Radix 0.15.0

Install this version#

No prebuilt archives were published for this version. It is listed here because its release notes are part of the record.

Release notes#

Intensive single-day release that introduces the norma-faber annotation-driven standard library framework (11 core commits), replaces auto-generated grammar docs with a hand-written EBNF spec and prose-based grammatica tutorials, and integrates the faber-trials LLM learnability research harness into the main repo.

Scale#

Major tracks#

Norma-faber: stdlib annotation framework#

Norma-faber defines standard library methods in Faber source (.fab files) with codegen annotations that map to each target language. Four annotation types control the pipeline:

• @ innatum — maps a Faber collection type to its native target type (e.g., @ innatum ts "Array", py "list")
• @ subsidia — declares an external implementation dependency
• @ radix — declares morphological variants
• @ verte — maps a method name (simple rename) or supplies a template with § / §0/§1 indexed placeholders

Key milestones in this release:

• Design document defining the approach for collections (fake morphology) vs IO (real morphology) (adc0cfc4e)
• Faber parser parses all four annotations into the AST, with tests (225afbbd5)
• Parser rejects multiple `@ verte` targets on a single line — each must be its own annotation (cbf856706)
• Richer grammar documentation of annotations in EBNF.md and fons/grammatica/structurae.md, plus working exemplum (e04851ede)
• Rivus parser (Faber's bootstrap compiler) implements parallel annotation parsing with error codes P301–P308 (14134c1be)
• Vertical slice proves the architecture end-to-end: lista.adde and lista.addita codegen correctly for TypeScript and Python via the registry (b5f9ef75b)
• All 5 codegen targets (TS, Py, Rust, C++, Zig) wired to the norma registry; placeholder character fixed from 'S' to § (45d34369b)
• Build-time generated registry (scripta/build-norma.ts) reads .fab files and produces norma-registry.gen.ts and norma-registry.gen.fab; add bun run build:norma command (f297c6dd3)
• Prefer bun run over direct scripta/ invocation for proper environment setup (2c055cb7c)
• 47 of 53 lista methods migrated to fons/norma/lista.fab with indexed §0/§1 placeholders; § regex fixed from matching ASCII 'S' (b781954da)
• Progress, architecture diagrams, and full lista migration plan documented in consilia/futura/norma-faber.md (f1b40501d, 7f62b6848)
• Development workflow section (build/test cycle for adding methods) (a005aef55)

Documentation restructure#

• Hand-written `EBNF.md` (416-line spec) replaces auto-generated GRAMMAR.md (1525 lines), becoming the authoritative language specification for both faber and rivus compilers (5103db701)
• Grammar extraction script (scripta/extract-grammar.ts) removed — docs no longer derived from parser source (5103db701)
• Grammatica docs rewritten as prose tutorials with code examples and etymology-based explanations: fundamenta.md, typi.md, functiones.md, structurae.md, importa.md, operatores.md, regimen.md, errores.md — ~2686 lines added, ~948 removed (4ef97717a)
• Grammatica files moved from grammatica/ to fons/grammatica/ (5103db701)

LLM research harness#

• `probationes/` directory created by merging faber-trials/ into the main repo — a full LLM learnability research pipeline (5103db701)
• Harness includes: analyzer.ts, grader.ts, loader.ts, pipeline-runner.ts, runner.ts, prompt.ts, api.ts, types.ts (5103db701)
• Tasks: complete.yml, complex.yml, predict.yml, translate.yml (5103db701)
• Thesis document: probationes/thesis.md (526 lines) (5103db701)
• Relative paths in probationes harness fixed to use import.meta.dir for correct resolution from the repo root (455be4da7)
• Trials no longer run post-AI review — runner updated to skip the review step (9555f9fe6)
• README reframed as LLM-oriented IR ("Latin hypothesis") with Research & Evidence section citing Framework 1.1 trial results; systems languages (Zig, Rust, C++) as primary value prop (d2b8d710f)

Other changes#

• Test source field standardization: faber: and input: fields renamed to source: across all 71 YAML test files; harness consolidated to a single TestCase type; added faber: false skip flag (ac4d4057d)
• CHECKLIST website preparation: compiler descriptions and CLI usage sections added to both fons/faber/CHECKLIST.md and fons/rivus/CHECKLIST.md (e9110fc7b, 9f6d16bc4); bracket status markers ([x], [ ], [~]) replaced with Unicode circles (●, ○, ◐) for better HTML rendering (462b62d2d); table formatting cleaned up in both CHECKLIST files (20f3f919f)

---

All releases · Install the current release