Source Registry
Lexery - Source Registry
Machine- and human-readable inventory of everything that has fed this wiki. If a claim cannot point here (or to a child note), treat it as opinion until ingested.
This page is the authoritative list of upstreams: Git remotes, local clones, documentation trees, automation outputs under raw/, and MCP-attached systems (Supabase SQL, R2 object stores, CI workflows). When you add a new ingest path or connector, update the registry table first, then mirror artifacts into LLM Wiki/raw/ or cite the MCP tool name so future sessions can reproduce the query.
Sibling: Lexery - Source Map (hierarchy + trust). Chronology: Lexery - Log. Topology: Lexery - Repo Constellation.
Registry table
| Source | Type | Path / URL | Last Ingested | Status |
|---|---|---|---|---|
| Lexery monorepo | Git repo | your local Lexery monorepo root |
2026-04-09 | Active |
| Public beta repo | Git repo | your local Ukrainan-Lawyer-LLM-BETA clone (historical) |
2026-04-09 | Historical |
| Bridge repo | Git repo | your local bridge clone (historical) |
2026-04-09 | Historical |
GitHub lexeryAI/Lexery |
Remote | https://github.com/lexeryAI/Lexery |
2026-04-09 | Active |
GitHub seredyuk2077/Ukrainan-Lawyer-LLM-BETA |
Remote | https://github.com/seredyuk2077/Ukrainan-Lawyer-LLM-BETA |
2026-04-09 | Historical |
GitHub seredyuk2077/Ukrainan-Lawyer-LLM |
Remote | https://github.com/seredyuk2077/Ukrainan-Lawyer-LLM |
2026-04-09 | Dormant |
| Linear workspace | API | Linear MCP | 2026-04-09 | Active |
| Codex session checkpoint | Local file | codex/SESSION_*_CHECKPOINT.md (monorepo) |
2026-04-09 | Session-scoped |
| Brain docs (~257 files) | Repo docs | apps/brain/docs/** |
2026-04-09 | Active |
| Root docs | Repo docs | docs/** |
2026-04-09 | Active |
| Supabase Legal Agent DB | MCP | Cursor MCP user-supabase-lexery-legal-agent-db; execute_sql queries |
2026-04-09 | Active |
| Supabase Legislation RAG DB | MCP | Cursor MCP user-supabase-legislation RAG; execute_sql queries |
2026-04-09 | Active |
| Cloudflare R2 legislation bucket | MCP | Cursor MCP user-cloudflare-r2-legislation; 374 acts as canonical JSON objects |
2026-04-09 | Active |
| Cloudflare R2 lexery-legal-agent bucket | MCP | Cursor MCP user-cloudflare-r2; retrieval traces, MM docs |
2026-04-09 | Active |
| GitHub Actions CI/CD | Workflow files | .github/workflows/; e.g. lldbi-brain-admin.yml (brain-admin automation) |
2026-04-09 | Active |
MCP usage note: Before calling any MCP tool, read the tool descriptor JSON under the Cursor mcps/ folder for that server. SQL and object-store tools are live; treat query results as point-in-time unless also captured under raw/codebase-snapshots/ or raw/architecture-docs/.
Raw Sources Inventory
Frozen or semi-frozen artifacts live under LLM Wiki/raw/. They are not a substitute for Git main, but they make diffs, PR context, and dashboard stats grepable inside the vault.
raw/github-prs/
- 10 PRs exported as paired JSON + Markdown per PR (
pr-1…pr-10), plus aggregateall-prs.json. - Use for: review narrative, titles, bodies, and metadata without opening GitHub in-browser.
raw/github-commits/
- 67 commits (30d window) captured in recent-commit listings; companion files include branch list and uncommitted diff snapshot for the Lexery workspace.
- Use for: velocity, branch topology, and “what changed locally before commit” forensics.
raw/architecture-docs/
- 4 primary docs:
LEXERY_LEGAL_AI_AGENT_ARCHITECTURE.md,MEGA_DIAGRAM_FULL.md,README.md,CURRENT_PIPELINE_STATE.md. - Use for: long-form architecture and pipeline state when the wiki page should cite a verbatim upstream export.
raw/codebase-snapshots/
brain-config.ts— 634 lines (vault snapshot); brain configuration surface for agents and flags.test-inventory.txt— 93 test paths (one path per line) enumerated for brain/tools coverage mapping.- Supabase live stats — dated markdown snapshot (e.g.
supabase-live-stats-2026-04-09.md) plus optional schema dumps when generated. - Monorepo packages — package/workspace listing export (e.g.
monorepo-packages-2026-04-09.md,root-package.json). - Workflow files — copy of selected GitHub Actions YAML (e.g.
workflow-lldbi-brain-admin.yml) for offline diff against.github/workflows/.
Regenerate these paths via #Automation Scripts after meaningful repo or infra changes; log the run in Lexery - Log.
Automation Scripts
All runnable Node scripts live in LLM Wiki/_system/scripts/. Shell helpers install-schedule.sh and uninstall-schedule.sh exist for scheduling but are not counted in the twelve core .mjs tools below.
| Script | Role (short) |
|---|---|
scan-codebase.mjs |
Walk the monorepo (respecting conventions) and emit snapshot-friendly extracts for the wiki. |
scan-supabase.mjs |
Query or document Supabase-facing stats/schema for raw/codebase-snapshots/. |
auto-fill.mjs |
Template and metadata passes to reduce manual wiki boilerplate. |
sync-git.mjs |
Local Git state sync (branches, diffs, commit lists) into raw/github-commits/. |
sync-github.mjs |
GitHub API or gh-backed export of PRs/issues into raw/github-prs/. |
sync-linear.mjs |
Linear issue/project sync hooks for roadmap notes (pairs with Linear MCP). |
generate-delta.mjs |
Compute deltas between wiki pages and fresh raw exports for review. |
update-log.mjs |
Append structured entries to Lexery - Log or log-shaped artifacts. |
ingest.mjs |
Orchestrate ingest steps from raw → vault pages (batch or scoped). |
suggest-links.mjs |
Propose wikilinks and backlink fixes from scan output. |
lint.mjs |
Validate frontmatter, broken links, or script invariants on wiki files. |
run-maintenance.mjs |
Single entrypoint to run the common maintenance chain in documented order. |
Typical order (adjust per task): sync-git.mjs / sync-github.mjs → scan-codebase.mjs / scan-supabase.mjs → ingest.mjs or auto-fill.mjs → lint.mjs → update-log.mjs. Use run-maintenance.mjs when the runbook matches your environment.
Not yet ingested
These are explicit gaps — absence from the table is not evidence of non-existence.
| Kind | Examples | Why ingest |
|---|---|---|
| Design | Figma, brand decks | UI truth vs shipped Portal |
| Comms | Telegram threads, email decisions | Informal ADRs |
| Qualitative | Customer interviews, support exports | Product–market fit signals |
| Quantitative | Analytics (Plausible, PostHog, etc.) | Traction and funnels |
| Ops | Deploy logs, Azure portal screenshots, Supabase dashboard | Lexery - Unknowns Queue items 1, 4 |
When any of the above lands in vault or _system/, add a row to the main table and note the ingest in Lexery - Log.
Trust & freshness rules
- Git
main/ active dev branch > stale local-only branches for "what ships." apps/brain/docs> generic memory for pipeline semantics.- Linear > wiki for issue state — but code wins on behavior (see Lexery - Drift Radar).
- Session checkpoints are handoff artifacts; re-verify before architectural claims.
- MCP query results are as fresh as the last invocation; prefer re-running
execute_sqlor R2 reads for production decisions, and snapshot outcomes intoraw/when the wiki needs a durable citation. raw/exports may lagmainby one maintenance run — check file dates and Lexery - Log before treating a snapshot as current.
Cross-links for common questions
- Where did Brain architecture come from? → Bridge repo +
apps/brain/docs/architecture(see Lexery - Legacy Architecture Bridge, Lexery - Brain Architecture). - What is public vs private? → Lexery - Repo Constellation, Lexery - GitHub History.
- What is planned vs shipped? → Lexery - Linear Roadmap + code grep, not wiki alone.
- Where are PR and commit receipts? → #Raw Sources Inventory under
raw/github-prs/andraw/github-commits/.
Maintenance hooks
- Daily: note new commits per repo in Lexery - Log (see Lexery - Maintenance Runbook).
- Monthly: drop stale "Last Ingested" rows into a refresh queue or bump the date after deliberate re-read.
- After MCP-heavy work: if you relied on Supabase or R2 for a factual claim, consider a one-off
scan-supabase.mjsor manual copy intoraw/codebase-snapshots/so the vault stays self-contained.
Ingest pipeline (conceptual)
Primary repos (git) ──► diff / path filter ──► human or model summary ──► wiki page patch
│ │
└──────────────────────────────────────────────┴──► Lexery - Log (timestamped)
MCP (Supabase / R2) ──► execute_sql / object reads ──► optional raw snapshot ──► registry row + wiki cite
- Select scope: never ingest entire
node_modulesor build artifacts — mirror.gitignoreand repo conventions. - Prefer paths:
apps/brain/docs/architecture/app/,codex/*.md,docs/,.github/workflows/for operational truth. - De-dupe: if the same fact appears in Linear and an ADR, link both but state one canonical sentence on the architecture page.
- Personal vs org: local paths in this registry are Andriy's machine; teammates should clone and update paths or use remote URLs only in shared copies.
- Raw folder hygiene: name snapshots with ISO dates where multiple revisions are expected (
*-2026-04-09.md).
Conflict resolution
| Situation | Rule |
|---|---|
Wiki vs main code |
Code wins; wiki follows in the same maintenance cycle |
| Wiki vs old Linear comment | Recent code + doc beats stale ticket text |
| Two wiki pages disagree | Merge into one canonical page; leave the other as a short pointer |
| Wiki vs MCP snapshot | Live MCP or Git beats an old raw/ file; refresh the export or bump the date |
Links
- Lexery - Source Map
- Lexery - Index
- Lexery - Log
- Lexery - Repo Constellation
- Lexery - Unknowns Queue — turns missing sources into answerable work.