Hierarchy + KG + Capabilities v2 + 93.05% line coverage. Full-spectrum campaign launched 2026-04-27. Local pre-flight running; DigitalOcean infrastructure provisioning queued; 48 a2a-gate scenarios + 4 ship-gate phases scheduled. Live status below; this page auto-updates as gates complete.
Real-time progress as the orchestrator drives 4 ship-gate phases + 51 a2a scenarios across a 4-node DigitalOcean mesh.
1600 lib tests, 0 failures. Coverage 93.05% (gate floor 92%). All MCP tools + HTTP routes exercised. Fmt + clippy --pedantic clean.
Federation W=2 of N=3 quorum across 3 DigitalOcean droplets. Sustained burst from node-1; convergence verified.
v0.6.2 → v0.6.3 migration roundtrip. v15 schema applied (temporal-validity columns + entity_aliases). All rows queryable post-migrate.
50 cycles of kill_primary_mid_write. Zero data loss; pending actions converged; no daemon crashes. Soak windows (3-21d) deferred per directive.
v0.6.2 certified the a2a-gate's 35-scenario v0.6.0 surface across 6 cells (ironclaw + hermes + openclaw × off / tls / mtls). For v0.6.3, all 42 existing scenarios re-run unchanged (capabilities-v2 is additive — old clients reading v1 paths continue to work). Six new scenarios cover the v0.6.3-only surface.
Assert schema_version="2" on every cell. All 5 new top-level blocks (permissions, hooks, compaction, approval, transcripts) present with correct shape. v1 fields (tier, version, features, models) preserved.
memory_get_taxonomy with namespace_prefix + 8-level depth. Verify subtree_count aggregation honest under limit truncation. truncated flag set when limit drops rows.
memory_kg_query with as_of past timestamp returns only edges valid at that time. memory_kg_invalidate flips an edge to invisible from present-tense walks but visible in past-tense walks.
memory_kg_timeline returns the chronological sequence of edge_added + edge_invalidated events for a memory or entity. Filter by relation type and date window.
memory_entity_register idempotent on (canonical_name, namespace) — re-register reuses entity_id and merges new aliases. memory_entity_get_by_alias resolves any alias to canonical entity.
memory_check_duplicate near-match semantics. Configurable similarity threshold. Returns ranked candidates not deduplicates by itself.
v0.6.3's testing runs in parallel mode using ai-memory itself as the coordination bus. 17 agents at peak, distributed across DigitalOcean droplets + GitHub Actions runners. Full time math here → · Architecture here →
Version bump + CHANGELOG promotion + final gate run. Concurrent: terraform provisions + scenario authoring.
4 ship-gate phases on 9 droplets + 4 a2a batches on 4 droplets — all concurrent. Wall = max of any leg.
GitHub Actions matrix: 5 platform builds + 5 publish channels in parallel (already optimal).
brew · cargo · docker · apt · dnf — 5 concurrent agents validate schema_version=2 on each.
From the v0.6.3 grand-slam charter — six streams shipped: