A number on a fintech screen looks as if it was found, the way a stone is found. Most of the time it was made. Someone defined a field, someone or something wrote a value, a clock attached a time, and a later view selected that cell to show. The birthplace is the subject of this note. The use of the number is not.
The desk treats a data point as a small manufactured object. That keeps the lesson inside description. It does not become a claim that the point is reliable enough to act on, and it does not become a claim that it should be ignored. Reliability is a separate question, and this page does not score it.
A point has a birthplace
Birthplaces are ordinary. A person types into a form. A terminal writes a code. A report is copied from a public table. A model emits a label after reading other fields. Each birthplace leaves a different kind of trace, and some leave almost none. The absence of a trace is itself information about the file, not a hole the reader is obliged to fill.
Ask three documentary questions. What is the field called in the source, not on the screen? What unit and what format does the source declare? What time does the source attach, and is that time the event or the moment of recording? If the source does not say, the honest sentence is that the source does not say.
Public statistical tables and technical schemas are the right kind of source for this habit. They show definitions. They are not a partnership, and they are not a dataset this desk claims to maintain. We do not publish original figures, because a lesson about birthplaces does not need a private number.
- Source
- The document or system that first states the field. Without it, the point is an orphan.
- Field
- A named place in a record. The name can differ from the label a screen later chooses.
- Gap
- A cell the format allows and the file leaves empty. Empty is a value of a kind. It is not a request to guess.
The file is not the event
A purchase, a wage, a late bill, a change of address: these are events in a life or a firm. A file holds signs that someone mapped onto those events. The mapping can be late, partial, or wrong, and the file will still be internally neat. Neatness is a property of the table, not a property of the world.
This is the usual confusion in product language. “The data shows” often means “the field contains.” Those are different sentences. The first borrows authority from the world. The second stays with the record. This note uses the second.
When a later system reads the file, it inherits the mapping. It does not recover the event by being more complex. A model can restate a field in a new sentence. The restatement is a new text. If the source cell was empty, a fluent sentence does not make it full. Fluency is a style, not a source.
For the same reason, two files that use one label — “customer,” “merchant,” “balance” — are not automatically about the same object. The label is cheap. The definition, if it exists, is in a schema or a methodology note. Read that, or say it is missing.
Missing fields stay missing
Product interfaces dislike blanks. They substitute a dash, a zero, or a phrase like “not available,” and sometimes they drop the row. Each substitute is a choice. A dash is not the same as a zero. A dropped row is not the same as a row that says the value was never collected. The lesson is to restore the distinction when you read, even if the screen has smoothed it.
Guessing a missing cell from neighbouring cells is a method with a name, and it produces a new point, not the old one. This desk does not demonstrate that method and does not recommend it. Naming it is enough: an imputed value is authored. It should not wear the clothes of the source.
There is no neutral “complete” file in this lesson. Completeness is relative to a stated list of fields. If the list is not stated, completeness cannot be claimed. A reader can still describe what is present. Presence does not require a percentage, and this page does not invent one.
- A zero can be a real measure or a stand-in for empty. The source has to say which.
- A timestamp can mark the event or the upload. They are different clocks.
- A renamed column is a new label. Keep the old name beside it when you can.
Reading the edge of a table
The edge of a table is where the definitions run out: footnotes, excluded categories, revised vintages, broken series. Public data releases often put the caution there, in smaller type. Moving the caution into the same sentence as the figure is a reading skill. It is not a way to produce a better figure.
Fintech screens rarely show that edge. They show the cell that fits the layout. The note’s job is to remember that the layout is a crop. What was cropped may be the definition. Without the definition, the cell is a mark whose birthplace is unknown.
Stop at the birthplace. Do not continue into whether the point should change a person’s saving, spending, or choice of firm. Those questions need a situation this page does not have, and a permission this page does not claim. The educational end is the ability to say where a point came from, or to say that the origin is not in the file.
Open question: if the cell were blank in the source, what did the screen put in its place?