Glossary
Three names where useful: the warehouse word / the analytics word / the maths word.
| Term | Meaning | Day |
|---|---|---|
| Additive / semi-additive / non-additive | Can be summed over all dims / over some dims but not time / never (ratios) — extensive vs intensive quantities | 29, 34 |
| ADR, RevPAR, Occupancy | Room revenue ÷ nights sold; room revenue ÷ rooms available; nights sold ÷ rooms available | 21 |
| Asset (Dagster) | A piece of data the orchestrator knows how to produce — a function value in a DAG | 16 |
| Asset check | A test attached to an asset; can block downstream | 20 |
| Backfill | Recomputing past partitions | 18 |
| Bridge / reconciliation | Step-by-step explanation of the difference between two totals | 41 |
| Business date | The trading day a transaction belongs to, after the cut-off shift — phase-shifted calendar | 10 |
| Bus matrix | Processes × conformed dimensions grid | 29 |
| CDC | Change data capture from a database's log | 6 |
| Columnstore | Storage by column with compression and segment elimination | 9, 26 |
| Conformed dimension | A dimension shared with the same keys across facts | 29 |
| Contract | Machine-checked promise about a source's schema and values | 8 |
| Context transition | DAX turning a row context into a filter context inside CALCULATE/iterators | 37 |
| Data diff | Symmetric difference of two versions of a table — A △ B | 44 |
| Delete-insert | Replace a partition's rows in one transaction | 24 |
| Error budget | Allowed misses under an SLO | 43 |
| Fan-out | Row multiplication from joining to a finer grain | 29 |
| Freshness | How recent the latest data is vs a deadline | 13, 20 |
| Grain | What one row means — unit of observation — the index set of a function | 8, 29 |
| Idempotent | Running twice equals running once — f(f(x)) = f(x) | 3, 15 |
| Incremental model | Processes only new/changed data plus a look-back | 32, 47 |
| IO manager | Dagster component that stores and loads asset values | 17 |
| Look-back window | Re-pulling the last N days because they still change | 6 |
| MAD / robust z | Median absolute deviation; outlier-resistant scale | 13 |
| Manifest | Ledger of every file landed with hash and metadata | 3 |
| Medallion / layers | Raw (bronze), staging (silver), marts (gold) | 1 |
| MERGE / upsert | Insert new keys, update changed ones | 24 |
| Menu engineering | Popularity × contribution margin classification (Stars, Plowhorses, Puzzles, Dogs) | 34 |
| OTB, pickup, pace | On-the-books room nights; change over a period; vs same lead time last year | 21 |
| Partition | An independently processed slice, usually by date — block of a block-diagonal system | 18 |
| PBIP | Power BI project format: text files for git | 36 |
| Quarantine | Table of rejected rows with reasons, replayable | 13 |
| Query folding | Power Query steps translated to source SQL | 36 |
| RCSI | Read committed snapshot isolation: readers don't block on writers | 26 |
| Resource (Dagster) | Injected connection/config object | 17 |
| RLS | Row-level security in the database or Power BI | 27, 39 |
| ROAS, CPA, CTR | Value ÷ spend; spend ÷ conversions; clicks ÷ impressions | 28 |
| Run key | Sensor deduplication key for run requests | 19 |
| Same-store sales | Growth over locations open in both periods | 34 |
| SCD Type 1/2/3 | Overwrite / new versioned row / previous-value column | 30 |
| Semantic layer | Single store of metric definitions for all tools | 34 |
| Sensor | Function that watches for events and requests runs | 19 |
| SLO | Service level objective, e.g. sent by 08:00 on 98% of days | 43 |
| Surrogate key | Meaningless warehouse key for a dimension row | 12, 30 |
| Watermark | Highest key safely processed; next run starts there minus overlap | 4, 24 |