[wu-sheng]

Why

A trace could be linked four different ways — ?traceId= with a page-decided popout, &source=native|zipkin from the copy buttons, &traceType=SKYWALKING_NATIVE|OTLP markers (where OTLP meant the Zipkin popout), and ?traceqlId=&ds= for TraceQL — plus ?evaluationTraceId= on the Evaluation records tab. A link without a marker opened in whatever popout the page guessed, so a trace could be read from a store it is not in and shown as empty. This makes one form for every trace link, names the query that reads it, and stops guessing.

What

One link form: ?traceId=<id>&traceType=<source>. The source is one of the names a layer template's traces.sources already uses, and says which OAP query reads the trace: native (native trace query), zipkin (Zipkin query API), traceql-native / traceql-zipkin / traceql-otlp (TraceQL over that datasource). Span keys (traceSegmentId, traceSpanIndex, traceSpanId) and the window keys (traceAt, traceEnd, traceLookback) are unchanged.

Where	Before	Now
Native trace card ⧉ url (Traces tab, Trace inspect)	traceId=X&source=native	traceId=X&traceType=native
Zipkin trace card ⧉ url	…&source=zipkin&traceAt=T	…&traceType=zipkin&traceAt=T
TraceQL card ⧉ url	traceqlId=X&ds=D	traceId=X&traceType=traceql-D
Evaluation records row trace link	SKYWALKING_NATIVE / OTLP (Zipkin) / traceqlId=X&ds=otlp&traceqlSpanId=S	native / zipkin / traceId=X&traceType=traceql-otlp&traceSpanId=S
Record list trace icon	SKYWALKING_NATIVE / OTLP / traceqlId&ds / no store: bare id	the widget's own recordTrace.source / no store: bare id
Span detail cross-trace ref	traceId=X	traceId=X&traceType=native
Log rows (Logs, Log inspect, AI chat logs)	traceId=X&traceAt=T	traceId=X&traceType=<the page's source>&traceAt=T
Evaluation records trace condition	?evaluationTraceId=X, written as you type	read once from a trace link when the page opens or moves to another group

• Each popout opens only for its own source. A link with no traceType, or a value outside the five, opens a notice (TraceLinkNotice) that shows the id and the bad value and offers one button per source; the button sets traceType on the same link. Nothing falls back to a page default any more.
• A dashboard record list with no recordTrace — the bundled General, MySQL, PostgreSQL, Redis, Virtual Database and Virtual Cache lists among them — now opens that notice instead of the layer's popout. Bundled templates are untouched; the 1.1.0 changelog carries a Breaking entry with how to set the store.
• An id a Zipkin source cannot hold (a SkyWalking id with dashes, say) reads No trace with that id. at once in the TraceQL and Zipkin popouts and Pick span…, instead of The trace source is not answering. or a failed request. The BFF decides it with Zipkin's own id rule (normalizeZipkinTraceId, mirroring OAP's Span.normalizeTraceId) before asking OAP — the Zipkin trace route and the TraceQL zipkin datasource only — and answers notFound + invalidId; a 400 OAP itself returns stays a failure.
• Fix: a native span's cross-trace ref on a Zipkin page now stays native.
• Layer dashboard editor fix. Save dropped empty grids of scopes no page renders, even when the template began with them. General Service and Istio Managed SVCs ship such grids, so Reset to bundled → Save wrote OAP's copy back: Check diff & push stayed disabled while the sidebar marked the layer diverged. Save now keeps an empty grid the template began with and drops only one the edit created; a new test holds every bundled layer template to "Reset to bundled, Save, unchanged".
• Docs: operate/traces.md, operate/genai-evaluation.md, components/dashboard-widgets.md, customization/layer-templates.md; 1.1.0 changelog; new notice strings in all 8 locales.

e2e

• genai: URL checks use the new values; the trace-condition spec opens the tab from a trace link, closes the popout, and checks the condition was filled.
• ui/trace-deeplink: names the source, and adds bare id → notice → pick native → waterfall.
• istio/zipkin-traces: names zipkin. ui/popouts: a log-row link names native.
• The e2e cases run on ubuntu-24.04. ubuntu-latest is moving to 26.04 (Oct 19 – Nov 19, 2026), where kind cannot load the multi-platform OAP image into its node (ctr images import --all-platforms: content digest not found), so the kind cases (istio, deployment) failed before any test ran; the compose cases pass on either image. 24.04 stays supported while it is one of the two latest images.

Validation

• pnpm -r run type-check, UI and BFF builds, pnpm lint (repo-wide gate: i18n parity, docs links, source budget, design tokens, security, template + i18n validation), e2e spec type-check, license headers — green. pnpm -r run test:unit: api-client 201, conversation-view 163, BFF 2178, UI 1627.
• Live, against the public demo OAP (no proxy): the notice for a missing / unknown type; the chooser opening a native trace (61 spans) from a stored Virtual Database record list; No trace with that id. for a SkyWalking id asked of zipkin and traceql-zipkin; copy buttons on Trace inspect writing and reopening the new form; Istio Managed SVCs Reset to bundled → Save enabling Check diff & push (not pushed).
• Earlier in the work, the genai browser specs passed 9/9 against the local e2e stack. Not re-run on the final commit: the local Docker engine stopped answering, so the e2e cases are left to CI.
• Reviewed with Codex (gpt-6-astra, max) over several rounds, including whole-change impact evaluations; findings fixed, final review clean.