Version history

Full release artifacts and per-version notes live on GitHub Releases. The milestones below each mark an acceptance-verified capability jump; evidence is archived under docs/audit/.

v0.8.x — Knowledge & closed loop (late Oct 2026)

• v0.8.4 (2026-10-26) — 3-minute English-UI demo video (README / VitePress / standalone player) + EDA pipeline title, online-viewer height and ontology baseline i18n fixes.
• v0.8.3 (2026-10-26) — Results-library re-acceptance closeout (L1/L2/L4) + clean-tree release prep; global zero-state test fixes.
• v0.8.2 (2026-10-25) — Results-library reviewer fixes: PATCH role gate, hot toggle, absolute URLs, evidence chain.
• v0.8.1 (2026-10-25) — diag-bridge results library: three tools (diag_results / diag_result_get / diag_result_annotate) + the /plugins results panel + five governance switches, off by default, visibility = diagnostic-token owner. Readable model metadata and one-click ontology-draft auto-fix; the ontology graph tab became an interactive node-edge view (1/2-hop neighbourhood drawer). Same round completed the full acceptance: 821 checks green.
• v0.8.0 (2026-10-24) — Ontology service + AI-first model loop: 10 entity kinds × 8 closed relation kinds × three evidence levels × five-state lifecycle; four compiled artifacts (process model / ProblemIR delta / scene patch / guard set) under a tighten-only invariant with per-rule traces; agents build the virtual model themselves and propose adoption via optimizer_profile_propose — an HITL approval publishes and rebinds, verified live across two rounds (v1→v2→v3) with zero device writes. 72/72 ontology e2e · 46/46 model-loop e2e. See Ontology & AI-first model loop.

v0.7.6x–7x — Governance & optimization (Oct 2026)

• v0.7.71 (2026-10-24) — Process optimizer & solver: process model → AML training (G1–G3 gates) → profile freeze → hash-pinned ProblemIR → OSQP/SLSQP solve → independent TS crosscheck → governed dispatch; declarative tool-face governance (all 104 host tools carry explicit references declarations).
• v0.7.70 (2026-10-21) — Data workbench + EDA studio (/data · /studio): compound queries, bulk export, golden trajectories, FDC/SPC (Nelson 1-8 / EWMA / CUSUM), OCAP, virtual metrology, derived datasets. See Data workbench & EDA studio.
• v0.7.66 (2026-10-17) — Unified tool-call preflight (tool-verify: argument shape + reference ownership + chained lineage backtracking).
• v0.7.63–65 (2026-10-15/16) — Data-analysis ledger finalized (analysis_reports + /analysis page + IDD six-intent catalog), dead tool surfaces removed. See Data-analysis ledger.
• v0.7.62 (2026-10-13) — Data-analysis system: analysis_save / analysis_query first-class records with line-scoped permissions.
• v0.7.58 (2026-10-11) — Production hardening: hold-write heartbeat lifecycle, silent mock fallback blocked, frame fingerprints + object GC, Timescale daily chunks + columnstore compression, heap-watermark auto-respawn.

v0.7.4x–5x — Platform foundation (Sep 2026)

• v0.7.54 (2026-09-30) — Full protocol-matrix regression (Modbus/OPC UA/MQTT/HTTP/serial) and MES integration closeout.
• v0.7.52 (2026-09-28) — Team knowledge-base toggles, lead→worker node delegation, generic closed-loop optimization templates.
• v0.7.50 (2026-09-28) — Dual-mode optimization loop (production shadow/adoption), any-scenario twin modeling, aw mcp gateway, line-operations page, timezone base.
• v0.7.45 (2026-09-21) — Scheduled tasks, DCW write-hold windows, real-scenario merged acceptance all green.
• v0.7.42 (2026-09-19) — Site-wide zh/EN i18n completion and end-to-end acceptance.

Upgrading

Releases follow semver. aw start verifies the config root on every launch and migrates the legacy data/ layout into it (newest file wins), so data survives upgrades; SQLite schema migrations run server-side at boot.