Profiling page: per-turn phase timings, live in the web dashboard

The engine already tracks where each turn's wall time goes (expert-disk
service, I/O wait, expert matmul, attention, lm_head) — it just only spoke
at exit or under PROF=1. Stream it instead:

- glm.c: mux serve emits a per-turn "PROF" protocol line next to TIERS/HITS
  (window deltas per request, same convention as the STAT hit%); the phase
  window base is now always captured (a few loads per request).
- openai_server.py: parses PROF into a 120-turn rolling window and serves it
  at /profile (read-only, same trust level as /health).
- web: new Profiling tab — stat tiles (tok/s, wall, tokens/forward, disk
  service), wall-time composition bars for the last turn and the window,
  per-turn throughput and stacked phase columns with hover readouts, and a
  table of recent turns. Disk service is shown apart from the stack: it
  overlaps with compute, so only the I/O wait the compute thread felt counts
  inside wall time. Phase colours are a CVD-validated set with gaps + legend
  + table so identity never rides on colour alone.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WhTmF8yvEBgSkUKSVfZF7P

web/src/lib/api.test.ts

@@ -1,6 +1,6 @@
 import { afterEach, describe, expect, it, vi } from "vitest"
 
-import { extractSSE, getHealth, serverEndpoint, streamChat } from "./api"
+import { extractSSE, getHealth, getProfile, serverEndpoint, streamChat } from "./api"
 
 afterEach(() => vi.unstubAllGlobals())
 
@@ -36,6 +36,19 @@ describe("runtime API", () => {
       headers: expect.objectContaining({ Authorization: "Bearer secret" }),
     }))
   })
+
+  it("requests the profiling history next to the OpenAI v1 prefix", async () => {
+    const turn = {
+      wall_s: 2.5, prompt_tokens: 7, completion_tokens: 12,
+      expert_disk_s: 0.4, expert_wait_s: 0.1, expert_matmul_s: 0.9,
+      attention_s: 0.6, lm_head_s: 0.2, forwards: 15,
+    }
+    const fetchMock = vi.fn().mockResolvedValue(new Response(JSON.stringify({ seq: 1, turns: [turn] })))
+    vi.stubGlobal("fetch", fetchMock)
+
+    await expect(getProfile("http://localhost:8000/v1/")).resolves.toEqual({ seq: 1, turns: [turn] })
+    expect(fetchMock).toHaveBeenCalledWith("http://localhost:8000/profile", expect.anything())
+  })
 })
 
 describe("chat request extensions", () => {

web/src/lib/api.ts

@@ -49,6 +49,23 @@ export interface HealthResponse {
   hwinfo?: HwinfoHealth
 }
 
+export interface ProfileTurn {
+  wall_s: number
+  prompt_tokens: number
+  completion_tokens: number
+  expert_disk_s: number
+  expert_wait_s: number
+  expert_matmul_s: number
+  attention_s: number
+  lm_head_s: number
+  forwards: number
+}
+
+export interface ProfileResponse {
+  seq: number
+  turns: ProfileTurn[]
+}
+
 export interface TokenUsage {
   prompt_tokens: number
   completion_tokens: number
@@ -100,6 +117,12 @@ export async function getHealth(baseUrl: string, apiKey = "", signal?: AbortSign
   return (await response.json()) as HealthResponse
 }
 
+export async function getProfile(baseUrl: string, apiKey = "", signal?: AbortSignal): Promise<ProfileResponse> {
+  const response = await fetch(serverEndpoint(baseUrl, "profile"), { headers: headers(apiKey), signal })
+  if (!response.ok) throw new Error(await responseError(response))
+  return (await response.json()) as ProfileResponse
+}
+
 export function extractSSE(buffer: string) {
   const frames = buffer.split(/\r?\n\r?\n/)
   const rest = frames.pop() || ""