ac01fffd9a
Architectural snapshot of the lakehouse codebase at the point where the
full matrix-driven agent loop with Mem0 versioning + deletion was
validated end-to-end.
WHAT THIS REPO IS
A clean single-commit snapshot of the lakehouse code. Heavy test data
(.parquet datasets, vector indexes) excluded — see REPLICATION.md for
regen path. Full lakehouse history at git.agentview.dev/profit/lakehouse.
WHAT WAS PROVEN
- Vector retrieval across multi-corpora matrix (chicago_permits + entity
briefs + sec_tickers + distilled procedural + llm_team runs)
- Observer hand-review (cloud + heuristic fallback) gating each candidate
- Local-model agent loop (qwen3.5:latest) with tool use + scratchpad
- Playbook seal on success → next-iter retrieval surfaces it as preamble
- Mem0 versioning + deletion in pathway_memory:
* UPSERT: ADD on new workflow, UPDATE bumps replay_count on identical
* REVISE: chains versions, parent.superseded_at + superseded_by stamped
* RETIRE: marks specific trace retired with reason, excluded from retrieval
* HISTORY: walks chain root→tip, cycle-safe
KEY DIRECTORIES
- crates/vectord/src/pathway_memory.rs — Mem0 ops live here
- crates/vectord/src/playbook_memory.rs — original Mem0 reference
- tests/agent_test/ — local-model agent harness + PRD + session archives
- scripts/dump_raw_corpus.sh — MinIO bucket dump (raw test corpus)
- scripts/vectorize_raw_corpus.ts — corpus → vector indexes
- scripts/analyze_chicago_contracts.ts — real inference pipeline
- scripts/seal_agent_playbook.ts — Mem0 upsert from agent traces
Replication: see REPLICATION.md for Debian 13 clean install + cloud-only
adaptation (no local Ollama).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Truth rules — file-backed policy
Phase 42 PRD: "truth/ dir at repo root — rule files, versioned in git."
This directory is the canonical home for TruthStore rules loaded from
disk. Each *.toml file holds a set of TruthRule records for one
task class. The truth crate's load_from_dir(path) walks this
directory, parses every .toml file, and registers the rules it finds.
Structure
truth/
├── README.md ← this file
├── staffing.fill.toml ← rules for task_class="staffing.fill"
└── staffing.any.toml ← rules for task_class="staffing.any"
File naming is informational — load_from_dir respects whatever
task_class the rule declares internally, NOT the filename. Using
task-class-matching filenames is a convention for humans reading the
git tree.
Rule shape
[[rule]]
id = "worker-active"
task_class = "staffing.fill"
description = "Worker must be active"
condition = { type = "FieldEquals", field = "worker.status", value = "active" }
action = { type = "Pass" }
condition.type is one of:
Always— always trueFieldEquals { field, value }FieldMismatch { field, value }FieldEmpty { field }FieldGreater { field, threshold }FieldContainsAny { field, needles }
action.type is one of:
Pass— rule informational; no enforcementReject { message }— short-circuit with errorRedact { fields }— mutate the context, strip fieldsBlock { message }— hard stop, alert
Composition
The crate's default_truth_store() continues to register rules
in code for backward-compat. Operators can layer file-backed
rules ON TOP via load_from_dir:
let store = truth::default_truth_store();
let store = truth::load_from_dir(&store, "/home/profit/lakehouse/truth")?;
File-loaded rules are additive — they do NOT replace in-code rules.
This lets the staffing team tune rules at the file level (edit a
threshold, add a new FieldContainsAny blocklist) without waiting
for a code deploy.
Explicit non-goals
- No hot reload — per Phase 42 PRD ("Truth reload is explicit
in this phase"). Operators bounce the gateway or POST
/v1/contextrefresh endpoint (future) to pick up changes. - No inheritance — each file stands alone; rule IDs must be unique across all files. Duplicate-ID detection is a load-time error.