Frontends that bring real code into Sophia (B-16, Real Code Corpora). Metamath first. Source:
sophia_ingest.rs.Sources: Real Code Corpora; SPEC-006 Worlds, Artifacts and Evidence §1. code:
sophia_ingest.rs,metamath.rs
store_metamath stores a parsed database:
- statements: one record each, named
mm.<db>.<label>; - proofs: one record each, named
mmproof.<db>.<label>; - links: a
proofannotation on each statement and its proof; - the library: three nominal collections,
mmlib.<db>.proofs,.axiomsand.definitions, filled withStore::add_members.
Being an axiom, or the proof chosen for a statement, is a library’s choice, not a property of the statement. One statement can be an axiom in one library and proved in another, or even in another part of the same library: set.mm’s personal sections prove ax-1 from other axiom systems.
examples/mm_overlap.rs compares set.mm and iset.mm (statements and proofs shared).
store_code and store_module store code (Julia, Rust, Python) through store_items:
- records with Moonkale’s ids;
- generic functions keyed by (module path, name), never by name alone (
binding, the one grouping rule; Julia Name Resolution):- a Julia method belongs to
extendswhen it has one (Base.show(…) = …,A.f(…) = …, orshow(…) = …afterimport Base: show), else to its own module’s function; - the entity is
jl.<module path>.<name>, sojl.LinearAlgebra.dotandjl.LinearAlgebra.BLAS.dotare two functions, andjl.Base.showholds the package’s methods of Base’sshow; - a function named like a type in the same module is its constructor:
jl.<module>.<T>.new; - a qualified method that cannot be placed (
Base.$f(…)in an@evalloop) belongs to no generic function; - outside Julia, the module is the package name;
- a Julia method belongs to
- named definitions (types, constants, macros) as
jl.<module path>.<name>; - generic uses as
nominal_refs, keyed the same way (the crate resolves a use in its own module).
CodeReport::extended counts the generic functions of other top-level modules (Base’s, mostly) that the package adds methods to. Tested by tests/generics.rs on tests/julia-semantics/generics; checked against Julia itself by examples/julia_conformance.rs.
Sources and manifests (2026-10-20). store_module and store_metamath_source first store the source files and an ingest manifest (source), so every record can be rebuilt from the store alone (sophia reconstruct ingest.jl.<pkg>). store_metamath (parsed assertions only) remains for callers that have no text.
Source spans (2026-10-21). store_module also records a source-span fact per definition (source id and byte range, types), in one transaction (record_facts). A Julia record’s bytes are WP-66’s normal form, which does not print back; the span lets the interface show the text it was read from (sophia_interface). tests/reconstruct.rs checks that every stored Julia record has a span into a stored source whose text contains its name. store_code (single files, Rust and Python) does not record spans yet.
Naming is shared (2026-10-24). Binding, binding and entity_path are re-exports from moonkale-content-id (SPEC-009 §8), so Moonkale and Sophia name a definition by one rule; a change is a moonkale-wp66@N bump on both sides. The crate’s binding also refuses a key whose segment is not a valid store path. Checked on moving the pin: LinearAlgebra re-ingested gives the same 905 names and the same 145 definition ids as with dacef2e6, and reconstruct rebuilds all 3 152 records identically.