a parameter cannot say that a missing row is absent, so a decision supplied as data brings back the rows its mask deleted

[FBumann]

Prompt: Close the issue and open a follow up scoped to the absence question.
And link related PRs

Note

The following content was generated by AI.

What can be improved?

A missing row means two things by kind: a masked variable is absent and takes the row, while a missing parameter row reads as 0 and the row stands. A quantity that moves from decision to data changes meaning. parameters: could take absence: like variables:.

Evidence, proposal, related work

Follow-up to #750, scoped to the absence: question. The null point of #750 is settled by #788: a null or NaN parameter value is refused when the data is attached, so a gap in parameter data is only a missing row.

The law as written

absence.md: "A where: does not set a variable to zero. It leaves the variable unbuilt", and "A missing parameter row is not absence. It reads as the value that contributes nothing: 0 as a coefficient." Each rule holds on its own. The two differ only by the kind of name.

Where it breaks: a decision fixed as data

#748 (spec.fix, for #303) turns a variable into a parameter under the same name. For a variable with where: and absence: undefined, the row p <= cap * avail was not built where cap did not exist. Once cap is a parameter it reads 0 there, and p <= 0 stands. #249 (closed) found this first.

#748 reconstructs the absence from the program:

• a read inside sum, sum(by=) or sum_back is safe, since an absent summand and a 0 one give the same sum;
• a read under a mask that holds every conjunct of the variable's mask is safe;
• an unguarded read directly in a constraint adds the variable's mask to that constraint's where:;
• a read through shift, through at, or in an unguarded case is refused, because the mask cannot be carried to the row there.

Proposal

absence: zero | undefined on parameters:, with the values and meaning it has on variables::

• zero, the default, is today's law: a missing row reads 0.
• undefined: a missing row is absent. It spreads through arithmetic, shift and at, takes the row, and is one summand fewer inside a summing operator, as a masked variable is.

With it, fix copies the variable's absence, and every read keeps its meaning with no rewrite and no refusal. #748 shrinks to the dtype rule, the bounds assumption and the rows that become assumptions.

Related

Ref	State	Relation
#750	closed	the parent issue; this one takes its main proposal
#788	merged	a gap arrives only as no row, so absence: has one form to decide
#748	draft PR	spec.fix; shrinks if this lands
#303	open	the need fix answers
#749	draft PR	spec.sensitivity; meets the reverse case, a deleted dual that should count as 0. A read keyword like shift's edge=0 would answer it, and waits for a second caller
#296, #671	open, PR open	coverage: total | masked says whether a table may have gaps; absence: says what a gap reads as. Orthogonal
#773	open	default: fills a missing row. It interacts with absence: undefined: a default and an absent gap cannot both apply
#249	closed PR	the first fix, with a coverage key main no longer has
#259	open	sum(..., where=). A separate need
#13	declined	parameters, variables and expressions stay three declarations

Open

• The key's name: absence:, as on variables, or a word that reads better on data.
• The engine rule: a consumer drops the row where an undefined parameter has no value, instead of reading 0. specsolve needs that rule.
• How absence: undefined and default: (a parameter cannot declare a default, so a bound that data prep fills no longer tells a where: which rows the modeller gave #773) combine on one parameter: refuse both, or let the default win.