Skip to content

v1.0.0-rc1 — Architectural stabilization milestone

Pre-release
Pre-release

Choose a tag to compare

View all tags
@Vinix24 Vinix24 released this 12 May 08:40
· 1156 commits to main since this release
v1.0.0-rc1
86c8c7e

v1.0.0-rc1 — 2026-05-09 (Architectural stabilization milestone)

Chain summary: Wave 0 + Wave 0.5 of the strategic replan — architectural decisions LOCKED via 14 ADRs, central VNX state proven on real production data (855k snippets across 4 projects, 0 verifier discrepancies), CI gate enforcing OAuth-only Claude routing, smart context injection validated at +30 percentage-point dispatch quality lift on 658 outcome-tagged dispatches.

This is the architectural stabilization moment: dispatch envelope, receipt schema, NDJSON ledger format, and ADR-locked invariants are now backwards-compatibility-honoring. Future v1.x is feature additions on a stable foundation; future v2.0 reserved for breaking changes.

Added — ADR backfill (10 new ADRs, 003-014)

  • ADR-003 — OAuth-only Claude routing via claude -p subprocess (no SDK, no API key) (#439)
  • ADR-004 — VNX positioning: self-hosted alternative to Anthropic Managed Agents (#439)
  • ADR-005 — Append-only NDJSON audit ledger as primary orchestration substrate (#439)
  • ADR-006 — Mandatory staging→promote with human approval gate (#439)
  • ADR-007 — Multi-tenant project_id stamping with composite UNIQUE rebuilds (#439)
  • ADR-008 — Dual-LLM adversarial review (codex_gate + gemini_review) with contract_hash evidence binding (#439)
  • ADR-009 — Schema-first migrations via PRAGMA introspection (no hardcoded column projections) (#439)
  • ADR-010 — Subprocess adapter (claude -p) as canonical Claude routing (#439)
  • ADR-011 — Manager+worker hierarchy with explicit depth>1; subagents Conditional/Pilot for read-only parallel-fanout (#439, #447)
  • ADR-012 — Hybrid interactive+headless (no retire-interactive) (#439)
  • ADR-013 — Worker pool size as configuration, not constant (workers = N) (#445)
  • ADR-014 — Autonomous mode = pre-approved chain dispatch with SHA-256 chain-spec hash as consent token (#445)

Added — Structural enforcement

  • CI gate ADR-003: No Anthropic SDK Imports — blocks any import anthropic / from anthropic / import claude_agent_sdk in scripts/, dashboard/, tests/. Magic-comment opt-in for explicit exceptions only. (#439)
  • scripts/check_adr_003_no_sdk_imports.py + tests/test_adr_003_no_sdk_imports.py (23 unit tests).

Improved — Migration architecture (Phase 6 P4)

  • One-shot data import script scripts/migrate_to_central_vnx.py (#432): real-data success on 4 production source DBs (855k code_snippets, 505 dispatches, 0 verifier discrepancies)
  • Migration 0016 schema-first rewrite per ADR-009 (#446): replaces hardcoded 12-column projection with PRAGMA introspection; orphan project_id stamping uses migrating-project_id (not vnx-dev fallback); _import_table streams via cursor.fetchmany(500) instead of materializing full table — resolves OI-1375, OI-1376, OI-1377.
  • Composite UNIQUE rebuilds for 5 round-6 tables (session_analytics, vnx_code_quality, dispatch_quality_context, dispatch_metadata, dispatch_experiments)
  • Schema audit step _audit_unique_constraints() catches T3 single-column UNIQUE patterns at startup
  • Structural test parsing 0015_complete_project_id.sql enforces ALTER↔IMPORT_TABLES symmetry (would have caught all round-7/8/9 gaps)
  • Dispatch-register identity fallback per-field (was all-or-nothing); dual-writer locking on data file (was sentinel only); ghost-gate receipt rerouting fixed (#446 sibling)

Added — Repo hygiene (OI-1373 cleanup)

  • Tier 1: 8 business agent drafts moved from public roadmap/features/phase-16-business-domain-bootstrap/agent_drafts/ to private claudedocs/business-domain/ (#440)
  • Tier 2: 7 strategy files (KICKOFF.md, ROADMAP.md, backlog.yaml, roadmap.yaml, 3 dispatch_plans) from .vnx-data/strategy/ to claudedocs/strategy/; !.vnx-data/strategy/ carve-out removed from .gitignore (#441)
  • Tier 3: .vnx-data/state/PROJECT_STATE_DESIGN.md to private; 2 .vnx-data/state/ carve-outs removed (in #443)
  • Tier 4: 18 phase FEATURE_PLAN.md files from public roadmap/features/<phase>/ to claudedocs/roadmap/features/ (#442)
  • Tier 5: 9 stale docs from docs/internal/plans/ and docs/research/ archived to claudedocs/archive/{headless-t0-research,research-2026-spring}/ (#443)
  • Pattern: filesystem mv + git add -u (NOT git mv) — preserves files locally on disk while removing from git tracking. History scrub intentionally NOT performed.

Added — Strategic documentation

  • claudedocs/2026-05-09-vnx-strategic-replan-proposal.md v1.1 — 6-wave plan (Wave 0-6) replacing the overtaken-by-reality PRD-UH-001 v1.3 framing
  • claudedocs/PRD-VNX-UH-002-v1.0-DRAFT.md — "VNX 2.0: self-hosted alternative to Anthropic Managed Agents" with 8 structural moats, ICP refinement, and §11.b open horizons
  • claudedocs/2026-05-09-vnx-replan-inventory.md — full repo inventory: shipped/in-flight/planned/dead with provenance
  • claudedocs/2026-05-09-vnx-industry-research.md — 15-topic competitive landscape (May 2026): no competitor stacks 4+ of VNX's 5 axes
  • claudedocs/2026-05-09-vnx-platform-features-verification.md — Perplexity-verified 8 platform claims; 2 outdated (subagent session-resume + Anthropic Code Review GA), 1 partial (Agent SDK orchestration), 5 confirmed
  • claudedocs/2026-05-09-vnx-market-validation.md — verdict CONDITIONAL GO; OpenClaw ban (April 2026) retroactively validated M6/ADR-003 OAuth-only as structural ToS-policy moat
  • claudedocs/2026-05-09-vnx-smart-context-design.md — three-state intelligence pipeline design + +30pp lift PoC
  • claudedocs/2026-05-09-vnx-subagent-tactical-research.md — verdict Conditional/Pilot; T1-only VNX_T1_TACTICAL_SUBAGENTS=1, read-only parallel-fanout, 4-week pilot
  • claudedocs/2026-05-09-vnx-f46-f58-reconciliation.md — F46-F58 phantom-features claim refuted: 6 real-clean-PR + 6 cluster-titled + 1 verify + 0 phantom
  • claudedocs/2026-05-09-p4-migration-architecture-lessons.md — bug taxonomy + 14 action items from the 9-round PR #432 chain
  • claudedocs/blog-drafts/2026-05-09-intelligence-30pp-lift-DRAFT.md — 2300-word public-facing deep-dive draft on the +30pp result
  • PRD-UH-001 v1.3 archived to claudedocs/archive/

Reports

  • claudedocs/2026-05-09-vnx-replan-inventory.md
  • claudedocs/2026-05-09-vnx-industry-research.md
  • claudedocs/2026-05-09-vnx-platform-features-verification.md
  • claudedocs/2026-05-09-vnx-market-validation.md
  • claudedocs/2026-05-09-vnx-smart-context-design.md
  • claudedocs/2026-05-09-vnx-subagent-tactical-research.md
  • claudedocs/2026-05-09-vnx-f46-f58-reconciliation.md
  • claudedocs/2026-05-09-p4-migration-architecture-lessons.md

Validated — Smart context injection (gevalideerd op real production data)

  • IntelligenceSelector.select() pipeline measured: 88.9% dispatch success WITH intelligence vs 58.3% WITHOUT across 658 outcome-tagged dispatches (+30.6 pp lift)
  • 922 rows in intelligence_injections ledger growing hourly
  • Pipeline plumbing 9/10; content quality 2/10 — "the lift we got despite injecting governance noise; imagine signal" (Wave 5 design target)

Open items resolved

  • OI-1373 (49+ strategic/business docs in public repo) — fully closed via Tiers 1-5
  • OI-1375 (migration 0016 hardcoded 12-column projection) — closed by #446
  • OI-1376 (migration 0016 orphan COALESCE project_id fallback) — closed by #446
  • OI-1377 (_import_table materializes full source table) — closed by #446

Known open items deferred to v1.0.0 final

  • OI-1374 (dashboard /agent-stream SSE lifecycle regression — separate-PR cleanup)
Assets 2
Loading