Before Juno can map a column or understand a question, it needs to know what your estate holds.
Two steps read the data description, two add language, and the last keeps it current.
Registration recorded the description of each table: the names, the data types, the keys, the row count and a sample of values. The index starts there, with one entry per table and one per column.
Juno then profiles each column from the sample: the pattern the values follow, the range they cover and how many distinct values there are. A column of ten-digit numbers that joins to the supplier table looks very different from a column of dates.
Names come next. Technical names are expanded into business words from a dictionary of common systems and abbreviations, so MENGE becomes “quantity” and LGORT becomes “storage location”. Your own aliases and synonyms join the same entry when you teach them.
The result is published as one searchable record. It refreshes when a sync scan adopts a change, when a training run adds vocabulary, and when your team confirms or corrects a mapping.
Three records of the same estate: its structure, its values and its language.
The structure: table names, column names, data types, lengths, keys and the joins between tables. This is what the mapping stage compares with the ontology.
The pattern, the range and the distinct values in each column, read from the sample. This is what recognises a supplier number, a plant code or a location.
The same tables and columns in the words your people use, with synonyms, acronyms and the aliases your team teaches. This is what a question is matched against.
The index holds the description and a sample. The rows stay in the source system. Product.
A value a person types is searched across every indexed column, with the shape of the values as the evidence.
| Column that carries this value | Distinct values | Pattern | Rows that match | Ontology attribute |
|---|---|---|---|---|
| Warehouse system · STOCK.LOCATION_ID | 1,284 | L + five digits | 3,912 | Location · location code |
| SAP · LIPS.LGORT_EXT | 1,284 | L + five digits | 1,204 | Location · location code |
| Supplier portal · SHIP_TO_LOC | 860 | L + five digits | 112 | Location · location code |
| SAP · MARD.LGORT | 46 | four characters | 0 | Storage location · code |
One value, looked up across every indexed column. Showcase estate.
A person typing “how much stock sits at L04542” uses a value, so Juno searches the value index for it. Three columns in three systems carry values of that shape, and all three map to the location code of the ontology. The storage location column has a different shape and a different meaning, so it stays out of the answer.
The same lookup works for a supplier name, a material number, a plant or a customer. It is the step that turns a value a person knows into the business object Juno holds, and it runs before a query is written.
The index is a living record, so a mapping proposed today rests on what your estate holds today.
Showcase estate.
A. Novak registered 23 tables by 10:07 on Monday. The index run that followed read every column, profiled the values in each one and expanded the technical names. It found 9,340 distinct supplier numbers in one SAP column and the same ten-digit shape in the Oracle supplier column of the acquired division.
That record is what made the rest of the week quick. Mapping proposals arrived at 08:30 on Tuesday with the value evidence behind them. When L. Tan asked her question on Wednesday, “NCNR” and the plant code in it were matched against the same index, and the query went to the right tables at the first attempt.
How Juno recognises that a supplier in SAP, in Oracle and on the portal is one supplier, and how much of the data agrees.
It holds the description of each table and what a sample of values looks like: the pattern, the range and the distinct count. The rows stay in the source system and are read when a question or a rule needs them.
Enough to recognise the shape of a column, and it follows the access rights of the connection. Your team sets the size per data set, and a column your policy keeps out of the sample is described by its name and type alone.
It is indexed by name and type, and the value profile records that the values vary. Columns like that are usually mapped by a person, and your aliases carry most of the meaning.
Yes. The catalogue screen searches it by table, by column, by value and by business word, which is how a data lead answers where a field lives and who owns it.
Each one builds on the one before it.
Reads your rows in place.
Connect and move ↗One record per real thing.
Master and govern ↗Apps for every decision.
Operate and decide ↗Asks and acts, at your level.
Ask and act ↗Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.