Condensation Failure G: The Locatable Boundary of an Audit Format—Tight Lower Bounds on the Earliest-Failure Cell, the Conjunctive Three-Question Format, and the Rigidity of the Four-Object Format
We study the locatable completeness of an audit format: whether a decidability claim can pass a fixed set of review tests, and whether the failure informationidentifies the class of the defect. Under the uniform setting of n = 4 defect classes,mreview tests, and a 0–1 failure matrix A ∈ {0,1}m×4 (rows index tests, columnsindex defect classes) we carry out the full enumeration and deliver proof-level conclusions; in the general statements n and m are pinned by convention to n = 4 andm ∈ {1,2,3,4}. The conclusions unfold along two localisation disciplines. First,the earliest-failure discipline used throughout the series takes Φ(j) = minFj asits localisation information and requires Φ to be injective on defect classes; weprove that this forces m ≥ n, so under the earliest-failure discipline no completelocatable three-question format exists at n = 4, and the permutation matrix atm=nis a tight example. Second, after relaxing to full-failure-pattern localisation,which asks the failure sets to be pairwise distinct and non-empty, the informationtheoretic counting bound drops to m ≥ ⌈log2(n + 1)⌉, that is m ≥ 3 at n = 4, and3 is attained: we exhibit an explicit three-question conjunctive composite formatwhose failure patterns are {1}, {2}, {3}, {1,2,3}, complete, locatable and minimal, at the price that the earliest-failure coordinate degenerates to (1,2,3,1) andthe localisation information moves from a single coordinate to the whole failureset. Third, when m = n = 4 holds together with the earliest-failure discipline andthe single-defect discipline, the failure matrix must be a permutation matrix, andthe four-object format is unique up to relabelling, giving 4! = 24 formats; but theearliest-failure discipline alone does not force the single-defect form, since there are1536 earliest-failure locatable formats at m = 4, of which the permutation matricesare only 24. Fourth, minimality acquires an exact row–column criterion (Proposition 6 of Chapter 9): a format is non-minimal if and only if there is a coordinatek such that no column is exactly {k} and no two columns differ only in position k;hence 3000 of the 32760 full-failure-pattern locatable formats at m = 4 are minimal,and the presence of an all-zero row is only the trivial sub-case of the criterion. Themechanism admits a one-sentence summary: earliest failure compresses the localisation information into a single coordinate, and there are only m coordinates, hencem ≥ n; full-failure-pattern localisation keeps 2m patterns, hence three questionssuffice. At the level of readings, under the full-failure-pattern discipline there are840 three-question formats at n = 4 and 32760 full-failure-pattern locatable formatsat m = 4, both echoing Theorem 2 and the inventory of three-question formats.On the complexity side, deciding completeness and deciding either kind of locatability are polynomial-time; inside the family that admits conjunctive compositetests, choosing a minimum complete format is exactly minimum set covering andis therefore NP-hard, while minimisation inside the single-object-presence familyremains open, and the minimisation complexity of locatable formats is not provedhere. The localisation counts of this paper rest on the single-defect claim hypothesis: a claim entering review is either qualified or missing exactly one class of object,and claims missing several classes are described structurally only and do not enterthe counts; the hypothesis and the union lemma for the failure patterns of multidefect claims are registered explicitly, and without the hypothesis a format suppliesat most inclusion-level information. The outstanding items are reconciled one byone against the five carried entries of Section 10.2: the complexity of minimisinglocatable formats, whether a fifth cell brings a new defect class, constructions forn > 4, the extension to valued formats, and the enumeration of composite-defectidentities.
Authors
- changzheng zhou (ORCID: https://orcid.org/0009-0002-4186-7877)
- ziqing zhou (ORCID: https://orcid.org/0009-0005-6883-6300)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24408847
- Primary Topic
- Computability, Logic, AI Algorithms
- Type
- preprint