Language Advantage Is a Function of the Executor: criteria self-audit and undecidability audit in cross-language code generation by LLM coders (timetable-ab)
Identification
- Clé publique
- radar:r-e3bc55403957cc3a2597ba99836d60ef
- Auteurs
- 许宸玮
- Année déclarée
- 2026
- Langue déclarée
- und
- Type source
- Dataset
- Première détection
- 2026-11-13T08:57:06.011867+00:00
- Dernière observation des métadonnées
- 2026-11-13T08:57:06.011867+00:00
- Dernière vérification HTTP
- Non effectuée
Institutions déclarées par la source
北华大学 文学院
Résumé documentaire — langue originale
Current state (v1.7.6, 2026-11-02). This description states current values. Earlier revision notes remain in the version history; where an older note quotes a figure this one supersedes, this one governs. The headline numbers. Language effect: inline JavaScript/Common Lisp = 1.008x (95% CI [0.829, 1.116]); abstraction = 1.007x ([0.879, 1.095]). Both intervals cross 1.0, so "identical" cannot be excluded. What survives is the writing style, not the language: collapsing repetition into one place saves 17% of characters (ratio 1.200x; as a saving interval 7.2%-30.7% in Common Lisp and 6.8%-26.6% in JavaScript, derived as 1 - 1/ratio from the ratio intervals [1.0771, 1.4423] and [1.0725, 1.3619] given in v4-ratios.tsv). Two earlier figures are withdrawn and named here rather than deleted: the "28-40% saving" (v1.6.0) and the "1.93x" language ratio quoted in older notes, which compared different abstraction levels, unequal JSON ecosystems and repeatedly registered runs. What is re-checkable, and what is not. Two languages, a 15-operator specification, 36 invariants, plus 24 cells with archived prompts and frozen spec/criteria hashes. A second layer of 13 independent implementations under 80 invariants is kept as a prior account. Runs that did not archive their task descriptions are reported as observations, never as evidence. Criteria self-audit (v1.7.0-v1.7.2). 88 criteria = 84 content + 2 form + 2 method (the numbers 81 and 82 are deliberately vacant, because H1/H2 live in a separate file), each newly added one proven able to fail by a counterexample. Three findings are recorded rather than hidden: the 36 scale criteria kill 0 of 12 mutants while the gate only ran the scale set; every assertion sits behind JSON.parse, so serialisation fidelity is constrained by nothing; and three mutants - leading whitespace, require("fs"), a JSON.parse call - are invisible to all 84 content criteria, caught only by H1/S1/S2 respectively. At v1.7.0, 8 of 84 content criteria had a counterexample; experiments/timetable-ab/counterexamples/ now ships a clean 84/84 base plus a mutation matrix that records, for each mutant, exactly which criteria turn red. Undecidability is still the point. Established (3), refuted (3), undecidable (6). Missing conditions: C1 no weak-executor channel; C2 no exclusive machine or clock alignment; C3 a single model family. Incomplete rather than missing: C4 silent-failure coverage - 12 mutants audited, 8 killed, 4 survivors all classified, zero Class-A blind spots, which is not the same as "no blind spots"; C5 scale (at most 30 ops, at most 1k lines). Protocol results. Expansion-time rejection (19/19 with validation off, load-time failure with it on); zero ceremony cost (0/0 manual edits); live-image REPL saves characters (9028 to 5957). S-expressions are not the most compact (CL 7506 vs JS 5933 characters at 5 ops); skill-pack intervention showed no main effect (0/6). Gate and packaging (v1.7.5-v1.7.6). The regression gate now guards the v3-full criteria set by fingerprint, the frozen files by hash, and the release metadata itself: version must agree between CITATION.cff and .zenodo.json, the citation DOI must be the concept DOI, authors and contributors must agree on both sides, and both the notes and this description must mention the current version. Every added step was put through a destruction test - edited to break, observed to fail, restored, observed to pass - because a check that has never failed is not known to be a check. Still open: the counterexample step is not wired into the gate. What this dataset does not do. It does not read "not detected" as "not present"; it does not extrapolate beyond the measured scale; it does not delete data - the timing columns, the retired line-count metric and all nineteen recorded failures of the measurement apparatus remain in the chronicle, annotated with their scope. Included: run artefacts, frozen specifications with recorded sha256 freeze hashes, the 36 invariants with their go-red proofs, mutation scores with sample-size bounds, CONDITIONS.md, results.tsv with RESULTS-SCHEMA.md, an append-only chronicle (E-01 through E-05), CLAIMS.md, and the criteria-engineering documents (SPEC-AUDIT.md, CRITERIA-ENGINEERING.md, C4-REPORT.md, ATTACK-WEAK-COVERAGE.md, DO-NOT-MODIFY.md). Excluded: compiled binaries, build caches, machine-local temporary files, personal data. Intended use: replication of the measurement protocol; reuse of the mutation-testing gate as a way to audit one's own criteria; and an empirical basis for benchmark-validity work in LLM code generation.
Identifiants
- DOI
- 10.5281/zenodo.23884677
- OAI
- oai:zenodo.org:23884677
Repères documentaires
- Famille documentaire
- dataset
- Types déclarés par les sources
- Dataset
- Éditeurs
- Zenodo
Les mentions d’entités ne constituent pas une preuve de reconnaissance juridique. La juridiction étudiée est distincte du pays d’affiliation des auteurs.
Sources
DataCite · Document 1
Citer cette référence
许宸玮. Language Advantage Is a Function of the Executor: criteria self-audit and undecidability audit in cross-language code generation by LLM coders (timetable-ab). 2026. https://doi.org/10.5281/zenodo.23884677
JSON · CSL-JSON · BibTeX · RIS · CSV · JSON-LD ·
APA, Chicago, ISO 690 : formats bibliographiques fondés sur les champs disponibles. OSCOLA / Bluebook : données structurées uniquement ; aucune conformité juridique complète n’est revendiquée sans vérification des éléments nécessaires.
- APA
- 许宸玮. (2026). Language Advantage Is a Function of the Executor: criteria self-audit and undecidability audit in cross-language code generation by LLM coders (timetable-ab). https://doi.org/10.5281/zenodo.23884677
- Chicago
- 许宸玮. Language Advantage Is a Function of the Executor: criteria self-audit and undecidability audit in cross-language code generation by LLM coders (timetable-ab). 2026. https://doi.org/10.5281/zenodo.23884677
- ISO 690
- 许宸玮. Language Advantage Is a Function of the Executor: criteria self-audit and undecidability audit in cross-language code generation by LLM coders (timetable-ab). 2026. https://doi.org/10.5281/zenodo.23884677
Warning: Undefined array key "last_link_verified" in /home/clients/e27d81e12f04d2bfd747aa264540ac56/sites/_UDTR_APP/legal_library_reference.php on line 31
UDTR — Bibliothèque juridique mondiale des droits de la Nature. Language Advantage Is a Function of the Executor: criteria self-audit and undecidability audit in cross-language code generation by LLM coders (timetable-ab). Notice documentaire: https://www.declarationuniverselledesdroitsdelarbre.org/global-rights-of-nature-radar/reference/r-e3bc55403957cc3a2597ba99836d60ef/ · Dernière vérification HTTP : non effectuée
Données de cette version figée
Dossiers et relations documentaires
Rôles des sources
Source bibliographique / publication d’origine · https://zenodo.org/doi/10.5281/zenodo.23884677
Un lien vers une autorité ne prouve pas à lui seul une affirmation juridique. Les résumés et les textes intégraux restent soumis aux conditions de leur source.
Statut documentaire
Pertinence à examiner
Bibliothèque
Cette identité n’est pas encore intégrée à la Bibliothèque juridique.