Proposal: OpenTelemetry semantic conventions for Content Telemetry

proposals/opentelemetry-semantic-conventions.md

@@ -0,0 +1,33 @@
+Status: discussing
+
+Proposed by: Alex Springer
+
+# OpenTelemetry names for Content Telemetry
+
+**What I'm seeing**
+
+- Agents already record OpenTelemetry. None of it says what content was grounded, cited or shown.
+- The OpenTelemetry GenAI conventions have no citation or provenance fields, only `gen_ai.data_source.id` and a retrieved document's id and score.
+- OpenTelemetry now wants domain conventions in separate registries ([OTEP 4815](https://github.com/open-telemetry/opentelemetry-specification/blob/main/oteps/4815-semantic-conventions-schema-v2.md)). The GenAI conventions moved out on that basis.
+
+**What I'd like**
+
+- An OpenTelemetry registry mapping the five content events, turns and fields to OpenTelemetry names. The specification stays normative.
+- It lives in the neutral contenttelemetry GitHub org, with a one-way dependency on the specification, like the profiles.
+- A working draft exists and passes these checks. It will be published there if this is aligned.
+
+**Why it matters**
+
+- Implementers report from instrumentation they already have.
+- The section 5.7.5 rules become a check anyone can run against their own emitter.
+- It gives us something concrete to take to the OpenTelemetry GenAI SIG.
+
+**Questions for discussion**
+
+- `content_telemetry.*` or `content.*` as the prefix.
+- Turns as spans only.
+- OpenTelemetry signals are not conformance until converted to section 7.1 documents. Is that line clear enough?
+
+**Declared interest**
+
+- SPUR tech lead and Content Telemetry maintainer.