{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp09_archetype_breadth150_20260804","cell_id":"priority_based_admission__computer_science","arm":"BREADTH_PROBE_ONE_SHOT","candidate_id":"priority_based_admission__computer_science__P1","proposal_index":1,"version":0,"title":"Priority Admission of Fuzzer Witnesses to a Runtime-Bounded Regression Suite","problem":"A software project discovers more bug-triggering inputs through fuzzing than its continuous-integration runtime budget can retain as permanent regression tests. Every input may document a real defect, but inputs differ in defect severity, behavioral novelty, reproducibility, redundancy, and execution cost. Unstructured accumulation lets redundant or slow witnesses consume the bounded suite budget while later witnesses for distinct, consequential failure modes remain outside routine regression checking.","actors":["Software maintainers who own the regression suite","Fuzzing operators who submit bug-triggering inputs","Continuous-integration operators who manage test-runtime capacity","Developers responsible for defects represented by admitted or archived witnesses"],"observable_state":"For each candidate witness, maintainers can observe whether it reproduces deterministically, the affected component and failure class, its behavioral or coverage overlap with admitted witnesses, minimized input size, execution time, and whether the represented defect has another permanent witness. The constrained state is visible when the aggregate runtime of eligible candidates exceeds the regression suite's declared fuzz-witness budget.","consequence":"If admission remains implicit, the permanent suite can spend scarce runtime repeatedly checking similar failures while failing to continuously check distinct or higher-consequence defects. Developers then receive weaker protection against recurrence, and removal or exclusion decisions become difficult to explain or contest.","affected_objective":"Preserve a bounded continuous-integration runtime budget while maintaining permanent regression witnesses for the most consequential and behaviorally distinct known failure modes.","intervention":"Create an explicit admission boundary between the complete witness archive and the runtime-bounded permanent regression suite. Classify each reproducible, minimized witness using published criteria: represented defect consequence, absence of another admitted witness for the same failure mode, behavioral novelty, execution cost, and confidence that the result is not flaky. Admit candidates by eligibility ranking within a fixed runtime budget, reserve a declared fraction of that budget for witnesses representing previously uncovered high-consequence failure classes, and archive every nonadmitted witness with its score and reason. Permit a maintainer who owns the affected component to request review, and periodically reconsider archived witnesses when admitted tests are removed, accelerated, or found redundant.","structural_mapping":[{"archetype_element":"Constrained admission boundary","domain_realization":"The boundary between the full archive of reproducible fuzzer inputs and the subset executed on every qualifying continuous-integration run."},{"archetype_element":"Candidates with priority-relevant attributes","domain_realization":"Minimized bug-triggering inputs characterized by defect consequence, failure-mode novelty, reproducibility, redundancy, and runtime cost."},{"archetype_element":"Explicit priority policy","domain_realization":"A documented eligibility-ranking rule that favors distinct, consequential, deterministic witnesses while accounting for their consumption of the runtime budget."},{"archetype_element":"Scarce protected capacity","domain_realization":"A fixed fuzz-regression runtime allowance, including a reserved fraction that ordinary redundant witnesses cannot consume."},{"archetype_element":"Defined handling for nonadmitted candidates","domain_realization":"A durable archive retaining the witness, reproduction metadata, admission score, exclusion reason, and eligibility for later reconsideration."},{"archetype_element":"Review or appeal path","domain_realization":"Component owners can challenge classification evidence, and an independent suite maintainer records whether the admission decision is affirmed or changed."},{"archetype_element":"Starvation prevention","domain_realization":"Archived witnesses gain reconsideration priority as their represented failure class remains uncovered over successive review cycles, without guaranteeing admission beyond the runtime budget."}],"mechanism_mapping":[{"mechanism_slug":"eligibility_ranking","role":"Converts observable witness attributes into a comparable, auditable admission order before the permanent-suite boundary.","counterfactual_removal":"Without ranking, admission would revert to arrival order or informal advocacy, so runtime could be captured by redundant or conspicuous witnesses."},{"mechanism_slug":"reserved_capacity_fraction","role":"Protects part of the runtime budget for witnesses representing uncovered high-consequence failure classes.","counterfactual_removal":"Without a reservation, numerous ordinary witnesses could exhaust the budget before a qualifying critical witness is considered."},{"mechanism_slug":"review_or_appeal_path","role":"Allows incorrect severity, redundancy, flakiness, or runtime classifications to be corrected using recorded evidence.","counterfactual_removal":"Without review, a mistaken classification could silently exclude the only durable witness for a consequential defect."}],"causal_chain":["The permanent regression suite has a declared runtime ceiling, while reproducible fuzzer witnesses compete for membership.","Each candidate is minimized and classified using explicit consequence, novelty, reproducibility, redundancy, and cost evidence.","Eligibility ranking and the reserved-capacity rule govern admission rather than arrival time or developer pressure.","Admitted witnesses consume the bounded runtime allowance; nonadmitted witnesses remain acknowledged and reproducible in the archive.","The routine suite therefore directs its limited recurring execution capacity toward distinct, higher-priority failure witnesses.","Logged decisions, appeals, and periodic reclassification expose misclassification, gaming, bias toward particular components, and starvation for correction."],"baseline":"Maintain the permanent suite through ad hoc maintainer judgment or admission by discovery order until the runtime budget is reached; subsequently discovered witnesses are manually substituted, deferred, or forgotten without a uniform explanation or reconsideration rule.","nearest_rivals":["Regression-test prioritization, which changes execution order but does not decide membership in the bounded permanent suite","Corpus minimization or deduplication, which removes redundant inputs but does not allocate capacity among distinct defects of unequal consequence","Test sharding or additional CI capacity, which changes execution capacity but does not govern admission when a runtime ceiling remains","First-come-first-served admission, which is explicit but ignores consequence and novelty differences among eligible witnesses"],"remaining_contrastive_claim":"The distinctive intervention is governed membership at the permanent-suite boundary: candidates that are all legitimate defect witnesses receive unequal admission under an explicit, reviewable priority policy, while excluded witnesses receive defined archival handling. Merely reordering tests, deduplicating inputs, or adding workers does not instantiate this boundary decision.","authority_safety":{"decision_authority":"The repository's designated regression-suite maintainer may approve the policy and individual admissions; affected component owners may supply evidence and appeal but cannot unilaterally self-prioritize their witnesses.","authorized_first_step":"Run a read-only shadow classification on a bounded historical sample and compare the hypothetical admitted set with the existing suite; do not add, remove, or suppress any test.","excluded_actions":["Automatically deleting nonadmitted witnesses","Changing production code or release gates","Allowing submitters to assign their own final priority","Using undocumented exceptions to consume reserved capacity","Inferring defect consequence from developer identity, team status, or customer prestige"],"halt_rollback":"Halt the shadow exercise if required metadata cannot be reconstructed, materially similar witnesses receive inconsistent classifications, or protected attributes or organizational status influence ranking. Because the first step changes no suite membership, rollback consists of discarding the shadow ranking and retaining the existing suite and archive unchanged."},"negative_tests":{"strongest_counterevidence":"The candidate policy would be undermined if simple corpus minimization fits every distinct reproducible failure witness within the runtime ceiling, or if measured classifications are too unstable for independent maintainers to reach materially consistent admission decisions.","problem_falsifier":"The problem is falsified for the studied repository if all distinct reproducible fuzzer witnesses fit within the declared recurring runtime budget after ordinary minimization, leaving no actual admission scarcity.","intervention_falsifier":"The intervention is falsified if, under a fixed budget in shadow evaluation, it admits fewer independently judged high-consequence distinct failure classes than the baseline, frequently excludes the sole witness for such a class, or requires assessment effort comparable to the test runtime it is meant to allocate.","risks":["Severity or novelty can be misclassified, excluding an important witness.","Submitters may exaggerate defect consequence or manipulate coverage evidence.","Components with better instrumentation may appear more novel and receive disproportionate capacity.","The reserved fraction may sit unused or be eroded through exceptions.","Low-priority witnesses may remain archived indefinitely.","Ranking administration may consume excessive maintainer time.","An archive without integrity or reproducibility checks may become a silent graveyard.","Optimizing for short runtime may systematically disfavor expensive but necessary end-to-end witnesses."]},"next_evidence_step":"For one repository, select at most 60 archived and admitted fuzzer witnesses from a fixed historical interval. Two maintainers independently classify them using a preregistered rubric, measure each witness's runtime and reproducibility in an isolated test environment, and construct a shadow admitted set under the current fuzz-regression runtime ceiling. Record inter-rater disagreements, represented failure classes, sole-witness exclusions, assessment time, appeals that would change classifications, and differences from the existing set. Make no test-suite changes from this exercise.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Not assessed because runtime isolation prohibits inspecting other experiment candidates or proposals.","revision_record":{"parent_version":null,"progress_targets_addressed":[],"conceptual_changes":[],"operational_changes":[],"evidence_changes":[],"claim_changes":[]}}