trace-id propagation through /v1/iterate (multi-call observability)

Closes J's 2026-05-02 multi-call observability gap: a single
/v1/iterate session with N retries used to surface in Langfuse as
N+1 disconnected traces (one per /v1/chat hop + one for the iterate
request itself), with no parent/child linkage. Operators couldn't
scroll the retry chain in one trace tree to spot where grounding
failed.

## Wire-level change

- New header constant `shared.TraceIDHeader = "X-Lakehouse-Trace-Id"`
- `langfuseMiddleware` honors the header on inbound requests: if
  set, reuses that trace id instead of minting a new one. Stashes
  the trace id on the request context so handlers can attach
  application-level child spans.
- `validatord.chatCaller` forwards the header to chatd. Every chat
  hop in an iterate session lands as a child of the parent trace.

## Application-level spans

- `validator.IterateConfig` gains `Tracer` (optional callback).
  When wired, each iteration attempt emits one Langfuse span
  via `validator.AttemptSpan`:
    Name: iterate.attempt[N]
    Input: { iteration, model, provider, prompt }
    Output: { verdict, raw, error }
    Level: WARNING when verdict != accepted
- `validatord.iterTracer` is the production hook — bridges
  `validator.Tracer` → `langfuse.Client.Span`.
- `IterateRequest`/`IterateResponse`/`IterateFailure` gain
  `TraceID`; each `IterateAttempt` gains `SpanID`. The /v1/iterate
  caller can pivot from the JSON response straight into the
  Langfuse trace tree.

## What an operator sees post-cutover

  GET /v1/iterate {kind=fill, prompt=...} → Trace TR-1
    ├─ http.request span (from middleware)
    ├─ iterate.attempt[0] span (validator.Iterate emit)
    │     input: prompt+model
    │     output: { verdict: validation_failed, error: ..., raw }
    ├─ chatd /v1/chat call (X-Lakehouse-Trace-Id: TR-1)
    │     ├─ http.request span (chatd middleware)
    │     └─ chatd-internal spans (existing)
    ├─ iterate.attempt[1] span
    └─ ...

All in one Langfuse trace tree, not N+1 separate traces.

## Hallucinated-worker safety net is unchanged

The /v1/iterate flow's hard correctness gate is still
FillValidator + WorkerLookup. Phantom candidate IDs raise
ValidationError::Consistency which 422s and forces the iteration
loop to retry. The trace-id propagation is the OBSERVABILITY layer
on top — it makes the existing safety net's outcomes visible per-call,
not a replacement for it.

## Verification

- internal/validator: 4 new tests
  - TestIterate_TracerEmitsSpanPerAttempt — span/attempt count + SpanID
  - TestIterate_NoTraceIDSkipsTracer — no orphan spans without trace_id
  - TestIterate_ChatCallerReceivesTraceID — propagation contract
  - (existing iterate tests updated for new ChatCaller signature)
- internal/shared: 1 new test
  - TestLangfuseMiddleware_HonorsTraceIDHeader — cross-service linkage
- cmd/validatord: existing HTTP tests still PASS via the dual-shape
  UnmarshalJSON contract.
- validatord_smoke.sh: 5/5 PASS through gateway :3110 (unchanged).
- Full go test ./... green across 33 packages.

## Architecture invariant added

STATE_OF_PLAY "DO NOT RELITIGATE" gains a paragraph documenting
the X-Lakehouse-Trace-Id header contract + the iterate.attempt[N]
span emission. Future-Claude won't re-propose "wire trace-id
propagation" — the header IS the wiring.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

internal/shared/langfuse_middleware.go

@@ -1,6 +1,7 @@
 package shared
 
 import (
+	"context"
 	"net/http"
 	"os"
 	"time"
@@ -8,6 +9,30 @@ import (
 	"git.agentview.dev/profit/golangLAKEHOUSE/internal/langfuse"
 )
 
+// TraceIDHeader propagates a Langfuse trace id across services. When
+// validatord makes a /v1/iterate call that internally calls chatd's
+// /v1/chat, validatord sends this header so both daemons' middleware
+// emit spans under the SAME trace tree (rather than two unrelated
+// traces). Closes the multi-call observability gap J flagged
+// 2026-05-02 ("we need to make sure they have the corpus of
+// information to complete the process and we want to spot errors").
+const TraceIDHeader = "X-Lakehouse-Trace-Id"
+
+// traceIDCtxKey is the context value key for the per-request trace id.
+// Handlers downstream of langfuseMiddleware can pull it via TraceIDFromCtx
+// to attach child spans (e.g. iteration-attempt spans inside validatord).
+type traceIDCtxKey struct{}
+
+// TraceIDFromCtx returns the per-request Langfuse trace id, or "" if
+// the middleware didn't set one (Langfuse not configured / /health
+// bypass / no Client wired).
+func TraceIDFromCtx(ctx context.Context) string {
+	if v, ok := ctx.Value(traceIDCtxKey{}).(string); ok {
+		return v
+	}
+	return ""
+}
+
 // langfuseMiddleware emits one Langfuse trace per HTTP request, with
 // a single span carrying start/end timestamps + status code. Per
 // OPEN item #2 (closed by the wave that adds this file): production
@@ -36,15 +61,28 @@ func langfuseMiddleware(serviceName string, lf *langfuse.Client) func(http.Handl
 			start := time.Now()
 			sw := &statusWriter{ResponseWriter: w, status: http.StatusOK}
 
-			traceID := lf.Trace(r.Context(), langfuse.TraceInput{
-				Name: serviceName + " " + r.Method + " " + r.URL.Path,
-				Tags: []string{serviceName, r.Method},
-				Metadata: map[string]any{
-					"path":        r.URL.Path,
-					"method":      r.Method,
-					"remote_addr": r.RemoteAddr,
-				},
-			})
+			// If the caller forwarded a trace id (cross-service parent
+			// linkage) reuse it instead of starting a new trace. Spans
+			// from this service then attach to the parent trace tree
+			// so an /v1/iterate session shows as one trace with
+			// children for each /v1/chat hop.
+			traceID := r.Header.Get(TraceIDHeader)
+			if traceID == "" {
+				traceID = lf.Trace(r.Context(), langfuse.TraceInput{
+					Name: serviceName + " " + r.Method + " " + r.URL.Path,
+					Tags: []string{serviceName, r.Method},
+					Metadata: map[string]any{
+						"path":        r.URL.Path,
+						"method":      r.Method,
+						"remote_addr": r.RemoteAddr,
+					},
+				})
+			}
+
+			// Stash the trace id on the request context so downstream
+			// handlers can attach finer-grained spans (e.g. one per
+			// iteration attempt inside validator.Iterate).
+			r = r.WithContext(context.WithValue(r.Context(), traceIDCtxKey{}, traceID))
 
 			next.ServeHTTP(sw, r)
 
@@ -89,13 +127,18 @@ func (sw *statusWriter) WriteHeader(code int) {
 	sw.ResponseWriter.WriteHeader(code)
 }
 
-// loadLangfuseFromEnv builds a langfuse.Client from environment
+// LoadLangfuseFromEnv builds a langfuse.Client from environment
 // variables. Returns nil if any of LANGFUSE_URL / LANGFUSE_PUBLIC_KEY
 // / LANGFUSE_SECRET_KEY is unset (best-effort: missing config means
 // no tracing, never a startup error). Same env names as the bare
 // /etc/lakehouse/langfuse.env file used by the multi_coord_stress
 // driver — operators ship one env file across every daemon.
-func loadLangfuseFromEnv() *langfuse.Client {
+//
+// Exported 2026-05-02 so daemons that need to emit application-level
+// child spans (validatord's iterate-attempt spans) can hold their own
+// reference to the same client `shared.Run` is already wiring into
+// the middleware.
+func LoadLangfuseFromEnv() *langfuse.Client {
 	url := os.Getenv("LANGFUSE_URL")
 	pk := os.Getenv("LANGFUSE_PUBLIC_KEY")
 	sk := os.Getenv("LANGFUSE_SECRET_KEY")

internal/shared/langfuse_middleware_test.go

@@ -119,6 +119,50 @@ func TestLangfuseMiddleware_RealRequestEmitted(t *testing.T) {
 	}
 }
 
+// TestLangfuseMiddleware_HonorsTraceIDHeader locks the cross-service
+// trace-linkage contract: when validatord forwards a /v1/iterate's
+// trace id to chatd via X-Lakehouse-Trace-Id, chatd's middleware
+// MUST reuse that id rather than minting a new one. Without this,
+// every chat hop spawned its own orphan trace and an iterate session
+// with N retries showed in Langfuse as N+1 disconnected traces.
+//
+// We verify by inspecting the trace id stashed on the request
+// context — handlers downstream pull it via TraceIDFromCtx and
+// attach finer-grained spans to it.
+func TestLangfuseMiddleware_HonorsTraceIDHeader(t *testing.T) {
+	// nil client path is fine — TraceIDFromCtx still propagates the
+	// header value because the middleware sets the ctx value before
+	// minting any new id. But that's only true when the client is
+	// non-nil (the nil-client branch is a passthrough that never sets
+	// the ctx value). So we wire a real (mocked) Langfuse to exercise
+	// the header-honoring branch.
+	lfMock := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_, _ = io.ReadAll(r.Body)
+		w.WriteHeader(http.StatusOK)
+	}))
+	defer lfMock.Close()
+	lf := langfuse.New(lfMock.URL, "test-pk", "test-sk", nil)
+
+	mw := langfuseMiddleware("test-service", lf)
+	var observed string
+	h := mw(http.HandlerFunc(func(_ http.ResponseWriter, r *http.Request) {
+		observed = TraceIDFromCtx(r.Context())
+	}))
+	srv := httptest.NewServer(h)
+	defer srv.Close()
+
+	req, _ := http.NewRequest("POST", srv.URL+"/api/echo", nil)
+	req.Header.Set(TraceIDHeader, "trace-from-parent")
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		t.Fatalf("POST: %v", err)
+	}
+	resp.Body.Close()
+	if observed != "trace-from-parent" {
+		t.Errorf("expected forwarded trace id, got %q", observed)
+	}
+}
+
 // TestLangfuseMiddleware_StatusCaptured locks the status-writer
 // wrapping: when the handler returns 500, the middleware must see
 // 500 in the span output (otherwise error traces all show 200 and

internal/shared/server.go

@@ -80,7 +80,7 @@ func Run(serviceName, addr string, register RegisterRoutes, auth AuthConfig) err
 	// gets free production-traffic trace visibility when those env
 	// vars are set. Missing any of the three → nil client → middleware
 	// becomes a passthrough.
-	lf := loadLangfuseFromEnv()
+	lf := LoadLangfuseFromEnv()
 	if lf != nil {
 		// Make sure pending events flush on graceful shutdown so the
 		// last few requests' traces don't get lost.