Retire the inference_turn event in favour of $ai_generation

Both events fired from the same hook with the same payload, so every
turn was reported twice, and only $ai_generation reaches PostHog's LLM
analytics views — inference_turn's numbers were readable in the raw
event stream and nowhere else. This is a deliberate removal of the
duplicate, not an oversight: nothing outside this repository consumed
inference_turn.

Author: TheGreatAxios

docs/TELEMETRY.md

@@ -6,13 +6,12 @@ includes prompts, code, file contents, or paths.
 
 ## What's collected
 
-Five events, each with a small set of properties:
+Four events, each with a small set of properties:
 
 | Event | When | Properties |
 |---|---|---|
 | `cli_start` | Once per used session (see First-run disclosure) | (none beyond common properties) |
 | `session_end` | When a TUI session finishes | `status`, `turn_count`, `duration_ms`, `session_mode`, `exit_reason` |
-| `inference_turn` | Once per completed turn | `provider_id`, `model_id`, `input_tokens`, `output_tokens`, `cache_read_tokens`, `cache_write_tokens`, `thinking_tokens`, `duration_ms` |
 | `$ai_generation` | Once per turn — on completion, and again on a turn that ends in an error | `$ai_trace_id`, `$ai_provider`, `$ai_model`, `$ai_input_tokens`, `$ai_output_tokens`, `$ai_latency`, `$ai_is_error`, `$ai_error`, `cache_read_tokens`, `cache_write_tokens`, `thinking_tokens` |
 | `$ai_span` | Once per top-level tool call in a completed turn | `$ai_trace_id`, `$ai_span_id`, `$ai_parent_id`, `$ai_span_name`, `$ai_is_error` |
 
@@ -26,10 +25,9 @@ request IP; no location data is collected by the client.
 Every event is capped to an explicit property allowlist before it leaves the
 process — no other field can ever be attached, even by accident.
 
-`provider_id` (and its AI-event equivalent `$ai_provider`) is the canonical
-provider kind resolved by the runtime (e.g. `openai-compatible`), never the
-free-text name you gave the provider in onboarding or settings. `model_id`
-(equivalently `$ai_model`) is the model identifier exactly as
+`$ai_provider` is the canonical provider kind resolved by the runtime (e.g.
+`openai-compatible`), never the free-text name you gave the provider in
+onboarding or settings. `$ai_model` is the model identifier exactly as
 configured — it is the one user-entered string that is sent, so do not put
 anything identifying in a model name.
 

src/telemetry/ai-observability.ts

@@ -68,6 +68,9 @@ export function emitAiObservability(
 
   telemetry.capture("$ai_generation", {
     $ai_trace_id: traceId,
+    // The canonical provider kind, never ctx.source.sourceId: sourceId is the
+    // user-typed label from onboarding/settings, and free text must not leave
+    // the process under the no-PII contract.
     $ai_provider: ctx.source.provider,
     $ai_model: ctx.source.model,
     $ai_input_tokens: ctx.usage.input,

src/telemetry/index.ts

@@ -42,7 +42,7 @@ export type BatchTuning = {
 export const TELEMETRY_NOTICE =
   "Anonymous usage telemetry is enabled (no prompts, code, or paths collected). Disable in /settings > Telemetry. Docs: docs/TELEMETRY.md";
 
-export type TelemetryEvent = "cli_start" | "session_end" | "inference_turn" | "$ai_generation" | "$ai_span";
+export type TelemetryEvent = "cli_start" | "session_end" | "$ai_generation" | "$ai_span";
 
 // Fixed enum of AI observability span names. The raw tool name is never sent
 // as a property: an MCP tool name carries the server identifier it was
@@ -78,16 +78,6 @@ export function getSessionId(): string {
 const EVENT_PROPERTY_ALLOWLIST: Record<TelemetryEvent, readonly string[]> = {
   cli_start: [],
   session_end: ["status", "turn_count", "duration_ms", "session_mode", "exit_reason"],
-  inference_turn: [
-    "provider_id",
-    "model_id",
-    "input_tokens",
-    "output_tokens",
-    "cache_read_tokens",
-    "cache_write_tokens",
-    "thinking_tokens",
-    "duration_ms",
-  ],
   // PostHog's LLM analytics views read the $ai_-prefixed properties and
   // nothing else, so every field these two events exist to surface has to
   // carry the documented name: an unprefixed property still arrives, but

src/tui/runner.ts

@@ -1404,19 +1404,6 @@ export async function runTUI(initialConfig: Config): Promise<number> {
     hookManager,
     initialTurnCount: resumeSeed.turnsUsed,
     onTurnComplete: (ctx) => {
-      // provider_id is the canonical provider kind, never ctx.source.sourceId:
-      // sourceId is the user-typed label from onboarding/settings, and free
-      // text must not leave the process under the no-PII contract.
-      getTelemetry().capture("inference_turn", {
-        provider_id: ctx.source.provider,
-        model_id: ctx.source.model,
-        input_tokens: ctx.usage.input,
-        output_tokens: ctx.usage.output,
-        cache_read_tokens: ctx.usage.cacheRead,
-        cache_write_tokens: ctx.usage.cacheWrite,
-        thinking_tokens: ctx.usage.thinking,
-        duration_ms: ctx.durationMs,
-      });
       emitAiObservability(getTelemetry(), ctx, {
         sessionId,
         subagentToolName: taskToolDefinition.name,

tests/unit/telemetry.test.ts

@@ -144,40 +144,6 @@ test("capture strips properties not in the event's allowlist", async () => {
   expect(body.properties.secret_field).toBeUndefined();
 });
 
-test("capture strips properties not in inference_turn's allowlist", async () => {
-  const { impl, events } = recordingFetch();
-  const telemetry = createTelemetry({
-    settings: settingsWith("id"),
-    env: {},
-    fetchFn: impl,
-    apiKey: "test-key",
-  });
-  telemetry.capture("inference_turn", {
-    provider_id: "anthropic",
-    model_id: "claude-x",
-    input_tokens: 10,
-    output_tokens: 20,
-    cache_read_tokens: 1,
-    cache_write_tokens: 2,
-    thinking_tokens: 3,
-    duration_ms: 400,
-    prompt: "should-not-appear",
-  });
-  await telemetry.flush();
-  expect(events().length).toBe(1);
-  const body = events()[0];
-  expect(body.event).toBe("inference_turn");
-  expect(body.properties.provider_id).toBe("anthropic");
-  expect(body.properties.model_id).toBe("claude-x");
-  expect(body.properties.input_tokens).toBe(10);
-  expect(body.properties.output_tokens).toBe(20);
-  expect(body.properties.cache_read_tokens).toBe(1);
-  expect(body.properties.cache_write_tokens).toBe(2);
-  expect(body.properties.thinking_tokens).toBe(3);
-  expect(body.properties.duration_ms).toBe(400);
-  expect(body.properties.prompt).toBeUndefined();
-});
-
 test("capture strips properties not in $ai_generation's allowlist", async () => {
   const { impl, events } = recordingFetch();
   const telemetry = createTelemetry({