The first version of this vault (
vault/Design/,Background/,State of the Art/, the comment-only code) was written by an agent from Daniel’s first prompts. Daniel said on 2026-10-17 that it “may have misunderstood” him, and that the repository is brainstorming, not a plan. This note lists where the earlier reading and Daniel’s own statements differ, so each point can be confirmed or dropped. Until then, theDesign/notes are ideas, not requirements. The requirements are in Requirements.Sources:
meta/Prompts/Prompt.md,Prompt2.md,Prompt3.md; The Original Idea; MoonkaleNorth Star.md(Daniel’s words, 2026-10-14/15); Daniel’s message of 2026-10-17.
| # | The earlier vault says | Daniel said | To confirm |
|---|---|---|---|
| 1 | Rust and Julia split the work by concern: a Rust kernel and store, a Julia frontend and macros, joined by a C ABI (Repository Layout). | ”Julia for rapid development … run Rust in production.” Not the same thing implemented twice (Prompt3). | Settled 2026-10-17: Rust only. Julia maybe much later, for ML on the code graph (S-ADR-0004 Rust for the store, Julia for experiments). |
| 2 | ”Julia + Lean” means “a proof layer about Julia code, not dependent types inside Julia” (Roadmap M5). | ”A dependently typed Julia with Prolog-like elements … checked by several systems (dependent, cubical, algebraic)” (North Star, 2026-10-14). | Settled 2026-10-20: yes, sophialang, Julia-flavoured, with dependent types as an exchangeable checker that adds no run-time cost (SR-53, SPEC-002 Sophialang Principles). |
| 3 | The headline is cross-language interoperability through equivalence edges with evidence. Everything is ordered towards it (M6, M7). | That goal is in the Original Idea. But on 2026-10-17 the reason to continue is to specify Moonkale, and the North Star frames many languages as parsed into one graph, with proofs as annotations. | Settled 2026-10-20: the Moonkale interface first; then individual languages parsed into the store; reuse and interop are the long-term dream (SR-63, SR-73). |
| 4 | M0 is Metamath and M1 is a Lisp, chosen to test the plumbing. | Not mentioned by Daniel. | Settled 2026-10-20: sophialang first; Metamath or Lisp maybe later (S-ADR-0008 Sophialang before Metamath or Lisp). |
| 5 | The store is DuckDB first. | Lists six databases without a preference. | Settled 2026-10-20: try several and compare (Database Comparison); SQLite is the default meanwhile. |
| 6 | Precompilation of Julia is “where the project justifies itself” (M3). | ”I wanted to improve precompilation for Julia” is one motivation among several. | Settled 2026-10-20: a goal, but not the first priority. It needs analytics on top of the store (SR-66). |
| 7 | The code files contain only comments. | ”Please do not add any code yet” (Prompt2, at the time). | Settled 2026-10-20: yes, plan the milestones and implement (Milestones). |
What the earlier vault still gives
- Background cards and surveys are useful reference and stay as they are: Background Concepts, State of the Art.
- Design notes that agree with SPEC-001 and can be read alongside it: Hashing and Identity (Merkle hashing, binders, cycles); Naming and Change Propagation; Tests and Documentation as Nodes; Equivalence and Witnesses (its ladder of evidence strengths).
- Design notes to read as ideas only, until confirmed: Core Calculus (its
Prf/Cmpsplit is one option); Compilation as Query; Cross-Language Semantic Hazards; Content-Addressed Precompilation; Roadmap.