A closed list of things that can happen to a value

In most projects, the answer is free text. The mapping spreadsheet has a column called transformation logic. It holds whatever the analyst typed: convert as needed, see lookup tab, handle nulls per Radu's email. Every row is a small essay, and no machine can check an essay.

Our answer is a closed list. A value that crosses from source to target may undergo exactly eight operations. It copies unchanged, changes type, or converts units. It translates through a lookup, joins with another field, or splits into parts. It takes a default when empty. Or it passes through free-form logic.

The first seven are deterministic templates. Given the same input, they produce the same output, every time, and a machine can execute and verify them. The eighth is the escape hatch, and the whole section after next is about its cage.

The reason the list is closed is the reason this series exists. A transformation from a closed grammar can be compiled, run against real rows, and checked. The reconciliation in part 5 is derived from the confirmed map, and that derivation is only possible because every entry in the map is machine-readable. An essay in a spreadsheet cell can be right, but nothing can prove it. A closed verb can be wrong, and the tie-out will catch it.

One of Danubia's columns defeats all seven templates. The legacy system stored a credit limit in a currency that no longer exists, with a divisor buried in an old configuration table. Getting it across takes real logic: arithmetic, a join, a condition. No closed verb covers it.

For that column there is the eighth operation: free-form logic, written out, attached to the mapping. The escape hatch exists because real estates contain cases no grammar anticipates, and a grammar that pretended otherwise would push those cases back into spreadsheets.

But the hatch is caged, permanently, by two rules. Free-form logic is born on the suggested tier and can never confirm itself. No threshold promotes it, no automation ratifies it, and only a person who read the logic can confirm it. And free-form logic never counts as deterministic, however simple it looks, because nobody can verify by name what somebody wrote by hand.

The same cage holds a small set of remediation operations: masking a value, regenerating it, carrying it frozen. Each of those is a design decision about data, and no name similarity can derive a design decision. They are authored by people only.

The reasoning deserves one plain sentence. The moment free-form logic can promote itself, the closed grammar is decoration, because every hard case would arrive pre-approved through the hatch.