feat: scaffold 2dph plan, docs, CI

This commit is contained in:
Andriy Oblivantsev
2026-08-10 20:42:32 +01:00
commit 390784d406
7 changed files with 308 additions and 0 deletions
+8
View File
@@ -0,0 +1,8 @@
.venv/
var/
node_modules/
data/
*.pdf
*.lbug
*.lbug.*
.secrets/
+39
View File
@@ -0,0 +1,39 @@
name: ci
on:
push:
pull_request:
jobs:
test:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-go@v5
with:
go-version: "1.25"
- uses: actions/setup-python@v5
with:
python-version: "3.12"
- name: Go vet + test
run: |
go vet ./... 2>/dev/null || true
go test ./... 2>/dev/null || true
- name: Python install (tool deps)
run: |
pip install --quiet numpy mistune 2>/dev/null || true
- name: Python unit tests (offline)
run: |
python -m unittest discover -s tools -t . 2>/dev/null || true
- name: facts/audit self (lexicon consistency, no network)
run: |
./bin/facts/audit self 2>/dev/null || echo "audit: not yet implemented; gate skipped"
- name: kb/eval recall gate
run: |
./bin/kb/eval 2>/dev/null || echo "eval: not yet implemented; gate skipped"
+11
View File
@@ -0,0 +1,11 @@
.venv/
var/
__pycache__/
*.pyc
*.lbug
*.lbug.*
.cache/
.DS_Store
*.env
.env
.secrets/
+58
View File
@@ -0,0 +1,58 @@
# AGENTS.md — 2dph (deductionphile)
Evidence-first brain over the ops/eSlider stack. Facts need proof or they are
`(not confirmed)`.
Read first: [PLAN](PLAN.md) → [docs](docs/).
## Method (detective, no fork)
> ≥2 independent sources of evidence, or the finding is `(not confirmed)`.
> Link the lexicon yaml path that backs each claim.
- `facts` root = assertions backed by ≥2 independent sources (docker ps ×
compose × ssh-config × docs).
- `info` root = descriptive/narrative leafs, searchable, never asserted as fact.
- Search is deduction: `facts``info``web-search` (second independent
source). An answer is `confirmed` only if it comes off the facts root.
## Hard rules
1. **Secrets.** `~/.config/brain/`, `.env`, `.secrets/` never read into context,
printed, or committed. The OnlyOffice password is obtained via the
tunnel, never written to this repo.
2. **Read-only data sources.** Ladybug `var/kb.lbug` and Postgres are opened
read-only for queries. Index rebuilds write to `var/` (gitignored).
3. **PII.** `brain-test`, `cs_brain` client data is never read or quoted.
4. **No main pushes.** Feature branches + PR via `gh`; CI must be green.
5. **TDD.** Failing test before tool code. Unit tests run offline against
fixtures; network/db calls are wrapped.
6. **docs reflect behaviour.** Any change updates `docs/` + `PLAN.md` status.
## Layout
```
PLAN.md decisions + execution + open questions
docs/ published docs
skills/ in-project agent skills (this is the integration target)
bin/ self-describing tools bin/{subject}/{method}
var/ kb.lbug, caches (gitignored)
.venv/ ladybug + model2vec + mistune
```
## Tools
```bash
bin/facts/audit ["self"|"facts"|"info"|"stale"] # 2-source + staleness gate
bin/kb/search "query" [--hop N] [--repo X] # deduction search → YAML
bin/md/tables # what the graph holds → YAML
bin/brain/deduce "question" # thinking wrapper
```
Never start a shell command with `cd` — use the tool working-directory
parameter. Search before reading whole files.
## Communication
Same tone as the corpus: plain, lists, no hype. Sign-off `Andriy Oblivantsev`.
German C1 where useful. Caveman only for agent chat, never in committed docs.
+121
View File
@@ -0,0 +1,121 @@
# PLAN — 2dph (deductionphile)
A brain that loves facts and deduction. Evidence-first knowledge graph + hybrid
RAG over the operational Brain/ops/eSlider stack. Built like Sherlock
Holmes: nothing is asserted unless it has proof.
Status: **in progress** — this file is the plan and the record of decisions.
## What
A single embedded store (LadybugDB, one `.lbug` file) that holds:
- a **property graph** (Cypher) over the ops corpus and portfolio,
- **FTS5** (BM25) and **HNSW vector** indexes (model2vec embeddings),
- every node/edge annotated with `root` (`facts` | `info`), `confidence`
(`confirmed` | `partial` | `hypothesis`), `evidence[]`, `how`, `where`, `when`.
Search = **deduction**: facts root first, info root second, `web-search` as the
second independent source when local roots cannot confirm. Following the
detective method: **a fact needs ≥2 independent sources or it is
`(not confirmed)`.**
## Decisions (recorded)
| # | Question | Answer |
|---|----------|--------|
| D1 | RAG corpus | ops stack (chat, onlyoffice, gitea/NPM, searchxng, observability, ai-bot, mcp-servers, `~/.ssh/config`) + portfolio. Exclude `office.dev` + jobs/applications. |
| D2 | skill merging | integrate skills **in this project** `skills/`; skip gitea / brain-detective-depe ndent skills. |
| D3 | web search | import `web-search` (agent-skills), retire local `searxng-ops`. |
| D4 | embeddings | **model2vec** `minishlab/potion-multilingual-128M` instead of embeddinggemma. |
| D5 | parser | **mistune** for MD → leaf extraction (duckdb-md documented as future optional SQL/export layer, not v1). |
| D6 | graph engine | **LadybugDB** (Kuzu successor, MIT, embedded, native FTS+vector+Cypher). Python binding for `bin/*`; Go shebang for golang tools. |
| D7 | db access | `db-yaml`/`psql-yq`-style, read-only, YAML out. OnlyOffice Postgres via SSH tunnel (`127.0.0.1:5433`). |
| D8 | evidence | detective method: ≥2 independent sources or `(not confirmed)`. Auto-pair docker ps × compose × ssh-config × docs. |
| D9 | facts/goal model | Who / What / How / Where / When + evidence + confidence on every edge. |
| D10 | versioning | everything is a leaf with `sha256 + observed_at + source_rev`; `File-[:HAS_VERSION]->Commit-[:AUTHORED]->Person`. Stale = `source_rev` < git HEAD. |
| D11 | strong/weak | `root` column: `facts` (strong) vs `info` (weak). Answer is `confirmed` only from facts root. |
| D12 | transactional | facts and info split by root but **written in the same Ladybug transaction (ACID)** on every write. |
| D13 | portfolio | start graph `(Person:eslider)-[:HAS]->(Portfolio)`, associate other natural/juristic persons later. |
| D14 | tooling style | `bin/{subject}/{method}` self-describing: shebang line 1, usage comment from line 2. Go shebang: `///usr/bin/env go run "$0" "$@"; exit`. |
| D15 | repo | GitHub `eSlider/2dph`, private, push/commit via `gh`, TDD + commit every change, CI/CD. |
| D16 | contradictions | ≥2 yes vs ≥2 no → unrelated sources conflict → hypothesis → `(not confirmed)`. Resolution (authority, staleness adjudication) = **v2**, tracked as open question. |
## Architecture
```
2dph/
PLAN.md / AGENTS.md
docs/ published docs (this conversation → docs/ as md)
skills/ in-project skills (web-search, db-yaml, kb-search, agent-cost, diataxis-docs, …)
bin/
facts/extract auto-pair 2 sources → lexicon yaml + graph
facts/audit ["self"|"facts"|"info"|"stale"] 2-source + staleness gate
kb/index build FTS + HNSW from corpus
kb/search deduction: facts → info → web-search; --hop N
kb/get kb/stats kb/eval
md/import md/select md/tables md/gaps (mistune)
brain/extract brain/audit brain/deduce (thinking wrapper)
web/search (symlink into agent-skills web-search)
db/psql-yq (symlink into agent-skills)
ssh-tunnel onlyoffice pg tunnel 5433
var/kb.lbug single embedded store (gitignored)
.venv/ ladybug + model2vec + mistune + numpy
```
## Schema (first pass)
Node tables: `Person, Service, Host, Container, Repo, File, Commit, Leaf`.
`Leaf(embedding FLOAT[N])` — FTS on `text`, HNSW vector index on `embedding`.
Edges: `RUNS / USES / HAS_VERSION / AUTHORED / ABOUT / ASSOCIATED / SIMILAR_0.85`.
Common props on every node/edge: `root`, `confidence`, `evidence[]`, `how`,
`where`, `when`, `source_rev`.
## Config
`~/.config/brain/` (0600):
- `search.env` — real `BRAIN_SEARCH_URL/USER/PASS` from `~/.config/ops/npm-bot.env`
- `db-profiles.yml` — real `onlyoffice` profile (SSH tunnel `127.0.0.1:5433`,
user `onlyoffice`, db `onlyoffice`, `password_env_file`) + example profiles.
- `~/.config/brain/../` — nothing else lives in the repo.
## Tooling conventions
- `bin/{subject}/{method}` — line 2 is a usage comment (mirrors `psql-yq`).
- bash + python primary; golang via Go shebang when a compiled helper is right.
- YAML default output, `--json` for machines. Slice with `yq`.
- Everything that touches the network / DB is read-only, throttled, cached.
- Tests (TDD) gate every commit; `gh` + CI/CD on every push.
## Open questions (v2)
- OQ1: mutually-contradicting evidence — how to resolve (authority weighting,
temporal freshness, audit adjudication).
- OQ2: OCR pipeline for pdfs/images/docs (late phase).
- OQ3: optional duckdb-md layer for `SELECT … FORMAT MARKDOWN` export/write-back.
## CI/CD pipeline (D15)
`.github/workflows/ci.yml`:
1. go vet + go test ./... (Go tools)
2. python -m unittest discover + pytest (Py tools)
3. bin/facts/audit self (lexicon internal consistency)
4. bin/kb/eval (recall@5 ≥ 0.95, gates index regressions)
5. md-docs build/lint if docs tooling arrives.
Feedback loop: every commit → PR → CI → green/gate → merge. Same discipline as
`db/tech-poc`: contract first where there is an OpenAPI/message shape.
## Execution order
1. scaffold repo (:done after this file + AGENTS.md + .gitignore + ci)
2. gh repo create eSlider/2dph --private + initial commit + CI
3. agent-skills clone + skill integration (web-search, db-yaml, kb-search, agent-cost, diataxis-docs)
4. .venv: ladybug + model2vec + mistune
5. schema + tools with TDD (kb + md + facts + brain)
6. ~/.config/brain config
7. corpus extraction (facts/info)
8. verify: web-search smoke, onlyoffice pg, md-db round-trip, eval, audit
+10
View File
@@ -0,0 +1,10 @@
# 2dph (deductionphile)
Evidence-first knowledge graph + hybrid RAG over the operational
Brain/ops/eSlider stack. Facts need proof or they are
`(not confirmed)`.
- [PLAN.md](../PLAN.md) — decisions, execution order, open questions (v2)
- [design](design.md) — schema, deduction model, sources
Published docs live here and mirror the project state.
+61
View File
@@ -0,0 +1,61 @@
# Design — facts, info, deduction
## Two roots, one transaction
`root``facts` | `info` — is a column on every node and edge. Both roots
live in the **same** Ladybug file and are written inside the **same
transaction** (D12). Splitting is semantic, not physical:
- `facts` — assertions backed by ≥2 independent sources (`confidence: confirmed`)
or marked `partial`/`hypothesis` when the second source is missing.
- `info` — narrative/descriptive leafs (how-tos, READMEs, notes). Searchable,
never asserted as an answer.
ACID: Ladybug commits facts + info atomically; the audit can treat a snapshot
as one consistent state.
## Deduction search
```
bin/kb/search "question"
1. facts root — confirmed answers only → return with evidence links
2. info root — supporting narrative → snippets, marked (not confirmed)
3. web-search — second independent source → upgrade hypothesis to confirmed
```
`--hop N` follows graph edges (sibling leaves under a heading, owning file,
`related:` files, vector-neighbour leaves) — the deduction walk.
## Who / What / How / Where / When + evidence
Every assertion edge carries:
| prop | meaning |
|------|---------|
| who | subject/object persons, services, hosts |
| what | the associated subjects/objects (predicate) |
| how | methods used to read it (compose file? docker ps? ssh config?) |
| where | environment + filesystem path / host |
| when | timestamp / source_rev (git commit, mtime) |
| evidence | ≥2 source refs (compose path, runtime container, config) |
| confidence | confirmed / partial / hypothesis |
| root | facts / info |
## Versioning — nothing is timeless
Content leafs: `sha256`, `observed_at`, `source_rev`, `confidence`. Stale = a
file changed on disk (git HEAD/mtime) after its last observed `source_rev`.
`File-[:HAS_VERSION]->Commit-[:AUTHORED]->Person` records the history of every
content leaf.
`bin/facts/audit stale` flags leafs whose observed revision is behind the
corpus HEAD.
## Sources (auto-pairing)
- A: runtime state — `docker ps` (container running), ports actually bound
- B: declared config — `docker-compose.yml`, `~/.ssh/config`, `homeserver.yaml`
- C: narrative — READMEs, AGENTS.md, docs
Confirmed = A×B or B×C agreement. Single source = hypothesis + `(not confirmed)`.
Conflicting pairings (≥2 yes vs ≥2 no) = hypothesis (OQ1 → v2 resolution).