feat: add sqlite soak verification gate

Author: Jacobinwwey

docs/diataxis/en/explanation/development-progress-dashboard.md

@@ -73,6 +73,26 @@ The next gap is narrower now:
 - keep the new render substrate honest under real browser/Tauri runtime verification,
 - preserve a clean downgrade/materialization boundary for later Godot reuse.
 
+## 2026-05-27 Phase-1 SQLite Soak Gate Baseline
+
+- The embedded `graphdb/sqlite` baseline no longer stops at restart continuity plus workload-envelope proof.
+- Current HEAD now also carries a dedicated host-level soak/performance verifier:
+  - `npm run verify:foundation:sqlite-runtime:soak`
+  - emits structured JSON reports under `output/verification/foundation-sqlite-runtime/`,
+  - keeps release-grade evidence separate from the lighter `smoke` / `medium` / `heavy` matrix path.
+
+What this newly proves:
+
+- repeated restart cycles on both `dist` runtime and packaged sidecar paths,
+- structured startup / ingest / readiness / diagnostics / query duration summaries,
+- threshold-gated p95 / max latency checks for the sqlite baseline.
+
+What it still does not prove:
+
+- long-horizon multi-host evidence,
+- calibrated final release thresholds across heterogeneous machines,
+- a declaration that A8 is fully production-closed.
+
 Code-vs-plan reality for this slice:
 
 | Area | Prior expectation | Current HEAD reality | Status |

docs/diataxis/zh/explanation/development-progress-dashboard.md

@@ -73,6 +73,26 @@ Tauri-first reply rendering 基线已交付：
 - 让新的共享 render substrate 在真实 browser/Tauri 运行时下持续经受验证，
 - 为后续 Godot 降级 / 物化保留干净边界，而不是重新把 Godot 约束塞回当前 Tauri UX。
 
+## 2026-05-27 Phase-1 SQLite Soak Gate 基线
+
+- embedded `graphdb/sqlite` 基线现在不再只停留在“重启连续性 + workload envelope”证明。
+- 当前 HEAD 还新增了一条专用的主机级 soak/performance verifier：
+  - `npm run verify:foundation:sqlite-runtime:soak`
+  - 会把结构化 JSON 报告写入 `output/verification/foundation-sqlite-runtime/`，
+  - 并把发布级证据与较轻量的 `smoke` / `medium` / `heavy` 矩阵验证分开。
+
+这条新门禁当前能够证明：
+
+- `dist` runtime 与 packaged sidecar 两条路径上的重复重启周期，
+- startup / ingest / readiness / diagnostics / query 五类时延统计摘要，
+- 面向 sqlite 基线的 p95 / max latency 阈值门禁。
+
+这条门禁当前仍不能单独证明：
+
+- 跨多宿主、长时间窗口的稳定证据，
+- 面向异构机器的最终发布阈值校准，
+- A8 已经完全达到 production-closed。
+
 这条切片的代码 / 方案现实矩阵：
 
 | 区域 | 先前期望 | 当前 HEAD 现实 | 状态 |

docs/en/implementation_plan.md

@@ -30,7 +30,8 @@ Bring the implementation plan back in line with `main` reality at three levels:
    - avoid reintroducing transition-only assumptions into CI gates,
    - keep fixrisk closure criteria tied to repo-controlled reality.
 2. **P1: release-grade graph/store hardening**
-   - extend the current operational `graphdb/sqlite` baseline from restart/workload proof into soak and performance closure.
+   - extend the current operational `graphdb/sqlite` baseline from restart/workload proof into soak and performance closure,
+   - keep a dedicated host-level soak gate (`verify:foundation:sqlite-runtime:soak`) so release-grade evidence is not diluted into the lighter matrix proof.
 3. **P2: release-grade ANN calibration**
    - keep the `external_http` connector green under workload proof, then close recall/latency threshold calibration.
 4. **P3: Tauri-first reply/render surface expansion**

docs/en/task.md

@@ -23,6 +23,7 @@ Primary references:
 - [x] Agent-workspace browser/runtime/Tauri verification closure is real and repeatable on the current branch.
 - [x] Phase-3 tutor telemetry, tutor trace/provider trend diagnostics, conversation memory, and memory-policy diagnostics now have concrete backend implementations.
 - [~] Phase-1 A8 has advanced to an embedded `graphdb/sqlite` operational baseline and now has restart-durability proof, host-level dist/runtime + packaged sidecar proof, and a host-level workload matrix across `smoke` / `medium` / `heavy`; soak / longer-duration / performance hardening still remain before production closure.
+- [~] Phase-1 A8 now also has a dedicated host-level soak/performance verifier path (`verify:foundation:sqlite-runtime:soak`) with structured report output, but release-grade closure still requires sustained threshold tuning and repeated host evidence rather than one passing command.
 - [~] Phase-1 A9 now has a live `external_http` sync-backed connector baseline under real query traffic, host-level dist/runtime + packaged sidecar proof, and a host-level workload matrix across `smoke` / `medium` / `heavy`, but benchmark-backed rollout thresholds and release-grade calibration still remain before production closure.
 - [x] `KnowledgeLearningPlatform.ts` no longer uses placeholder-backed runtime surfaces for query comparison, staleness, learning-quality, and session-plan-quality diagnostics.
 - [x] Server bootstrap now injects an active local `tutorAdapter`; the remaining tutor gap is production-proven multi-provider routing rather than default activation.
@@ -38,6 +39,7 @@ Primary references:
 - [x] Runtime path adaptation has been integrated for sidecar and frontend data roots.
 - [x] Worker runtime resolution has been stabilized for packaged sidecar scenarios.
 - [ ] Soak / longer-duration / performance hardening for the embedded graph backend baseline remain pending after the new packaged/runtime and workload-matrix proofs.
+- [ ] Promote the new sqlite soak verifier from initial host-level gate to sustained release evidence with repeated runs and tuned thresholds.
 - [ ] Production ANN connector threshold convergence and release-grade calibration remain pending after the new host-level runtime and workload-matrix proofs.
 - [ ] Phase-2 quality/query/session diagnostics now need release-grade calibration on top of a release-grade graphdb/ANN baseline.
 - [ ] Tutor routing now needs multi-provider hardening beyond the active local-first adapter path.
@@ -59,6 +61,8 @@ Primary references:
   - Same orchestration path, but also restores transient tracked `src-tauri/bin/server-*` dirtiness introduced by the current verification run so the worktree can be kept clean.
 - `npm run verify:foundation:sqlite-runtime:matrix`
   - Highest-value host/runtime proof for the embedded sqlite graph backend across `smoke` / `medium` / `heavy` workloads.
+- `npm run verify:foundation:sqlite-runtime:soak`
+  - Dedicated P1 host/runtime soak and performance gate for the embedded sqlite graph backend. Writes structured JSON reports under `output/verification/foundation-sqlite-runtime/`.
 - `npm run verify:foundation:ann-runtime:matrix`
   - Highest-value host/runtime proof for the `external_http` ANN connector across `smoke` / `medium` / `heavy` workloads.
 - `npm run verify:agent-workspace:browser`

docs/zh/implementation_plan.md

@@ -31,7 +31,8 @@
    - 不再让过渡期假设重新进入 CI 门禁，
    - fixrisk 的闭环标准始终绑定仓库可控现实。
 2. **P1：release-grade graph/store 加固**
-   - 把当前 `graphdb/sqlite` operational baseline 从重启/工作负载证明继续推进到 soak 与性能闭环。
+   - 把当前 `graphdb/sqlite` operational baseline 从重启/工作负载证明继续推进到 soak 与性能闭环，
+   - 保留独立的主机级 soak 门禁（`verify:foundation:sqlite-runtime:soak`），避免把发布级证据稀释到较轻量的矩阵验证里。
 3. **P2：release-grade ANN 校准**
    - 保持 `external_http` connector 在 workload proof 下稳定，再收口 recall/latency 阈值校准。
 4. **P3：Tauri-first reply/render surface 扩展**

docs/zh/task.md

@@ -19,6 +19,7 @@
 - [x] agent-workspace 的 browser/runtime/Tauri 验证闭环已经是真实可重复的。
 - [x] Phase-3 中的 tutor telemetry、tutor trace/provider trend、conversation memory、memory-policy diagnostics 已有具体后端实现。
 - [~] Phase-1 A8 已推进到 embedded `graphdb/sqlite` operational baseline，并已具备 shutdown/fresh restart 的重启耐久性证明、主机级 dist/runtime + packaged sidecar 证明，以及覆盖 `smoke` / `medium` / `heavy` 的主机级 workload matrix；但 soak、长时段与性能级加固仍未达到生产闭环。
+- [~] Phase-1 A8 现在也新增了独立的主机级 soak/performance verifier（`verify:foundation:sqlite-runtime:soak`）并输出结构化报告，但发布级闭环仍需要持续阈值校准与多轮主机证据，而不是单次通过命令就结束。
 - [~] Phase-1 A9 现已具备 live `external_http` sync-backed connector baseline，并在真实 query 流量下得到证明；主机级 dist/runtime + packaged sidecar 证明和 `smoke` / `medium` / `heavy` workload matrix 也已具备；但 benchmark rollout 阈值与发布级校准仍未完成。
 - [x] `KnowledgeLearningPlatform.ts` 中 query compare / staleness / learning-quality / session-plan-quality 运行面已不再返回 placeholder。
 - [x] `server.ts` 已注入激活态本地 `tutorAdapter`；剩余导师缺口已不再是默认激活，而是生产级多 provider 路由。
@@ -34,6 +35,7 @@
 - [x] Sidecar 与前端数据根路径的运行时适配已集成。
 - [x] 打包 Sidecar 场景下的 Worker 路径解析已稳定。
 - [ ] embedded graph backend 基线在新的 packaged/runtime 与主机级 workload matrix 证明之外，仍待补齐 soak、长时段与性能级加固。
+- [ ] 把新的 sqlite soak verifier 从“初始主机门禁”继续推进为“可持续的发布级证据”，补齐多轮运行与阈值调优。
 - [ ] 生产级 ANN connector 在新的主机级 runtime 与 workload matrix 证明之外，仍待补齐阈值收敛与发布级校准。
 - [ ] query/quality/session 运行面虽已真实接通，但仍需在发布级 graphdb/ANN 基线上完成发布级校准。
 - [ ] tutor 运行路径接下来要从 local-first 激活态 adapter 扩展为生产级多 provider 路由。
@@ -55,6 +57,8 @@
   - 与上述统一编排相同，但会额外回滚本次验证新引入的受跟踪 `src-tauri/bin/server-*` 脏改动，用于保持工作区 clean。
 - `npm run verify:foundation:sqlite-runtime:matrix`
   - 当前 embedded sqlite 图后端最有价值的主机/runtime 证明，覆盖 `smoke` / `medium` / `heavy` 三档 workload。
+- `npm run verify:foundation:sqlite-runtime:soak`
+  - 面向 P1 的专用 embedded sqlite 主机/runtime soak 与性能门禁，会把结构化 JSON 报告写到 `output/verification/foundation-sqlite-runtime/`。
 - `npm run verify:foundation:ann-runtime:matrix`
   - 当前 `external_http` ANN connector 最有价值的主机/runtime 证明，覆盖 `smoke` / `medium` / `heavy` 三档 workload。
 - `npm run verify:agent-workspace:browser`

package.json

@@ -97,6 +97,7 @@
     "verify:foundation:sqlite-runtime": "npm run build && node scripts/ensure-sidecar-ready.js && node scripts/verify-foundation-sqlite-runtime.js",
     "verify:foundation:sqlite-runtime:heavy": "npm run build && node scripts/ensure-sidecar-ready.js && node scripts/verify-foundation-sqlite-runtime.js --heavy",
     "verify:foundation:sqlite-runtime:matrix": "npm run build && node scripts/ensure-sidecar-ready.js && node scripts/verify-foundation-sqlite-runtime.js --matrix",
+    "verify:foundation:sqlite-runtime:soak": "npm run build && node scripts/ensure-sidecar-ready.js && node scripts/verify-foundation-sqlite-runtime.js --soak",
     "verify:core-real-machine": "node scripts/verify-core-real-machine-tests.js",
     "verify:core-real-machine:clean": "node scripts/verify-core-real-machine-tests.js --restore-sidecar-binaries",
     "verify:agent-workspace:browser": "node scripts/verify-agent-workspace-browser.js",