Contracts Across Modalities: Detecting Embedding Staleness and Governance Drift in Multimodal Lakehouse Architectures
DOI:
https://doi.org/10.22399/ijcesen.5469Keywords:
Data contracts, multimodal lakehouse, vector database, staleness, embeddings, provenanceAbstract
Enterprise lakehouses increasingly colocate typed records, unstructured source objects, and vector embeddings derived from those objects. Table formats, object stores, and vector databases provide durable storage and efficient retrieval, but they do not make a derived vector self-invalidating when its source changes. An embedding can therefore remain queryable after a document is edited, redacted, or deleted; after consent is revoked; or after the active embedding model is upgraded. These failures are operationally quiet: the vector has the expected dimension, the index remains healthy, and no schema constraint necessarily fails. This paper defines a cross-modal contract as C = (R,E,S,V,G), where R is a source record or object, E is its embedding, S is a provenance signature captured at embedding time, V is a validity predicate over current content and model state, and G is the governance state that authorizes retention and use. We implement the contract in a reproducible pre-production benchmark containing 180 contracts: 60 tabular records, 60 documents, and 60 images. Thirty-six are unchanged controls; 144 contain seeded failures spanning source edits, redactions, deletions, model-version upgrades, consent revocations, vector corruption, and unversioned model-artifact replacement. Lightweight signature, model-version, and governance checks detect 120 of 144 failures (83.3%) with no false positives. They intentionally miss vector corruption and unversioned artifact replacement because neither changes the stored source signature, declared version, or governance metadata. Selective re-embedding detects those residual classes; the combined policy detects all 144 seeded failures in this benchmark. Median lightweight evaluation ranges from 6.5 to 27.8 µs per contract, whereas re-embedding ranges from 0.42 to 2.84 ms, a 62-102 cost difference. The contribution is not a claim that hashing solves semantic quality. It is a precise separation of three obligations that current systems often conflate: source freshness, model compatibility, and governance authority, together with an escalation path for failures that require inspection of the vector payload itself.
References
[1] R. Y. Wang and D. M. Strong, “Beyond accuracy: What data quality means to data consumers,” J. Manage. Inf. Syst., vol. 12, no. 4, pp. 5–33, 1996, doi: 10.1080/07421222.1996.11518099. DOI: https://doi.org/10.1080/07421222.1996.11518099
[2] C. Batini, C. Cappiello, C. Francalanci, and A. Maurino, “Methodologies for data quality assessment and improvement,” ACM Comput. Surv., vol. 41, no. 3, Art. no. 16, 2009, doi: 10.1145/1541880.1541883. DOI: https://doi.org/10.1145/1541880.1541883
[3] Y. L. Simmhan, B. Plale, and D. Gannon, “A survey of data provenance in e-science,” SIGMOD Rec., vol. 34, no. 3, pp. 31–36, 2005, doi: 10.1145/1084805.1084812. DOI: https://doi.org/10.1145/1084805.1084812
[4] P. Buneman, S. Khanna, and W.-C. Tan, “Why and where: A characterization of data provenance,” in Database Theory—ICDT 2001, pp. 316–330, doi: 10.1007/3-540-44503-X_20. DOI: https://doi.org/10.1007/3-540-44503-X_20
[5] J. Cheney, L. Chiticariu, and W.-C. Tan, “Provenance in databases: Why, how, and where,” Found. Trends Databases, vol. 1, no. 4, pp. 379–474, 2009, doi: 10.1561/1900000006. DOI: https://doi.org/10.1561/1900000006
[6] L. Moreau et al., “The Open Provenance Model core specification (v1.1),” Future Gener. Comput. Syst., vol. 27, no. 6, pp. 743–756, 2011, doi: 10.1016/j.future.2010.07.005. DOI: https://doi.org/10.1016/j.future.2010.07.005
[7] A. M. Chirigati, R. Rampin, D. Shasha, and J. Freire, “ReproZip: Computational reproducibility with ease,” in Proc. ACM SIGMOD, 2016, pp. 2085–2088, doi: 10.1145/2882903.2899401. DOI: https://doi.org/10.1145/2882903.2899401
[8] M. Boehm et al., “Automated provenance tracking in data science scripts,” in Proc. 26th ACM SIGKDD, 2020, pp. 2510–2518, doi: 10.1145/3394486.3403205. DOI: https://doi.org/10.1145/3394486.3403205
[9] M. Armbrust et al., “Delta Lake: High-performance ACID table storage over cloud object stores,” Proc. VLDB Endow., vol. 13, no. 12, pp. 3411–3424, 2020, doi: 10.14778/3415478.3415560. DOI: https://doi.org/10.14778/3415478.3415560
[10] S. Ait Errami, H. Hajji, K. Ait El Kadi, and H. Badir, “Spatial big data architecture: From data warehouses and data lakes to the LakeHouse,” J. Parallel Distrib. Comput., vol. 176, pp. 70–79, 2023, doi: 10.1016/j.jpdc.2023.02.007. DOI: https://doi.org/10.1016/j.jpdc.2023.02.007
[11] J. Wang et al., “Milvus: A purpose-built vector data management system,” in Proc. ACM SIGMOD Int. Conf. Manage. Data, 2021, pp. 2614–2627, doi: 10.1145/3448016.3457550. DOI: https://doi.org/10.1145/3448016.3457550
[12] M. Aumüller, E. Bernhardsson, and A. Faithfull, “ANN-Benchmarks: A benchmarking tool for approximate nearest neighbor algorithms,” Inf. Syst., vol. 87, Art. no. 101374, 2020, doi: 10.1016/j.is.2019.02.006. DOI: https://doi.org/10.1016/j.is.2019.02.006
[13] Y. A. Malkov and D. A. Yashunin, “Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 42, no. 4, pp. 824–836, 2020, doi: 10.1109/TPAMI.2018.2889473. DOI: https://doi.org/10.1109/TPAMI.2018.2889473
[14] J. Johnson, M. Douze, and H. Jégou, “Billion-scale similarity search with GPUs,” IEEE Trans. Big Data, vol. 7, no. 3, pp. 535–547, 2021, doi: 10.1109/TBDATA.2019.2921572. DOI: https://doi.org/10.1109/TBDATA.2019.2921572
[15] N. Reimers and I. Gurevych, “Sentence-BERT: Sentence embeddings using Siamese BERT-networks,” in Proc. EMNLP-IJCNLP, 2019, pp. 3982–3992, doi: 10.18653/v1/D19-1410. DOI: https://doi.org/10.18653/v1/D19-1410
[16] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of deep bidirectional transformers for language understanding,” in Proc. NAACL-HLT, 2019, pp. 4171–4186, doi: 10.18653/v1/N19-1423. DOI: https://doi.org/10.18653/v1/N19-1423
[17] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE CVPR, 2016, pp. 770–778, doi: 10.1109/CVPR.2016.90. DOI: https://doi.org/10.1109/CVPR.2016.90
[18] N. Dalal and B. Triggs, “Histograms of oriented gradients for human detection,” in Proc. IEEE CVPR, 2005, pp. 886–893, doi: 10.1109/CVPR.2005.177. DOI: https://doi.org/10.1109/CVPR.2005.177
[19] N. Halko, P.-G. Martinsson, and J. A. Tropp, “Finding structure with randomness: Probabilistic algorithms for constructing approximate matrix decompositions,” SIAM Rev., vol. 53, no. 2, pp. 217–288, 2011, doi: 10.1137/090771806. DOI: https://doi.org/10.1137/090771806
[20] I. T. Jolliffe, Principal Component Analysis, 2nd ed. New York, NY, USA: Springer, 2002, doi: 10.1007/b98835. DOI: https://doi.org/10.1007/b98835
[21] M. Mitchell et al., “Model cards for model reporting,” in Proc. FAT*, 2019, pp. 220–229, doi: 10.1145/3287560.3287596. DOI: https://doi.org/10.1145/3287560.3287596
[22] T. Gebru et al., “Datasheets for datasets,” Commun. ACM, vol. 64, no. 12, pp. 86–92, 2021, doi: 10.1145/3458723. DOI: https://doi.org/10.1145/3458723
[23] E. M. Bender and B. Friedman, “Data statements for natural language processing: Toward mitigating system bias and enabling better science,” Trans. Assoc. Comput. Linguist., vol. 6, pp. 587–604, 2018, doi: 10.1162/tacl_a_00041. DOI: https://doi.org/10.1162/tacl_a_00041
[24] P. Domingos, “A few useful things to know about machine learning,” Commun. ACM, vol. 55, no. 10, pp. 78–87, 2012, doi: 10.1145/2347736.2347755. DOI: https://doi.org/10.1145/2347736.2347755
[25] E. Breck, S. Cai, E. Nielsen, M. Salib, and D. Sculley, “The ML test score: A rubric for ML production readiness and technical debt reduction,” in Proc. IEEE Int. Conf. Big Data, 2017, pp. 1123–1132, doi: 10.1109/BigData.2017.8258038. DOI: https://doi.org/10.1109/BigData.2017.8258038
[26] N. Sambasivan et al., “Everyone wants to do the model work, not the data work: Data cascades in high-stakes AI,” in Proc. CHI, 2021, doi: 10.1145/3411764.3445518. DOI: https://doi.org/10.1145/3411764.3445518
[27] S. Zhang, A. Bouadi, and K. Werder, “Establishing data provenance for responsible artificial intelligence systems,” ACM Trans. Manage. Inf. Syst., vol. 13, no. 2, Art. no. 22, 2022, doi: 10.1145/3503488. DOI: https://doi.org/10.1145/3503488
[28] M. Veale, R. Binns, and L. Edwards, “Algorithms that remember: Model inversion attacks and data protection law,” Phil. Trans. R. Soc. A, vol. 376, Art. no. 20180083, 2018, doi: 10.1098/rsta.2018.0083. DOI: https://doi.org/10.1098/rsta.2018.0083
[29] Y. Cao and J. Yang, “Towards making systems forget with machine unlearning,” in Proc. IEEE Symp. Security Privacy, 2015, pp. 463–480, doi: 10.1109/SP.2015.35. DOI: https://doi.org/10.1109/SP.2015.35
[30] C. Guo, T. Goldstein, A. Hannun, and L. van der Maaten, “Certified data removal from machine learning models,” in Proc. ICML, 2020, pp. 3832–3842, doi: 10.48550/arXiv.1911.03030.
[31] R. Shokri, M. Stronati, C. Song, and V. Shmatikov, “Membership inference attacks against machine learning models,” in Proc. IEEE Symp. Security Privacy, 2017, pp. 3–18, doi: 10.1109/SP.2017.41. DOI: https://doi.org/10.1109/SP.2017.41
[32] M. Fredrikson, S. Jha, and T. Ristenpart, “Model inversion attacks that exploit confidence information and basic countermeasures,” in Proc. ACM CCS, 2015, pp. 1322–1333, doi: 10.1145/2810103.2813677. DOI: https://doi.org/10.1145/2810103.2813677
[33] J. Gama, I. Žliobaitė, A. Bifet, M. Pechenizkiy, and A. Bouchachia, “A survey on concept drift adaptation,” ACM Comput. Surv., vol. 46, no. 4, Art. no. 44, 2014, doi: 10.1145/2523813. DOI: https://doi.org/10.1145/2523813
[34] E. Caveness et al., “TensorFlow Data Validation: Data analysis and validation in continuous ML pipelines,” in Proc. ACM SIGMOD, 2020, pp. 2793–2796, doi: 10.1145/3318464.3384707. DOI: https://doi.org/10.1145/3318464.3384707
[35] S. Schelter, D. Lange, P. Schmidt, M. Celikel, F. Biessmann, and A. Grafberger, “Automating large-scale data quality verification,” Proc. VLDB Endow., vol. 11, no. 12, pp. 1781–1794, 2018, doi: 10.14778/3229863.3229867. DOI: https://doi.org/10.14778/3229863.3229867
[36] C. Wohlin et al., Experimentation in Software Engineering. Berlin, Germany: Springer, 2012, doi: 10.1007/978-3-642-29044-2. DOI: https://doi.org/10.1007/978-3-642-29044-2
[37] R. Gollapudi, “Operational drift and risk-bounded decision-making in production database systems,” J. Int. Crisis Risk Commun. Res., vol. 6, no. S3, pp. 132–147, 2023, doi: 10.63278/jicrcr.vi.3762. DOI: https://doi.org/10.63278/jicrcr.vi.3762
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2024 International Journal of Computational and Experimental Science and Engineering
This work is licensed under a Creative Commons Attribution 4.0 International License.
