{"paper":{"title":"What Should Explanations Contain? A Human-Centered Explanation Content Model for Local, Post-Hoc Explanations","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"A fourteen-code model derived from industrial user studies specifies what content local post-hoc explanations should contain.","cross_cats":[],"primary_cat":"cs.HC","authors_text":"Helmut Degen","submitted_at":"2026-05-14T00:05:00Z","abstract_excerpt":"Which categories of explanation content are relevant for users of industrial AI systems, and how can those categories be organized for local, post-hoc explanations? To address these questions, a hybrid inductive-deductive qualitative content analysis was applied to 325 meaning units drawn from six user studies in building technology, manufacturing, AI software development, and hospital cybersecurity. The inductive phase produced an initial twelve-code structure. A theory-informed coverage assessment and expert review then added two further codes, Rule base and What-if backward, that were not i"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The resulting fourteen-code model is organized into four groups: rule-based, causal, epistemic (actual), and epistemic (similar), with twelve codes grounded in the corpus and two as theoretical extensions. An eleven-member expert panel supported the content adequacy of all codes (I-CVI ≥ 0.82; scale-level agreement of 0.93 for relevance, 0.92 for boundary clarity, and 0.94 for understandability). A stratified subsample of 82 units yielded Krippendorff's α = 0.920 and Cohen's κ = 0.920.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the six user studies and 325 meaning units adequately represent the explanation needs of users across broader industrial AI applications, and that expert panel judgments on content adequacy will hold for actual end-users in deployment.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A 14-code content model for local post-hoc AI explanations, derived from 325 user statements and validated by experts with high reliability scores.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A fourteen-code model derived from industrial user studies specifies what content local post-hoc explanations should contain.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"802ca3a26dbac9d80354cfeac85a4c5a47e5e5616fbb3c037b79888406685f39"},"source":{"id":"2605.14207","kind":"arxiv","version":1},"verdict":{"id":"291ccc08-7eba-4bca-9d01-34dca96bd6dd","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:47:02.701492Z","strongest_claim":"The resulting fourteen-code model is organized into four groups: rule-based, causal, epistemic (actual), and epistemic (similar), with twelve codes grounded in the corpus and two as theoretical extensions. An eleven-member expert panel supported the content adequacy of all codes (I-CVI ≥ 0.82; scale-level agreement of 0.93 for relevance, 0.92 for boundary clarity, and 0.94 for understandability). A stratified subsample of 82 units yielded Krippendorff's α = 0.920 and Cohen's κ = 0.920.","one_line_summary":"A 14-code content model for local post-hoc AI explanations, derived from 325 user statements and validated by experts with high reliability scores.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the six user studies and 325 meaning units adequately represent the explanation needs of users across broader industrial AI applications, and that expert panel judgments on content adequacy will hold for actual end-users in deployment.","pith_extraction_headline":"A fourteen-code model derived from industrial user studies specifies what content local post-hoc explanations should contain."},"references":{"count":158,"sample":[{"doi":"","year":1994,"title":"Case-based reasoning: Foundational issues, methodological variations, and system approaches","work_id":"94941092-c87f-4c1f-bca5-b437f0d1ddd9","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1145/3313831.3376615","year":2020,"title":"Lau, Jose Echevarria, and Zoya Bylinskii","work_id":"11c5e348-c2a5-4713-8d81-3ff8503e4e3c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1109/fuzz-ieee.2019.8858846","year":2019,"title":"Adhikari, A., Tax, D.M.J., Satta, R., Faeth, M., 2019. LEAFAGE: Example-Based and Feature-Importance-Based Explanations for Black-Box ML Models, in: Proceedings of the 2019 IEEE International Conferen","work_id":"99d70b1d-db71-4462-898e-3fa54e0b9222","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"UP- GRADE: Universal Patching and Remediation for Autonomous De- fense.https://arpa-h.gov/explore-funding/programs/upgrade","work_id":"fc51045a-cadf-45b7-8420-09f3bd6d6e3a","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1109/fuzz-ieee55066.2022.9882743","year":2022,"title":"Aechtner, J., Cabrera, L., Katwal, D., Onghena, P., Valenzuela, D.P., Wilbik, A., 2022. Comparing User Perception of Explanations Devel- oped with XAI Methods, in: Proceedings of the 2022 IEEE Interna","work_id":"d3e6df67-a142-4ab1-8d26-5b2c6d383ed0","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":158,"snapshot_sha256":"8a84f5a896cce841d8f2a704fffd36d6bf6d85ecbd2860e53975070d9cba0a7f","internal_anchors":2},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}