Every stratum has its own null

In a toy meter-report game, an operator sends one fixed-width reading per active meter each day. A parser stores the reports, an API serves them, and a console shows the result. The game is invented, but the boundary problem is familiar: each layer has a convenient way to say nothing, and those ways do not necessarily mean the same thing.

Working model. An empty representation only becomes a useful fact when its layer and contract are known. Before translating it, ask what happened and what the next person should do.

What the layer shows	What happened in this toy contract	Next action
No input record for an active meter	The report never arrived.	Check delivery before asking about the reading.
A record with spaces in the reading field	The report arrived, but its sender left the field unanswered.	Return the report for completion.
A stored NULL with state = unverified	A raw value arrived, but verification has not accepted it as a reading.	Hold billing and check the retained raw value.
A stored NULL with state = retired	The meter is no longer expected to report.	Exclude it from the daily chase.

The last two rows deliberately share a SQL NULL. The accompanying state makes the reason legible. SQL alone cannot tell the operator whether to investigate or leave the meter alone. An API that omits the reading property or a console that prints — for every row can lose the distinction again. The consumer then has to guess from absence, or treat a retired meter as a late one.

This is where the apparently fussy distinction pays rent. A nightly check should count missing reports for active meters, not all rows without a numeric value. A bill should wait for verification, not turn an unanswered field into zero. The boundary can collapse states only after the consuming decision no longer needs them.

The model may be too elaborate for a system whose only operation is to ignore non-numeric readings. I have not implemented this toy pipeline. The next test is four fixture rows, one for each condition, passed through the parser, API, and console. If each still produces the intended operator action, the distinctions survived. If the UI shows four identical dashes and offers one action, the loss has merely moved to the last layer.

Use this fictional meter-report pipeline as a worked example of how an empty representation acquires meaning at a boundary. Here, no record for an active meter means a delivery gap; spaces in a received record mean unanswered input; a SQL NULL with unverified state means a retained raw value has not passed validation; and a SQL NULL with retired state means no report is due. Those conditions require different operator actions. Preserve the distinctions until the next decision no longer needs them. An API omission or a shared console dash can erase them after storage got them right. In a layered system you know, trace one field from source to display. Identify each contract, who assigns its meaning, and the first point where two states collapse. Test whether that collapse changes an alert, calculation, eligibility check, or recovery action. Challenge this model where downstream actions are genuinely identical or extra state causes more errors. Produce a compact state-to-action table and one fixture test that would catch a harmful collapse.