Skip to content

CLAUDE.md

Latest commit

 

History

History
63 lines (49 loc) · 3 KB

CLAUDE.md

File metadata and controls

63 lines (49 loc) · 3 KB
Raw

CLAUDE.md — nextgen-databases (COORDINATION REPO)

CRITICAL: This is a coordination repo. No database implementation lives here. (Read First)

nextgen-databases coordinates a portfolio of database projects. It does not hold their implementations. Each database and each database language has its own repo (see REGISTRY.adoc).

Before you create or edit a file here, STOP and ask: is this coordination, or is it about one specific database? If it is about one database, it belongs in that database's own repo.

NEVER (in this repo)

  1. NEVER add per-database source code, schemas, migrations, storage engines, or database-language implementations here.
  2. NEVER add per-database design docs, whitepapers, benchmarks, or datasets here.
  3. NEVER create a new top-level directory for a database or language — create or extend its own repo instead (REGISTRY.adoc).
  4. NEVER "just put it here for now." That is exactly the drift this repo is recovering from.

ALWAYS (what DOES belong here)

  1. Portfolio coordination: README.adoc, TOPOLOGY.adoc, ROADMAP.adoc, EXPLAINME.adoc.
  2. The registry of databases/languages → their repos: REGISTRY.adoc.
  3. Cross-database integration tests (tests/) and shared infrastructure (flake.nix, Justfile, stapeln.toml, opsm.toml).
  4. Governance & metadata: .github/, .machine_readable/, .well-known/, LICENSES/, CONTRIBUTING.adoc, SECURITY.adoc, 0-AI-MANIFEST.a2ml.

Where database content goes

See REGISTRY.adoc for the authoritative map. Examples: VeriSimDB → hyperpolymath/verisimdb; Lithoglyph → hyperpolymath/lithoglyph; Glyphbase → hyperpolymath/glyphbase; GNPL, the narration/projection language → hyperpolymath/gnpl; NQC → hyperpolymath/nqc.

Transitional note

lithoglyph/ and verisimdb/ are done: extractions completed 2026-07-27 and 2026-08-03, their 819 and 713 files removed, each leaving a pointer README. Everything they held is preserved at its original path in the split-history/{lithoglyph,verisimdb} tags on origin — never prune those tags or the _split_* branches.

verisimdb was a FORK, not a duplicate (108 files unique here, 331 substantive diffs), so seven files with no upstream counterpart were ported first — hyperpolymath/verisimdb#219. Do not assume the next extraction is a clean subset; check before deleting.

The directories quandledb/, nqc/, typeql-experimental/, verisim-core/, and verisim-modular-experiment/ are legacy content being extracted to their own repos — see docs/migration/RESITE-DATABASES-TO-OWN-REPOS.adoc. Do not grow them. A CI guard (.github/workflows/placement-guard.yml) and a local pre-write hook (.claude/hooks/block-db-writes.sh) enforce this: creating a new file inside those directories is blocked, because it belongs in that database's own repo.

Also read 0-AI-MANIFEST.a2ml (universal AI entry point) and AGENTS.adoc.