v29 (Hardening & Validation) completed with all 6 kanban tasks done:
Key metrics at v29 completion:
| Metric | Value | Notes |
|---|---|---|
| Entities in main KG | 9,665 | In gourmet_kg.db (kanban board DB) |
| KG edges | 14,504 | hierarchical, observed_in, causal, correlational |
| Domain observations | 137 | In gourmet.db |
| Causal signals | 68 | In gourmet.db |
| GNN AUC | 0.7716 | Target 0.95 — needs real temporal labels |
| Oracle V4 | 0 crashes | Hardened |
| Brier Score | 0.0413 | Maintained from v27 |
| Pipeline | 6 sources | Resilient fetcher live |
Three research streams, ordered by dependency:
Problem: The Living Oracle's compute_external_score() uses a deterministic sin() proxy — not real data. v29.1 built the resilient fetcher but never wired it in.
Tasks:
sin() proxy in living_oracle_v24.py with real VIX/news-sentiment/macro from resilient_fetcher.pycompute_external_score() to log provenance (which source was used each fetch)Success criteria: Grounded oracle produces different (and better-calibrated) predictions than the sin proxy on at least 60% of test days.
Problem: The gematria_kg.db has 1 entity and 0 embeddings. The main KG has 9,665 entities but no embedding vectors. Without embeddings, cross-domain convergence analysis is impossible.
Tasks:
gourmet_kg.dbgematria_kg.db with real entity data and embeddings (migrate from kanban board KG)Success criteria: gematria_kg.db has 9,665+ entities with embedding vectors. First convergence analysis report on disk.
Problem: 4+ months of silence on Substack. The calibration story (Brier 0.270→0.041) is the strongest narrative.
Tasks:
Success criteria: Post published. One concrete public output after 4+ months of silence.
t_v30_0 v30.0 Orchestrator: Grounding, Embeddings & Forward Motion (ready)
├── t_v30_1 v30.1: Living Oracle Grounding — Wire Resilient Fetcher (todo)
├── t_v30_2 v30.2: KG Embedding Generation & Convergence (todo)
└── t_v30_3 v30.3: Substack — Calibration Story (todo)
t_v30_4 v30.0 Cycle Synthesis (todo, depends on v30.1 + v30.2 + v30.3)
| Priority | Stream | Why First |
|---|---|---|
| 1 | Stream 1 (Grounding) | Unblocks everything downstream. The oracle is the signal source. |
| 2 | Stream 2 (KG Embeddings) | Can run parallel with Stream 1. Uses existing data. |
| 3 | Stream 3 (Substack) | Uses existing data. No blockers. Just needs writing. |
firecrawl)firecrawl-redis (port 6379) ✅firecrawl-rabbitmq (port 5672/15672) ✅firecrawl-nuq-postgres (port 5432) ✅gematria_kg.db: 1 entity, 0 embeddings ❌ (Stream 2 fixes this)Plan by OWL, 2026-06-09. Launching v30 kanban now.