"archetype_element":"Quantifier and Scope Map","domain_realization":"Separate the existence of one witness, failure to find a witness within a budget, nonexistence inside a finite declared space, and nonexistence across every unrestricted finite cut."},{"archetype_element":"Computability Impossibility Certificate","domain_realization":"Document the mapping from a program-input pair to the family of cuts C_n and the bounded-step acceptance rule, including the argument that a satisfying cut exists exactly when the source program halts."},{"archetype_element":"Decidable Subclass Map","domain_realization":"Define the combinations of finite assets, discrete edit points, bounded cut structure, finite parameters, and total rule constructs for which complete enumeration is guaranteed to terminate."},{"archetype_element":"Constructive Procedure Witness","domain_realization":"Provide the bounded enumerator, its coverage invariant, its strictly decreasing count of unexplored candidates, and a checker for returned cuts and infeasibility certificates."},{"archetype_element":"One-Sided Recognition Contract","domain_realization":"For unrestricted but effectively enumerable cuts and rule computations, guarantee eventual discovery of an accepting witness when one exists under a fair dovetailing schedule, without promising a negative result."},{"archetype_element":"Unknown and Nontermination Policy","domain_realization":"Keep UNKNOWN distinct from a checked witness, proven bounded infeasibility, out-of-scope syntax, evaluator nontermination, malformed input, and planner failure."},{"archetype_element":"Fallback Solution Contract","domain_realization":"Allow bounded reformulation, open witness search, or human editorial work while prohibiting any fallback from inheriting an unrestricted proof-of-infeasibility claim."},{"archetype_element":"Computability Guarantee Record","domain_realization":"Store the brief and rule versions, candidate-space contract, enumerator or dovetailing mode, bounds, coverage certificate or witness, result label, and checker version."},{"archetype_element":"Recheck Trigger","domain_realization":"Reclassify a brief when assets, edit-point granularity, rule language, generators, track limits, duration bounds, parameter domains, or external services change."},{"archetype_element":"Complexity Follow-On Gate","domain_realization":"After bounded decidability is established, separately assess state-space growth, pruning correctness, memory, and expected completion cost before operational use."}],"mechanism_mapping":[{"mechanism_slug":"halting_problem_reduction","role":"Tests the unrestricted existence-and-synthesis requirement by making satisfying cut C_n exist exactly when an arbitrary source computation halts within some finite number of steps.","counterfactual_removal":"Repeated search failure could be mistaken for an impossibility result, or further engineering could pursue a total unrestricted solver without examining whether the requested distinction is computable."},{"mechanism_slug":"many_one_reduction_proof","role":"Requires a total computable transformation from each source program-input pair to one editorial-feasibility instance whose yes answer preserves source halting.","counterfactual_removal":"The boundary claim would rest on a loose halting-problem analogy rather than an explicit target instance and answer-preserving map."},{"mechanism_slug":"reduction_direction_checklist","role":"Checks the source-to-target direction, validity of every C_n encoding, totality of its bounded-step evaluator, existential biconditional, and computation-model assumptions.","counterfactual_removal":"A reversed reduction or an invalid candidate encoding could incorrectly retire the unrestricted solver requirement."},{"mechanism_slug":"language_fragment_restriction","role":"Restricts bounded-mode rules to an enforceable total language over finite editorial objects and parameters.","counterfactual_removal":"A supposedly finite search could stall forever while evaluating one arbitrary rule script and therefore fail to support its termination guarantee."},{"mechanism_slug":"bounded_domain_exhaustive_search","role":"Enumerates every cut in the declared finite candidate space and supports both a checked witness and a complete within-bound infeasibility conclusion.","counterfactual_removal":"The service could not distinguish exhaustive failure from failure to explore a sufficiently large subset."},{"mechanism_slug":"constructive_algorithm_and_correctness_proof","role":"Provides the bounded enumeration procedure together with termination and coverage arguments showing that PROVEN-INFEASIBLE-WITHIN-BOUND follows only after every valid candidate is rejected.","counterfactual_removal":"A bounded solver might terminate yet omit legal candidates through an incomplete generator or unsound pruning rule."},{"mechanism_slug":"enumeration_and_dovetailing","role":"Fairly interleaves candidate generation and rule evaluations so a divergent evaluation cannot permanently starve a later accepting cut.","counterfactual_removal":"Sequential open search could become trapped on one nonterminating candidate even when a readily checkable witness exists elsewhere."},{"mechanism_slug":"semi_decision_with_explicit_unknown","role":"Wraps open witness search in an operational resource bound and returns UNKNOWN, never infeasible, when no accepting witness has yet been confirmed.","counterfactual_removal":"The production interface would again convert elapsed time into a false negative existence verdict."},{"mechanism_slug":"proof_checking","role":"Independently checks returned cuts against the frozen rules and validates bounded coverage certificates and any pruning justifications.","counterfactual_removal":"Users would have to trust the search engine's assertion that a cut satisfies the brief or that an enormous bounded space was fully eliminated."},{"mechanism_slug":"fallback_mode_router","role":"Dispatches each brief to exact bounded search, open recognizer mode, reformulation, or human escalation and attaches the earned guarantee to the result.","counterfactual_removal":"Witnesses, bounded infeasibility, incomplete search, and evaluator failure could collapse into an operationally misleading yes-or-no answer."},{"mechanism_slug":"computability_boundary_decision_record","role":"Versions the solvability classification, operational modes, public result language, supporting reduction, bounds, and conditions requiring reassessment.","counterfactual_removal":"Later changes to the rule language or candidate generator could silently invalidate the distinction between bounded proof and open search."}],"causal_chain":["An unrestricted editorial brief can define an effectively enumerable family of candidate cuts indexed by arbitrary natural numbers.","A total rule can test candidate C_n by simulating an arbitrary source program for only n steps, accepting exactly when halting has occurred within that bound.","Some satisfying cut therefore exists exactly when the source program eventually halts.","Conditional on independent validation of this construction, an always-terminating exact existence solver for the unrestricted brief class would decide the halting problem.","Enforcing finite cut structure and a total rule fragment removes the unbounded candidate and evaluation behavior used by the reduction.","Complete enumeration then supports exact witness or infeasibility results only within the declared finite boundary.","For open briefs, fair dovetailing preserves the ability to discover an existing checkable cut without claiming that unsuccessful search proves absence.","Explicit routing, UNKNOWN behavior, certificates, and versioned records prevent bounded or one-sided evidence from being promoted to an unrestricted infeasibility claim."],"baseline":"Run a heuristic or constraint-driven cut generator until it finds an acceptable candidate or exhausts its configured budget, then report CUT FOUND or NO CUT without separating complete finite enumeration, incomplete search, evaluator nontermination, and failure of the tool itself.","nearest_rivals":["A fixed-size constraint solver: it can decide a precisely finite formulation but does not justify an answer for arbitrary cut lengths, executable rules, or unbounded generated elements.","A heuristic, evolutionary, or generate-and-rank editor: it may produce useful candidate cuts but failure to produce one cannot certify infeasibility.","Manual editorial construction: it can incorporate tacit aesthetic judgment and reformulate the brief, but it is not a uniform terminating decision procedure for every machine-encoded feasibility instance.","Repeatedly increasing the search timeout: this can expose later witnesses but supplies no point at which absence becomes a proven unrestricted negative.","Sampling candidate cuts and validating each one: an accepted sample proves feasibility, while any finite collection of rejected samples leaves unrestricted nonexistence unresolved.","A proof-oriented solver used without a scope contract: a checked bounded certificate can be trustworthy, but proof search failure or timeout remains neither a proof nor a disproof."],"remaining_contrastive_claim":"The proposal's remaining contrast is an existential-synthesis architecture that deliberately gives the positive and negative sides different operational treatments: open search may confirm feasibility through a cut witness, while infeasibility is issued only after a finite, enforceable candidate space is completely covered. This is a scoped design contrast, not a claim of novelty, prevalence, demand, or effect size.","authority_safety":{"decision_authority":"The editor and designated editorial producer retain authority to judge narrative and aesthetic acceptability. A delivery or standards approver remains authoritative for obligations within their remit. The planner may certify only satisfaction of the frozen machine-stated rules and may not redefine the creative brief.","authorized_first_step":"On synthetic assets and non-production briefs, the pipeline engineer may formalize one candidate encoding and bounded rule fragment, submit the reduction for independent review, implement the enumerator and dovetailing scheduler, and test result labeling. This step may not modify an active project, reserve production resources, or issue a delivery decision.","excluded_actions":["Reject, cancel, or rescope a production because open search returns UNKNOWN or reaches a timeout.","Describe PROVEN-INFEASIBLE-WITHIN-BOUND as proof that no conceivable edit exists.","Publish or deliver a FOUND-WITNESS without the responsible editor and required standards reviewers approving it.","Treat satisfaction of machine-stated metadata rules as evidence of aesthetic, narrative, legal, or cultural acceptability.","Silently enlarge or shrink the asset set, edit grid, rule fragment, track limit, duration bound, or generator capabilities after issuing a certificate.","Allow an unbounded or external rule evaluator into exact bounded mode.","Represent the reduction-based boundary as established before its mapping and assumptions pass independent review."],"halt_rollback":"Halt the pilot if bounded mode labels a brief infeasible despite an independently verified in-bound witness, accepts a cut that violates a frozen machine rule, omits a valid candidate through enumeration or pruning, loses fairness in the open scheduler, or allows UNKNOWN to be rendered as NO CUT downstream. Disable machine infeasibility decisions, restore human editorial handling, preserve briefs and traces, invalidate affected certificates, and require corrected enumeration plus independent re-checking before resumption."},"negative_tests":{"strongest_counterevidence":"The strongest counterevidence would establish that every actual brief already has an enforced finite asset set, finite edit grid, fixed maximum structure, finite parameters, and a total rule evaluator, with a known complete terminating solver. The declared problem would then be decidable, and remaining difficulties would concern complexity, encoding quality, or workflow rather than a computability boundary.","problem_falsifier":"The problem is falsified if stakeholders ask only for candidate generation, never interpret unsuccessful search as infeasibility, already label incomplete results UNKNOWN, and make no class-wide promise to decide existence for arbitrary executable briefs.","intervention_falsifier":"The intervention fails if the reduction does not create well-formed target instances, bounded-mode membership can be bypassed, the candidate enumerator is incomplete, pruning removes a valid candidate, dovetailing is unfair, a nonterminating evaluator enters exact mode, or downstream interfaces still collapse UNKNOWN into infeasible. Its operational premise is also defeated if pilot users cannot express any bounded brief they are willing to evaluate.","risks":["Finite bounds may exclude the edit structure needed to satisfy a real brief.","Complete bounded search may be decidable but computationally impractical.","Incorrect or incomplete asset metadata can make a formally valid cut operationally unusable.","Machine-stated rules may omit narrative intent, pacing, performance quality, or contextual standards.","Open dovetailing can consume resources indefinitely unless operational budgets are enforced.","Aggressive pruning can invalidate an infeasibility certificate if its preservation argument is wrong.","Users may interpret a checked formal witness as creative endorsement.","Candidate cuts, scripts, or diagnostic traces may expose unreleased footage or confidential campaign requirements.","Frequent UNKNOWN results may encourage informal bypasses unless escalation ownership is explicit."]},"next_evidence_step":"Run one closed-book non-production study. Freeze a minimal cut encoding, a finite bounded-rule grammar, and an execution model. Have a reviewer who did not author the argument check the program-input-to-C_n reduction for source-to-target direction, total and computable construction, well-formed candidates, bounded evaluator termination, existential answer preservation, and scope assumptions. Implement a prototype using 20 synthetic briefs: five bounded feasible briefs with witnesses placed at different enumeration depths, five bounded infeasible briefs with independently enumerated finite spaces, three open briefs with accepting witnesses behind deliberately nonterminating earlier evaluators, three open briefs with no witness in the operational budget, two malformed or out-of-fragment briefs, and two briefs whose bounds change after certification. Record routing, explored candidates, evaluator steps, fairness trace, certificate or witness, output label, and reviewer judgment. Fail the first evidence if an in-bound witness is missed, a bounded infeasibility certificate omits a candidate, a later open witness is starved, an incomplete search returns infeasible, or a changed bound preserves the prior certificate.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Proposal 1 concerned universal behavioral verification of an already-authored interactive media package: whether every viewing session terminates within duration and avoids uncleared assets. It used finite-state authoring restrictions, over-approximate model checking, and release-preflight routing. Proposal 2 concerned universal semantic equivalence between an original and replacement render graph. It used a proof-carrying render-expression fragment, canonicalization or rewrite certificates, and counterexample-based substitution control. Proposal 3 instead concerns existential editorial synthesis: whether any cut satisfying a programmable brief exists and can be found. Its distinct causal path maps halting to the existence of some candidate C_n, then separates one-sided witness discovery through fair dovetailing from negative conclusions available only by exhaustive finite search. It is independently adoptable as an editorial planning service without verifying interactive-session behavior or authorizing render-graph substitutions.","revision_record":{"parent_version":null,"progress_targets_addressed":["Created one complete candidate at proposal index 3.","Selected an existential synthesis problem distinct from universal behavior verification and render equivalence.","Specified actors, state, consequence, causal mechanism, rivals, authority, safeguards, falsifiers, and bounded evidence.","Explained diversity from both earlier sealed proposals."],"conceptual_changes":["Initial version; applied computability-boundary mapping to existence and synthesis of constraint-conforming editorial cuts."],"operational_changes":["Initial version; introduced exact bounded feasibility mode, open dovetailed witness search, explicit UNKNOWN, coverage certificates, and editorial authority controls."],"evidence_changes":["Initial version; prior art remains unsearched and first evidence is limited to independent reduction review and 20 synthetic briefs."],"claim_changes":["Initial version; makes no novelty, prevalence, demand, or effect-size claim and conditions the unrestricted impossibility conclusion on a valid reviewed reduction."]}}