Every note of the Sophia vault, in reading order. How the vault is organised: Start Here. The mathematics behind the design is in the CT-ML wiki, Track F; links to it below are ordinary web links.
Now: Sophia and sophialang (2026-10-20)
The current work. Sophia is the content-addressed store for code in every language; sophialang is its own language. The first priority is the interface to Moonkale. Start at Status.
- Requirements and coordination: Requirements (SR-nn, each with its origin) · Moonkale Interface Needs
- Specifications (Specifications): SPEC-001 Sophia Store v0 · SPEC-002 Sophialang Principles · SPEC-003 Sophialang Core v0 · SPEC-004 Moonkale Interface · SPEC-005 Rendering Profiles · SPEC-006 Worlds, Artifacts and Evidence · SPEC-007 Sophia Query Language · SPEC-008 Module-Level Julia Resolution · SPEC-009 Moonkale Folder Store · SPEC-010 Store-Truth Folders and Generated Files
- Decisions: Decisions (S-ADR-0001 to S-ADR-0012)
- Theory: Theory of the v0 Store · Type Taxonomy · Computational Trinitarianism
- Research: Review 2026-10-20 Real Code 2 (external) · Review 2026-10-20 Real Code (external) · Review 2026-10-20 (external) · Review 2026-10-20 Query Language (external) · Review 2026-10-20 M3a (external) · Query Languages for Sophia · Zero-Copy Storage of Canonical Graphs · Compression and Canonical Form · Block Size and Hash Overhead · Adopting from Unison · Minimal Binaries · LLMs and Mixed-Language Code · Operating Systems and Sophia · Hardware for Sophia · Database Comparison
- Storage and materialisation (what is kept, what is recomputed): Storage and Materialization · Object Categories · Current Inventory · Object Lifecycle · Materialization Policy · Garbage Collection · Compression · AD as a Materialization Case · Storage Measurements
- Languages (support levels L1–L6, one folder each): Languages · Julia: Julia Overview, Julia Parsing and Normalization, Julia Name Resolution, Julia Multiple Dispatch and World Age, Julia Macros and Generated Functions, Julia Types, Julia AD and Compiler Backends, Julia Conformance, Julia SophiaLang Interoperability · SophiaLang Overview · Sophia Query Overview · Metamath Overview · Rust Overview · Python Overview · Lisp Overview · Lean Overview
- Plans and measurements: Milestones · Semantic Boundary Experiments · Real Code Corpora · Benchmark Plan · Benchmark Results
- Problems: Problem Log (SP-001 to SP-015)
- Research questions (Research Questions): RQ-001 Recursive Groups up to Bisimilarity · RQ-002 When Incremental Collection Equals a Fresh Build · RQ-003 Core Terms or Surface Syntax for Language Models
- Reading of the Earlier Vault: where the notes below may have misread Daniel. Read the sections below as ideas, not requirements.
0. The idea
- The Original Idea — the project as it was first written down: one graph database for code in every language and all its IRs, equivalence proofs instead of an FFI, hashes instead of names
- Design Overview — what that becomes once made precise, and the three claims the project makes
- Glossary — the vocabulary, including where it departs from the original text
1. Identity: what a node is
- Core Calculus — the one language every frontend elaborates into; definitional versus propositional equality; the
Prf/Cmpsplit - Hashing and Identity — Merkle hashing of canonical forms; binders, cycles, three hashes per definition
- Multi-AST Layering — terms, types and proofs as one syntax; source, core, MLIR and LLVM as layers connected by functors
- Graph Schema — node and edge kinds, the
EQUIVedge, three storage backends - Naming and Change Propagation — names as mutable labels over immutable hashes
2. Equivalence: when two nodes are the same
- Equivalence and Witnesses — the ladder from
alphatoasserted; congruence; the cross-language relation; refinement - Cross-Language Semantic Hazards — the concrete ways “the same” turns out to be false
- Effects Memory and Resources — effect rows, regions and borrows; four memory models under one roof
- Trusted Computing Base — what must be correct for any of this to mean anything
- Tests and Documentation as Nodes — evidence that is not proof, and prose that is not code
3. Compilation: a compiler as queries
- Compilation as Query — passes as rules, saturate then extract, incrementality for free
- Query Cookbook — concrete queries in SQL, Cypher and Datalog
- Content-Addressed Precompilation — the original Julia motivation, made precise
4. Plans and risks
- Roadmap — milestones M0–M7, and why M3 is where the project justifies itself
- Open Problems and Risks — what may sink it
- Repository Layout — where code and notes live; Code Map — the per-file design notes
5. State of the art
State of the Art is the index. By topic: content-addressed code · e-graphs · incremental computation · graph databases for code · code indexing · IR formats · graph IRs · equivalence checking · verified compilation · superoptimisation · interoperability · semantics of real languages · semantics frameworks · dependent types · Julia compilation
6. Background cards
Background Concepts is the index: LLVM and MLIR, Unison, hashing, rewriting, operational semantics and effect systems, type theory, Julia internals, databases.