Crate improv_core_model 

Improv core model: categories, items, measures, coordinates, formulas.

GUI-free, storage-free. This is the multidimensional “cube”: each measure is a tensor indexed by a subset of categories; a coordinate names one cell.

Structure follows the project steering doc (AGENT_STEERING.md / IMPROV.txt).

Re-exports

pub use extfn_def::ExternalFn;

pub use extfn_def::Language;

pub use formula::BinaryOp;

pub use formula::DimensionSpec;

pub use formula::Expr;

pub use formula::Formula;

pub use formula::FuncId;

pub use formula::UnaryOp;

pub use ids::CategoryId;

pub use ids::ItemId;

pub use ids::MeasureId;

pub use ids::Name;

pub use ids::ScenarioId;

pub use ids::ViewId;

pub use locale::Locale;

pub use parser::parse_definition;

pub use parser::parse_expr;

pub use parser::parse_formula;

pub use parser::Definition;

pub use parser::FormulaText;

pub use parser::LambdaText;

pub use parser::ParseError;

pub use value::Value;

pub use value::ValueError;

pub use value::ValueErrorKind;

pub use value::ValueType;

Modules

extfn_def

Definitions of external-language functions (Phase 6), as plain, serializable model data. The runtime that evaluates these lives in the improv_extfn crate; this module holds only the declaration so a Model can persist which functions a formula may CALL and how to type-check them.

formula

The dimension-aware formula AST.

functions

The built-in function registry: ONE table for every function the formula language knows, read by the parser (names, aliases, arity), the compiler (argument and result types, kind), the GUI and CNL (spellings), and the generated function reference (docs/src/functions.md).

ids

Opaque, stable identifiers and human-readable names.

locale

Locale-aware formula entry: the separators and translated function names a locale’s users type, mapped onto the one canonical, locale-independent formula language before parsing (. decimal, , argument separator, English names, as formulas are stored).

parser

Textual formula parser: "Revenue = Price * Quantity" -> Formula.

schedule

Pure refresh-scheduling decisions: given each external-sourced measure’s RefreshPolicy and when it last ran, decide which measures are due to refresh now. This is intentionally I/O-free and clock-free (times are plain seconds) so a daemon can drive it and tests can exercise it deterministically. The daemon (CLI serve-refresh) does the actual query/eval; this only picks what and when.

value

Value types and runtime values, including error values that propagate through the dataflow graph.

Structs

CanvasRect

Where a matrix sits on a view’s canvas: position of its top-left corner and its size, in abstract layout units (the GUI treats them as egui logical points; nothing here depends on egui — core_model is GUI-free).

Category

A dimension of the model (e.g. Time, Product, Region).

Coordinate

A coordinate maps categories to items: one cell of a measure’s tensor.

ExternalCall

A measure defined as func(arg_measures...) where func is a registered external function. Evaluated host-side per coordinate.

Filter

Restrict a category to a subset of its items in a view.

Item

A member of a category (e.g. 2025, “Widget A”).

MatrixPlacement

One matrix on a view’s canvas: a measure, that matrix’s own pivot layout, and where it sits (rect). A view holds several of these — Quantrix’s free-form canvas, where each matrix is independently pivoted (see docs/reviews/2026-09-22-gui-reconstruction-plan.md Step 3).

Measure

A named variable defined over one or more categories, e.g. Revenue[Time, Product].

Model

The whole model: the multidimensional cube plus its raw input data.

NamedLambda

A named LAMBDA: its parameters and its body, in which parameter i is the reference functions::lambda_param(i), or, where the body uses it as a category (MAX(x OVER c)), the category functions::lambda_category(i). A body reads only its parameters (no measures), so a call depends on just its arguments; it may call lambdas defined before it, never itself.

Scenario

A what-if scenario: a named overlay of input-cell overrides on the base model. Applying it produces a model whose inputs are the base inputs with these overrides taking precedence; derived measures then recompute normally.

SqlSource

How to refresh an SQL-backed input measure: the query and how its columns map to the measure’s dimensions + value. A live-query measure marked here is nondeterministic (its data comes from outside); determinism tests treat only measures WITHOUT an entry here as pure.

View

A saved canvas layout: one primary matrix described by the flat fields below, plus any number of placements (additional matrices), each with its own measure, pivot and geometry.

Enums

LambdaParam

How a named LAMBDA’s body uses a parameter, which says what a call passes for it (NamedLambda::param_kinds).

MeasureKind

A measure is either raw input data or derived from a formula.

RefreshPolicy

When an external-sourced measure (SQL or external-function) should refresh.