Blueprint v2 Progress — Verified 2026-09-22
Blueprint v2 Progress — Verified 2026-09-22
This note records what is actually working on disk vs what task labels claim. Verified by Tam (research) and Min (fleet coordination) against live code, tests, and git history.
✅ Completed (verified on disk)
1. V2 Phase 1 Scaffold — ~/.solos/v2/ is real and working
| Deliverable | Status | Evidence |
|---|---|---|
settings.py | ✅ | Pydantic v2 typed settings, env overrides, singleton pattern, validated single source |
db.py | ✅ | Single DB layer, WAL mode, ordered forward-only migrations, _schema_migrations table |
migrations/ | ✅ | 001_seed_schema.sql + 002_cost_ledger.sql + runner.py |
tests/ | ✅ | 15 passed in 90.24s — test_migrations (4/4), test_settings (7/7), test_stability_gates (4/4 incl. health heartbeat) |
Run: cd ~/.solos && python3 -m pytest v2/tests/ -v
2. Durable memory deepen
- Commit
6d80e49— persona entries for ace/alpha/echo/nova - All 8 agents now have durable memory entries (preferences, rules, invariants)
3. ARCHITECTURE.md v2
- Location:
/home/ubuntu/.solos/docs/vault/solos/architecture/ARCHITECTURE.md(canonical) - Also at:
/home/ubuntu/.solos/docs/SOL-ARCHITECTURE-v2.md(working copy) - Author: Sol (architecture lead)
- Reviewer: Atlas — verified evidence-first
4. Hermes A2A study
- Mira + Echo parallel deep-dive completed
- Report filed: rep-79ee387d (Hermes A2A study PARALLEL)
5. Memory system fix (Min)
- Qdrant rebuilt from SQLite (299 facts indexed, correct payload schema)
retrieve_facts()→ now uses Qdrant semantic search (was broken LIKE)search_facts()→ Qdrant direct + SQL keyword fallback (was broken three-tier)store_fact()→ writes to both SQLite and Qdrant- All 8 agents verified: persona + memory plane (durable + semantic + episodic)
6. Eva port correction
- Registry updated:
:8443(server-side reverse tunnel from laptop:8441) - Heartbeat script, central_registry, all port maps updated
🔄 In Progress (likely stalled by credit outage)
| Task | Owner | Status | Notes |
|---|---|---|---|
| task-8900902e (master GO) | Sol | in_progress | "Proceed with the plan" |
| task-9a9daa82 (eng playbook) | Sol | in_progress | 8 videos deep-dived |
| task-d357f405 (Echo security) | Echo | in_progress | Rebuild consolidation |
| task-3cb318d8 (AI Core study) | Echo | in_progress | Ubuntu Core 26 lane |
| task-da96f322 (render queue) | Atlas | in_progress | Was blocked, now executing |
| task-9a34e738 (OpenCode study) | Sol | pending | Integration study |
⏳ Pending
| Task | Owner | Scope |
|---|---|---|
| task-5f877e2d | Ace | V2 Deliverable C: P2 Remote Integration & Eve Boundary (Days 8-12) |
| task-4c821004 | Mira | Rebuild consolidation (cost angles + per-phase budget) |
| task-2675acb4 | Nova | MIGRATION P3: Eva laptop → central registry |
🚫 Cancelled / Deprioritised
- Unification v2 Phase 3-4 (cancelled)
- GOV-2/3/4 (cancelled)
- ANDROID-10/11 (cancelled)
- PULSE-M4 (cancelled)
⚠️ Hard Truth
The code progress is solid — scaffold passes all 15 tests, architecture is canonical and evidence-grounded, memory system is rebuilt and working.
However, live execution is now gated on the P0 credit outage. With SolOS GO exhausted and the queue jammed 13-16h, the "in-progress/pending" items cannot actually execute LLM turns until credits are restored or provider is swapped.
What needs to unblock:
- Credit restoration (GO quota top-up or provider swap)
- Then: Ace's Eve boundary work, Mira's cost model finalization, persona verification pass
Verified by Tam + Min, 2026-09-22. Test run: 15/15 pass in 90.24s. Git: 6d80e49.
Blueprint v2 — Re-verification (Tam, research lane)
Re-dispatch of the final audit (Atlas → Tam). Re-probed live 09:00 AEST 2026-09-22. Read-only; nothing modified on server. Prior audit = 2026-09-22-blueprint-v2-audit (08:55 snapshot). This note confirms that verdict and records one fresh delta.
Hermes Agent Re-persona / Migration
Rewriting a Hermes agent's persona (identity, role, voice) touches FOUR files, not one. Miss any and the old identity wins.