Backfilling the data/AI jargon I'd scoped out — corpus, vector DB, ETL and 33 more

The owner asked what "corpus" means, then immediately: why wasn't it in the deck? Checking honestly, he'd caught a real scoping miss. "corpus" existed only as a word inside one card's code snippet, not as a term you could look up, and whole common terms — vector database, data lake, data warehouse — weren't anywhere. The first jargon deck had been built around infra, process, and git, and quietly skipped the data and AI-infra category.

So the fix wasn't to add one word, it was to backfill the missing category. Thirty-six native-verified cards: corpus, dataset, vector database, embedding store, data lake, data warehouse, lakehouse, ETL, ELT, data pipeline, ground truth, data labeling, fine-tuning, feature store, schema drift, data contract, golden dataset, data lineage, PII, CDC, OLTP/OLAP, columnar storage, materialized view, denormalization, upsert, deduplication, data skew, cardinality, star schema, batch vs streaming, data residency, RLHF, distillation, quantization, data drift. The pairs carry the contrast that actually trips people up — a lake is raw and schema-on-read while a warehouse is cleaned and schema-on-write; ETL transforms before loading while ELT loads then transforms in-warehouse; OLTP serves the live app while OLAP serves analytics. Integrated into the jargon category, which is now 122 terms.

The reusable lesson: a curated deck has blind spots by construction, and when someone points at one missing item, the right move is to ask whether an entire category was skipped — then backfill the category, not just the word.