docs(kpi): K17 — #26 KpiHistory component doc + README/CLAUDE + cross-component interactions + completion-design update

docs/plans/2026-06-15-stillpending-completion-design.md

@@ -78,8 +78,11 @@ Wire up behavior that exists in code but is never started, and fill the event-lo
 #### M5 — Audit hardening (T1–T8)
 Hash-chain tamper evidence (off by default, `verify-chain` made real); Parquet export/archival (replace the 501); per-channel retention overrides; tag-cascade for `ParentExecutionId` (thread writing-execution id through trigger-driven runs); ExecutionId/ParentExecutionId + SourceNode backfill on historical rows; per-node stuck-count KPIs; structured response capture (headers/content-type, inbound request headers, per-method opt-out, `AuditInboundCeilingHits` metric); CLI `audit tree`.
 
-#### M6 — Notifications (T9–T11)
-Teams + other non-Email delivery adapters behind the existing `INotificationDeliveryAdapter` seam; `NotificationType` enum values; Central UI notification-list `Type` selector; historical/trend KPI charts (introduce a time-series store).
+#### M6 — KPI History & Trends (T11 delivered; T9/T10 deferred)
+Reshaped during the 2026-06-17 brainstorm (see `docs/plans/2026-06-17-m6-kpi-history-design.md`):
+- **T11 — DELIVERED** as the reusable **KPI-history backbone** (#26 KpiHistory), promoted from a notifications-only feature. A tall/EAV `KpiSample` store in **central MS SQL** (no new infra — supersedes the original "point-in-time only, no time-series store" stance), a `KpiHistoryRecorderActor` cluster singleton (`kpi-history-recorder`, not readiness-gated, best-effort with per-source isolation) sampling DI-registered `IKpiSampleSource`s every minute, a bucketed `GetRawSeriesAsync` + `KpiSeriesBucketer` query + scoped `KpiHistoryQueryService`, and a reusable custom-SVG `KpiTrendChart` (no third-party charting lib). Trends shipped for **all** current KPI sources — Notification Outbox, Site Call Audit, Audit Log, and Site Health — across four UI surfaces.
+- **T9 (Teams + other non-Email delivery adapters behind `INotificationDeliveryAdapter`) — DEFERRED to the next major version.** The seam exists; no code now. Transport choice (Incoming Webhook vs Microsoft Graph) and the Teams list-targeting model remain to be designed.
+- **T10 (`NotificationType` enum values + Central UI notification-list `Type` selector) — DEFERRED with T9.** A Type selector has no purpose until a second delivery type exists.
 
 #### M7 — OPC UA / MxGateway UX (T13–T17)
 Dedicated operator Alarm Summary page; MxGateway secured writes (operator+verifier); OPC UA address-space search + `BrowseNext` paging; type-info surfacing + bulk override CSV import; "Verify endpoint" connectivity button + cert-management UI.
@@ -96,7 +99,7 @@ Template tree search/filter; folder drag-drop + sibling reorder + root context m
 ## Dependencies & sequencing
 
 - **M1 → M5** — audit hardening builds on the wired purge/reconciliation.
-- **M6/T11** — depends on introducing a time-series store (new infra; size carefully).
+- **M6/T11** — delivered as the #26 KpiHistory backbone; reused **central MS SQL** (a tall/EAV `KpiSample` table) rather than introducing new infra. T9/T10 deferred to the next major version.
 - **M9/T26** — base-template versioning is the largest authoring item; may split.
 - **M4** — runs anytime; cheap and high-clarity, good to interleave.
 - **M3** — independent; can run in parallel with M1/M2.
@@ -121,7 +124,7 @@ Template tree search/filter; folder drag-drop + sibling reorder + root context m
 ## Open items / risks
 
 - M3 real-compile may surface latent invalid scripts in existing templates/fixtures — budget for fixture cleanup.
-- M6 time-series store is the one genuinely-new piece of infrastructure; scope it deliberately (could reuse MS SQL with a rollup table rather than a new dependency).
+- M6 KPI history (resolved): reused **central MS SQL** with a tall/EAV `KpiSample` table rather than a new dependency, so no genuinely-new infrastructure was introduced.
 - The Phase 2 roadmap is large; treat each milestone as a separate planning + implementation pass, not a single mega-effort.
 
 ## Next step