Architecture¶

Brief, current, precise. A PR that changes the structure described here updates this file in the same PR. The language is docs/SPEC.md; what may enter it is docs/design/ceiling.md; plans and refusals are docs/ROADMAP.md; measured results are docs/benchmarks.md, produced by the harness in bench/ — which is also how a claim here gets falsified.

python examples/walkthrough.py executes the pipeline below stage by stage and prints what each one produces — the same public calls lps.solve makes, so the demonstration cannot drift from the code. Its output is committed as examples/walkthrough.out and asserted line for line (tests/test_walkthrough.py), so reading it is the same as running it — and a stage that starts telling a different story shows up as a diff in that file.

Thesis¶

A YAML math spec is a closed AST known before any data is touched. That one property makes everything else legal: the whole model can be compiled — to eager xarray/linopy calls, or to a logical plan executed relationally under a fixed memory budget — with both paths provably meaning the same thing. Every rule below protects it.

Four directories, four fences. One produces the AST; three consume it and know nothing of each other. Each box below is a directory, and its subtitle is the import rule tests/test_architecture.py enforces off the path — so a module cannot step over a fence by being spelled differently.

flowchart TB
    Y[YAML file] --> LANG

    subgraph LANG["language/ — imports nothing but errors.py"]
        direction TB
        SCHEMA["_yaml.py → schema.py<br/>YAML 1.2, duplicate keys refused"]
        SCHEMA --> EXPAND["expansion.py · piecewise.py<br/>macros, expressions, formulations<br/>— no consumer sees any of them"]
        EXPAND --> RESOLVE["resolution.py · dimensions.py · degree.py<br/>names → typed nodes, dim sets, degree 1"]
    end

    LANG --> AST["core AST — the narrow waist<br/>fully resolved: names typed, dims checked, degree judged<br/>closed from both sides"]

    AST --> LOWER
    AST --> WALK
    AST -.->|"opt-in: lpspec.linopy"| BUILD

    subgraph REL["relational/ — imports nothing from the package but errors.py"]
        direction TB
        LOWER["lowering.py<br/><i>(flat: reads AST, writes plan)</i>"] --> PLAN["plan.py<br/>frozen logical plan"]
        PLAN --> COMP["compiler.py<br/>plan → lazy frames · reads nothing"]
        DR[("data<br/>parquet / any Arrow table<br/><i>sources.py, flat</i>")] --> BIND["binding.py<br/>→ BoundSources, frozen"]
        BIND --> EXEC
        COMP --> EXEC["executor.py + labels.py<br/>assemble the model frames"]
        EXEC --> TABLES["sinks/tables.py<br/>cols · obj · rows · A"]
        TABLES --> LPS["sinks/lp_file.py<br/>(mps planned)"]
        TABLES --> DIRECT["sinks/highs.py<br/>COO batches → HiGHS"]
        DIRECT --> SOL["result.py<br/>label join, never dense"]
    end

    subgraph TS["typeset/ — reaches the language and nothing else"]
        direction TB
        WALK["walk.py — one walk of the AST"] --> FMT["latex · typst · markdown<br/>one spelling table each"]
    end

    subgraph EAGER["linopy/ — the ONLY code importing linopy or xarray"]
        direction TB
        LOAD["loader.py<br/>data → xr.Dataset"] --> BUILD["builder.py<br/>evaluate AST"]
        BUILD --> MODEL[linopy.Model] --> SOLVE["linopy solve / writers"]
    end

    LOWER -->|"outside the plan:<br/>LanguageError naming the construct"| ERR["load error<br/>(no fallback)"]

    classDef laneL fill:#fdf6ec,stroke:#b7791f,stroke-width:2px,color:#111
    classDef laneR fill:#f0f7f0,stroke:#3a7d44,stroke-width:2px,color:#111
    classDef laneE fill:#eef1fb,stroke:#4a5fc1,stroke-width:2px,color:#111
    classDef laneT fill:#f7f0f7,stroke:#8b3a7d,stroke-width:2px,color:#111
    classDef waist fill:#e9edfa,stroke:#4a5fc1,stroke-width:3px,color:#111
    class LANG laneL
    class REL laneR
    class EAGER laneE
    class TS laneT
    class AST waist

Only six modules sit outside a fence, and each is legitimately both halves: lowering.py reads the AST and writes the plan, sources.py binds data to a validated schema, api.py runs the lot, errors.py is the leaf every fence points at, and __main__.py / _notes.py are plumbing. That is a category, not a leftovers bin — see What counts as language.

Eligibility is decided by attempting the lowering — lower_program returns a Program or raises lps.LanguageError — so it cannot drift from what the engine supports. Errors split model from run: everything under LanguageError is decidable without data, DataError is what a source failed to supply, and both are LinopyYamlError (errors.py). lps.check() is exactly parse → expand → validate → lower with no data bound, so a model repository can compile-check its math in CI. Expansion precedes validation in both lanes, because a formulation emits declarations and those are language too — a stray dim in generated math is the same error as a stray dim in a written one.

One contract, many consumers¶

The AST is a narrow waist. Everything upstream emits it, everything downstream reads it, and nothing else has to agree on anything — so the model you write once is the same model that gets checked, solved, typeset and read back.

flowchart TB
    Y(["your math, written once — one YAML file"])

    Y --> AST

    AST["the whole model, typed and checked<br/>before a byte of data is read<br/><b>lps.load_schema → MathSchema</b>"]

    AST --> SHOW
    AST --> CHECK
    AST --> RUN

    subgraph SHOW["show it — no data, no solver"]
        direction TB
        S1(["typeset the math for a paper or a review<br/><b>lps.to_latex · to_typst · to_markdown</b><br/>how names print: <b>SymbolTable</b>"])
        S2(["drive it from the command line<br/><b>python -m lpspec &lt;format&gt;</b>"])
        S3(["watch what a build is doing"])
    end

    subgraph CHECK["check it — no data, no solver"]
        direction TB
        C1(["will this build, and is the math sayable?<br/><b>lps.check</b>"])
        C2(["do the dimensions line up?<br/><b>lps.check</b> — same pass, same answer"])
        C3(["will that solver take it, and how big is it?"])
    end

    subgraph RUN["run it — the only part that touches your data"]
        direction TB
        R1(["stream it straight into a solver<br/><b>lps.solve</b> · <b>lps.build</b> for several sinks"])
        R2(["write an LP file for anything else<br/><b>lps.write</b>"])
        R3(["put the same math on a linopy model<br/><b>lpspec.linopy.build · .extend</b>"])
    end

    R1 --> ANS

    subgraph ANS["your answers, as tables you can join"]
        direction TB
        A1(["values and shadow prices, by name<br/><b>result.objective · .primal · .dual</b><br/>bridges out: <b>.to_pandas · .to_parquet</b>"])
        A2(["derived results and diagnostics"])
        A3(["change a number, re-solve, keep the labels"])
    end

    classDef built fill:#eef6ee,stroke:#3a7d44,stroke-width:1.5px,color:#111
    classDef plan fill:#fdf4e8,stroke:#b7791f,stroke-width:1.5px,stroke-dasharray:5 4,color:#111
    classDef waist fill:#e9edfa,stroke:#4a5fc1,stroke-width:3px,color:#111
    classDef fam fill:#fcfcfb,stroke:#c9c5be,color:#111
    class Y,R1,R2,R3,C1,C2,A1,S1,S2 built
    class C3,S3,A2,A3 plan
    class AST waist
    class RUN,CHECK,SHOW,ANS fam

Solid is what ships today; dashed is what the shape makes cheap — which is also to say the solid boxes are the ones with a name in them. That is the whole Python surface bar the error hierarchy, which is cross-cutting rather than a box; the table below is the same set, listed. None of the dashed boxes is a rewrite — each reads the same AST the engine reads, so a renderer is a tree walk, a check is a pass with no data bound, and a new output format is one function in relational/sinks/. typeset/ is that claim cashed: a spike that typesets any model the lanes can build, in one walk of the resolved AST, holding no opinion the lanes do not already hold — including a piecewise: block, which prints as the λ-formulation it expands to rather than as the sugar it was written as. How names print is the one thing it does not read off the model: a symbol table is presentation, so it is a sidecar file (examples/symbols/) rather than keys on MathSchema, and a model with no table still renders. It splits the way relational/sinks/ does — one walk over the AST, one module per output format — so a format is a spelling table rather than a second walk that could disagree about what the model says. python -m lpspec <format> is the shell front for it, one verb per entry in typeset.FORMATS rather than a list that could fall behind: a consumer that needs no data needs no runner.

That last sentence was prose for a while, and false: typeset/ reached api.load_schema, so rendering a model imported the executor, the plan and polars. Nothing failed, because it was the one fence of the four with no test behind it. It has two now — a path-scoped import rule like the other three, and a check on the transitive closure, since a renderer that imports only language/ still pays for polars if some language module does. Two properties carry the rest: data enters at exactly one place, which is why checking a model costs seconds and needs nothing but the file; and the waist is closed, which is what the ceiling in docs/design/ceiling.md protects — a new consumer is free, a new primitive is taxed. What is planned, and why, is docs/ROADMAP.md.

The Python surface¶

Sixteen names, and that is the feature. The model is the YAML file; Python is how you run it, so the surface is the diagram above written out — one group per box, nothing that constructs math, nothing that reaches the plan.

	`lpspec.`
run it	`solve` · `build` · `write` · `check`	the four verbs; `check` binds no data
load it	`load_schema` · `MathSchema`	parse and validate; the schema is the contract under the YAML
show it	`to_latex` · `to_typst` · `to_markdown` · `SymbolTable`	one per format, plus how names print
catch it	`LinopyYamlError` · `LanguageError` · `DataError` · `DimensionError` · `SchemaError` · `PiecewiseExpansionError`	one hierarchy, split model from run

Plus two on the opt-in shim — lpspec.linopy.build and .extend — and the methods of what solve hands back (Result: objective, primal, dual, to_pandas / to_dataarray / to_parquet, the status pair, close).

tests/test_architecture.py pins all of it: __all__ must match the table, and no public non-module attribute may exist outside it. Both directions, because either alone rots — the first catches a name documented and never exported, the second a helper that leaked into the namespace by being imported at the top of __init__.py. That check found one the day it was written.

There is deliberately no Python API for constructing a model, no way to hand in a plan, and no registry to populate. That is hard rule 5 below, and it is what makes a .yaml file the thing you review, diff and cite — rather than the serialisation of a Python object you would have to run to understand.

Hard rules¶

Enforced, not aspirational: tests/test_architecture.py encodes these as static checks and CI's bare-install job proves the dependency claims.

These rules constrain the language. What a construct may say, which layer may know what, and what a file means on its own — each survives any engine, and each decides what can enter docs/SPEC.md. How much a build costs is a property of the engine, measured in docs/benchmarks.md and not a rule. It was one once, phrased around a memory_limit that only one engine had, and that made an implementation choice load-bearing in the language's rulebook.

The layers are ordered, and imports prove it. Every module imports only downward, at module level, with no exception at all: DELIBERATE_LAZY_IMPORTS in tests/test_architecture.py is empty, and an undeclared in-function import fails the build. It held one entry until recently — lowering.py reached piecewise.py lazily, because a formulation expands before lowering while its link check asked lowering for a verdict. That check was really about degree, which is language; asked of language/degree.py instead, the cycle is gone rather than deferred, and a lazy import is once again only ever a leftover.
Core AST is the whole language. Both backends consume only core AST; macros, named expressions and piecewise: are expanded away before dispatch, and the plan/query/xarray are backend-private. The AST crossing that seam is fully resolved — names are typed Variable/Parameter/Dimension nodes — so a backend cannot hold its own opinion about what a name refers to. Resolving independently is how the two lanes silently disagreed about scoping before. The waist is closed from the front too: nothing under src/lpspec/language/ imports lowering, sources, api or any consuming subpackage, so what a model means cannot depend on what is done with it. That is the mirror of rule 2 and it is enforced the same way, off the path (LANGUAGE_MAY_IMPORT). load_schema sits inside that fence for the same reason: parsing and validating a model is the language's own job, and a consumer that binds no data has to reach it without reaching a runner. api.py re-exports it, so callers keep saying lps.load_schema.
The engine knows nothing about linopy, xarray or YAML. src/lpspec/relational/ goes polars → highspy → solver, with linopy's semantics as a spec to match rather than code to share; it never sees the schema, the AST, or the eager builder. Engine-internal naming encodes neither "polars" nor "yaml". Enforced more strictly than stated — the engine imports nothing from the package at all, bar declared dependency-free leaves (errors.py today, listed in ENGINE_MAY_IMPORT) — because a near-zero import surface is what keeps the subpackage extractable. Widening that list is a decision, not an accident.
One language, two lanes — not fast-vs-slow versions of each other. The streaming engine builds models declared in YAML; the linopy lane attaches YAML math to a linopy.Model already in memory, which is structurally eager. Both accept exactly the same language, and no helper registry exists that could create a divergence — that equality is what makes the differential tests an oracle rather than a comparison of dialects. A construct outside the language is a load error naming the construct and its rewrite, never a redirection to the other lane.
Backend-visible YAML files are self-contained. No Python-side state (registries, session objects) may change what a file means.
The public interface is a declared model, not a Python API. YAML is the format we ship and document; the contract underneath it is MathSchema, and whether that seam is ever blessed is open (see Composition). The Python surface is the runner (api.py); the plan is internal, and a stable plan-construction API is a later possibility, not a current contract. The whole of it is sixteen names, pinned by a test — so the surface grows through a list a reviewer reads, the way every other fence here does.

The relational lane¶

The spine is one module per box above. binding.py turns a caller's sources into the frames the engine reads; compiler.py turns plan nodes into lazy frames and reads nothing; executor.py fills the model frames; sinks/ drains them. Two more sit beside the executor rather than inside it, because each answers a question the executor merely uses: labels.py decides which coordinate gets which solver index, and result.py is what a caller reads a solve back through. The remaining four are not on the spine and the diagram does not draw them — plan.py is the vocabulary the spine speaks, frames.py and status.py are the two boundaries (a caller's table in, a solver's verdict out), and chunking.py and data_validation.py are single rules lifted out of whoever needed them first. The map below is the full list. The split is what makes the admissibility test below something you can perform rather than reason about — build a PolarsCompiler, hand it a node, read .explain() (tests/test_compiler.py does exactly that, over empty frames: a schema is all it takes to compile a query). It is also why a new sink is a function in one file instead of another method on the executor.

What binding produces is a value. BoundSources is frozen — parameters, dimensions, their cardinalities, and which parameters are boolean — because a query is written against data that has stopped changing. The variable frames are passed beside it and stay mutable, since a variable frame appears as its declaration is built and a constraint compiled afterwards has to see it. That is the one live registry in the lane, and keeping it out of the carrier is what makes it visible in a signature rather than only in a docstring.

Tidy tables. Parameters are (dims…, value); a variable frame is (dims…, var_label), one row per existing variable; a linear expression is (frame dims…, var_label, coeff) plus a constant part; constraint rows are (row, sense, rhs); the coefficient matrix is COO (row, col, coeff). Masks are row absence — no NaN sentinels, no -1 labels. Broadcasting is a join, sum drops coordinate columns, group_sum joins the dim table and projects a declared coordinate in place of the grouped dim. Neither aggregates: both rewrite a fragment's dim tuple, and duplicates collapse in the terminal SUM(coeff) GROUP BY row, col at assembly. Labels are dense 0..n-1 by construction, so var_label is the solver column index and row the solver row index — no remapping. That is also why value-only re-solve is cheap and structural editing is out of scope.

Labels are also row-major over the coordinate product, and that is a contract rather than a side effect of how they are computed: it is what makes a build reproducible run to run. Labeller.frame reaches it three ways depending on how much of the product survives the mask — arithmetic, factored, counted — and they must agree integer for integer, because a label is a solver index. That three-way agreement is why labelling is a module (relational/labels.py) and not three methods among twenty: its inputs are stated — the query, the dimension cardinalities, the program — so nothing else about a build can move an index.

The plan is affine-by-design. No node introduces variables or constraints as a side effect of an expression; formulations are model transformations. Variable types are not formulations — binary/integer are a vtype column, LP binary/general sections and HiGHS integrality, which keeps basic MILP inside the streaming lane. Reimplementing linopy's reformulation passes inside the plan is explicitly rejected: that duplicates the library this package consumes.

Labels are the one place order is load-bearing. var_label is the solver's column index and row is its row index, so both are assigned by sorting the masked coordinate product on its dimensions' declared ordinals and numbering the result. Variables and constraint rows are the same operation over different frames and it is written once (Labeller.frame) — twice is how the two would come to disagree about which coordinate gets which index. Everything else is order-free, which is what lets the query planner rearrange it.

That order is also what comes back. primal / dual / to_parquet sort on the label before handing rows over, so a read is row-major over the coordinate product — the same order the LP sink writes. Sorting is stated at the read rather than assumed from the inputs, because a where mask decides which rows of the product survive and a join decides nothing about the order they arrive in; without it two reads of one unchanged result came back differently.

A frame is the boundary in both directions. relational/frames.py recognises a caller's table through the Arrow PyCapsule protocol without importing any dataframe library, and Result.primal hands back a polars.DataFrame, which exports the same protocol. That symmetry is what keeps pandas and pyarrow off the dependency list: they are bridges out (to_pandas, to_dataarray), shipped with the [linopy] extra, not shapes the engine holds. The bare-install CI job runs the suite with neither present.

Sinks are capped, explicitly. Today every sink expresses the same three streams and no more: cols (bounds, objective coefficients, integrality), rows, and A in COO. The upgrade path is two further streams — sos_sets and genconstr — plus a semi-continuous threshold on cols. Unlike the three that exist, those two would land unevenly, because the destinations differ per sink (see "Capability is not the ceiling"); that unevenness is what Track 4 exists to make declared rather than discovered at solve time.

Module map¶

Module	Role
`language/_yaml.py`	the only place a file is read: YAML 1.2 booleans, duplicate keys refused
`language/schema.py`	pydantic schema incl. `expressions:` / `macros:` / `piecewise:`
`language/expression_parser.py`, `language/where_parser.py`	text → core AST; grammar only, dependency-free
`language/expansion.py`	named-expression / macro substitution (pre-dispatch)
`language/resolution.py`	one flat namespace; `NameNode` → typed `Variable`/`Parameter`/`Dimension` nodes
`language/dimensions.py`	static dim-set checking over the resolved AST
`language/degree.py`	degree 1: the ceiling's first clause, asked by both lanes and stated by neither
`language/helpers.py`	the closed set of built-in operators: their names and call shapes — no registry
`language/validation.py`	load-time: parse, expand, resolve, check everything — and `load_schema`, the language's front door
`language/piecewise.py`	`piecewise:` → λ-formulation declarations
`api.py`	the runner: `check` / `build` / `solve` / `write`, linopy-free; re-exports `load_schema`
`typeset/`	spike — resolved AST → LaTeX / Typst / Markdown. A reader, not a lane: no model, no data, no plan (README)
`__main__.py`	`python -m lpspec <format>` — a shell front for the verbs that bind no data
`sources.py`	bind runtime data (parquet paths / in-memory tables) to a validated schema; the `convex:` curvature guard, which is the one check that needs values
`lowering.py`	core AST → logical plan (defines the relational subset)
`errors.py`	the exception hierarchy; the one module either fenced side may import
`relational/plan.py`	frozen logical-plan dataclasses
`relational/frames.py`	the boundary — caller tables in, via the Arrow PyCapsule protocol
`relational/compiler.py`	plan → lazy frames; pure, reads nothing
`relational/chunking.py`	how a batched pass sizes its chunk: budget ÷ the width of one unit
`_notes.py`	attach context to an exception on the way out; no package imports, no opinions
`relational/status.py`	solve outcome on two axes; linopy's vocabulary, copied not imported
`relational/labels.py`	which coordinate gets which solver index; three routes to one number, which must agree
`relational/binding.py`	a caller's sources → `BoundSources`, the frozen frames every query is written against
`relational/executor.py`	assemble the model frames from the bound data
`relational/result.py`	what a solve returned: status, objective, and the label joins that read values back
`relational/data_validation.py`	is the bound data usable — one row per coordinate, labels that exist, single-valued coords
`relational/sinks/tables.py`	what every sink reads and no more — the four frames plus the batching scalars
`relational/sinks/`	how a built model leaves: `lp_file`, `solver_direct` (one module each, README)
`linopy/__init__.py`	opt-in shim: `build` / `extend` on a `linopy.Model`
`linopy/loader.py`	data coercion to `xr.Dataset`, master coords
`linopy/builder.py`	eager backend: core AST → `linopy.Model`
`linopy/semantics.py`	where this lane answers linopy's v1 arithmetic convention — one home, as linopy's own `semantics.py` is

Four subpackages, and the directory is the rule in every case. Everything under language/ produces the AST and may not reach a consumer of it; everything under relational/ is the engine and imports nothing else from the package; everything under linopy/ is the opt-in eager lane and is the only code allowed to import linopy or xarray; everything under typeset/ reads the AST and writes text, and reaches neither the plan nor any data. tests/test_architecture.py reads membership off the path in all four cases, so no fence can be stepped over by naming a file differently.

language/ and relational/ are the two halves of the waist and their fences point the same way — outward, at errors.py, the one leaf both may import. typeset/'s points there too, which is what makes "a new consumer is free" a measurable claim rather than a hope: it is enforced twice, once on the names a renderer imports and once on the transitive closure behind them, because a consumer's real cost is what it drags in, not what it spells.

What counts as language¶

A fence says what may not happen; it does not say what belongs. The test is:

A rule is language iff two consumers answering it separately would be a bug.

Not "is it about syntax", not "does it run early" — would a second opinion be wrong? Every "one implementation each" rule in this file is that test applied: names resolve once (resolution.py) because two lanes resolving separately disagreed three ways; the helper set is closed (helpers.py) and a test proves both lanes implement exactly it; a primitive's dim rule lives only in dimensions.py and lowering asks for the verdict rather than deciding again; degree lives only in degree.py for the same reason.

That last one was the counter-example that produced the rule. Degree 1 is the first clause of the ceiling and it lived in lowering.py — so the eager lane kept a hand-copy of one refusal sentence and delegated another to linopy, and one language rule had two spellings with only one of them tested. Nothing about x * y is relational; the ceiling doc says outright that degree is not a property of the plan. It was in a consumer because that is where it was first needed, which is how this kind of drift always starts.

piecewise.py was the same story, one layer up. It sat flat, outside language/, on the grounds that it emitted declarations (language) but consulted lowering.check_core_subset to do it (plan) — so a construct the SPEC documents lived outside the directory that owns the language, a formulation running in every lane enforced what a plan node can represent, and lowering.py had to import it back lazily. One cycle, one misplaced file, one fence crossed, all from a single call.

The call turned out to be about degree. Its two tested cases are p ** 2 and p * p; everything else check_core_subset would have caught — dim rules, call shapes, unknown helpers — resolution and dimensions.py already answer. So degree.check_expression states it where degree lives, the error still names the link the user wrote rather than the _link0 the expansion generates, and piecewise.py moved into language/ where the SPEC always implied it was. The lazy import went with it: DELIBERATE_LAZY_IMPORTS is empty.

The corollary is what the top level is for. A module stays flat when it is legitimately both halves: lowering.py reads the AST and writes the plan, sources.py binds data to a validated schema, api.py runs the lot. That is a real category and it is now a small one — a flat module should be arguable, and piecewise.py was not.

Applying the test to a consumer's own refusals splits them cleanly. lowering.py still refuses plan shapes — shift(by=) must be an integer literal, and group_sum(by=) a declared coordinate — because those are about what a plan node can represent, and a second opinion about them is not a bug, it is just the other lane's business. What it no longer does is state a rule about the language that another consumer then has to restate, or offer one for a formulation to borrow.

Naming across the layers¶

The same construct passes through three layers, and each names it in full — no abbreviations, so a name never has to be decoded. The layer is the suffix, which is what keeps the three vocabularies from colliding:

Layer	Suffix	Example
YAML block (`language/schema.py`)	`Block`	`VariableBlock`, `PiecewiseBlock`
Core AST (`*_parser.py`)	`Node`	`VariableNode`, `DimensionComparisonNode`
Logical plan (`relational/plan.py`)	none / `Declaration`	`Variable`, `VariableDeclaration`

Two rules follow from that table, and a PR that adds a construct keeps them:

A node names the coordinate map, not a surface spelling. The translation node is Translate, and it stayed that way when the surface collapsed to a single shift(…, edge=): the node is named for what it does to coordinates, so which keyword the language happens to expose does not reach it.
Nothing is abbreviated. Cmp became ParameterComparison, vtype became variable_type. The one place abbreviation survives is frame column names inside the engine, which are not Python identifiers.

Where a concept is already linopy's, use linopy's name¶

For anything this package shares with linopy — solve statuses, result shapes, solver metrics, duals — adopt linopy's primitive: its spelling, its field names, its decomposition. Result is the envelope (status + solution + report) and Solution the raw arrays, because that is what those words mean in linopy; status / termination_condition are two axes and is_ok is the rollup, because that is linopy's model. Our audience arrives from linopy/PyPSA, and a second vocabulary for one fact is a tax on every one of them. It also keeps the oracle honest: the two lanes can be compared exactly rather than through case-folding.

Copy it; do not import it. The engine may not import linopy (rule 2), so the tables live here — and a test imports linopy and asserts the copy still matches (tests/test_solve_status.py). A copy nobody checks is a copy that rots, and the failure should be a red test rather than a user handed two dialects.

This applies to vocabulary we share. Where the design genuinely differs it stays ours: we have no Solution of dense arrays to hold, because the values are read back by joining labels to coordinates.

Extension checklists¶

Add a macro or named expression: edit YAML. Nothing else.

Add a consumer of the AST (a renderer, a checker, a report): a directory beside typeset/, a fence test naming what it may import, and a walk. It reads language.load_schema and stops there — if it needs the plan it is a lane, not a consumer, and the ceiling doc is the conversation to have first.

Add a primitive: grammar (usually free — f(x, k=v) already parses) → signature in helpers.BUILTINS (arity and which arguments name dimensions — resolution, validation and lowering all read it from there, so the shape is declared once) → eager helper → plan node + locality class → executor → lowering case → differential test on both sinks → SPEC §5/§7, and this file if structural.

Three things are deliberately not per-primitive work, because they are one implementation each: a primitive's dim rule lives only in language/dimensions.py — both its dim set and its verdict on an operand that lacks the dim being reduced along, which lowering asks for rather than deciding again — its degree verdict lives only in language/degree.py, which both lanes ask; and the dense-label assignment that gives a coordinate its solver index lives only in relational/labels.py, shared by variables and constraint rows. What a lowering case still owns is what is about the plan: which node the call becomes, and the shapes that node cannot represent.