torchcell.paper

Code behind the manuscript’s generated artifacts. tables holds primitives that emit one table as both markdown and LaTeX; ontology_graph introspects the pydantic schema into a typed graph and ontology_svg lays it out as the zoomable SVG published at /ontology/; signal is a command-line tool that computes a built dataset’s gzip signal and its derived shape and graph role.

`ontology_graph`

Introspect the torchcell pydantic schema into a typed, renderable graph.

Classes 

`ontology_graph.OntologyField`	One pydantic field on a class, as it should appear in a diagram row.
`ontology_graph.OntologyClass`	A pydantic model or enum in the schema, plus its place in the ontology.
`ontology_graph.CompositionEdge`	A `has-a` edge: `source.field` holds one or more `target`.
`ontology_graph.OntologyGraph`	The whole schema as classes plus inheritance and composition edges.

Functions 

ontology_graph.build_ontology_graph

Introspect the live schema modules into an OntologyGraph.

`ontology_svg`

Lay out and render torchcell.paper.ontology_graph as a zoomable SVG map.

Classes 

`ontology_svg.Card`	A placed class box.
`ontology_svg.Lane`	A titled domain block containing one or more laid-out trees.
`ontology_svg.Layout`	A fully placed diagram, ready to serialise.

Functions 

`ontology_svg.build_layout`	Place every class into its lane, then pack lanes into the grid.
`ontology_svg.render_schematic_svg`	Render the legible, Nature-compliant structural panel.
`ontology_svg.render_svg`	Serialise a placed layout to one self-contained SVG document.

`signal`

CLI to compute a dataset’s gzip ‘signal’ + derived shape/graph-role from its built LMDB. Use as new datasets are added, without touching the paper table.

Functions 

signal.resolve_lmdb

Resolve a CLI path to a processed/lmdb dir.

`tables`

Reusable primitives for generating paper tables in BOTH markdown and LaTeX from a single in-code source of truth, plus data-derived columns such as a streaming-gzip “signal” (a Kolmogorov-complexity proxy) computed directly from a built LMDB.

Classes 

`tables.Cell`	A table cell that renders differently in markdown vs LaTeX.
`tables.DatasetSignalRecord`	One dataset's computed row -- the raw-data artifact the DB report emits.
`tables.SignalCache`	JSON-backed memo of `(n, bytes)` per id, keyed by LMDB fingerprint.
`tables.Column`	A table column.
`tables.Row`	One table row.
`tables.PaperTable`	Columns + rows rendered identically to markdown and LaTeX.

Functions 

`tables.human_bytes`	Format a byte count the way a paper table reads: 0.6 KB, 101 KB, 1.9 MB.
`tables.tex_escape`	Escape the LaTeX specials that show up in dataset table cells.
`tables.scientific`	Format a number in scientific notation (unit-free) for a table cell.
`tables.default_phenotype_bytes`	Extract the serialized phenotype from a torchcell LMDB record.
`tables.instance_bytes`	Serialize the full stored INSTANCE (perturbation + environment + phenotype).
`tables.stream_gzip_signal`	Return `(n_records, gzip_bytes)` for an LMDB in a single streaming pass.
`tables.read_first_record`	Return the first record of an LMDB (for schema-derived table columns).
`tables.phenotype_descriptor`	Derive `(shape, graph_role)` for one record from its stored instance.
`tables.lmdb_fingerprint`	Fingerprint an LMDB's `data.mdb` (mtime+size) so a rebuild invalidates any cached signal.
`tables.read_frontmatter`	Return the leading dendron `---...---` block of a note (verbatim), or a minimal one if absent -- so a regenerated note keeps its dendron identity.