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/validator/iterate.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"time"
 )
 
 // IterateRequest is the input to Iterate. Mirrors Rust's
@@ -20,14 +21,24 @@ type IterateRequest struct {
 	MaxIterations int            `json:"max_iterations,omitempty"`
 	Temperature   *float64       `json:"temperature,omitempty"`
 	MaxTokens     int            `json:"max_tokens,omitempty"`
+	// TraceID lets external schedulers force a parent trace id (so the
+	// iterate session lands as a child of an upstream Langfuse trace).
+	// Empty = the caller didn't provide one; validatord may still
+	// generate one from request context.
+	TraceID string `json:"trace_id,omitempty"`
 }
 
 // IterateAttempt is one row in the history. raw is capped at 2000
-// chars on the wire to keep responses bounded.
+// chars on the wire to keep responses bounded. SpanID points at the
+// Langfuse span emitted for this attempt (if a Tracer was wired in
+// IterateConfig); empty when tracing was disabled. Operators reading
+// `data/_kb/replay_runs.jsonl` or staffing-co-pilot session logs can
+// jump from a row to its full trace tree via this id.
 type IterateAttempt struct {
-	Iteration int            `json:"iteration"`
-	Raw       string         `json:"raw"`
-	Status    AttemptStatus  `json:"status"`
+	Iteration int           `json:"iteration"`
+	Raw       string        `json:"raw"`
+	Status    AttemptStatus `json:"status"`
+	SpanID    string        `json:"span_id,omitempty"`
 }
 
 // AttemptStatus is the per-attempt verdict. Tagged JSON so consumers
@@ -38,11 +49,14 @@ type AttemptStatus struct {
 }
 
 // IterateResponse is the success payload (200 + Report + accepted artifact).
+// TraceID echoes back the parent trace id so callers can pivot to
+// Langfuse and see the full session (chat hops + validator verdicts).
 type IterateResponse struct {
 	Artifact   map[string]any   `json:"artifact"`
 	Validation Report           `json:"validation"`
 	Iterations int              `json:"iterations"`
 	History    []IterateAttempt `json:"history"`
+	TraceID    string           `json:"trace_id,omitempty"`
 }
 
 // IterateFailure is the max-iter-exhausted payload (422 + history).
@@ -50,19 +64,53 @@ type IterateFailure struct {
 	Error      string           `json:"error"`
 	Iterations int              `json:"iterations"`
 	History    []IterateAttempt `json:"history"`
+	TraceID    string           `json:"trace_id,omitempty"`
 }
 
 // ChatCaller is the seam Iterate uses to invoke an LLM. Tests inject
 // scripted callers; production wires this to the chatd /v1/chat HTTP
 // endpoint. Implementations must return the model's textual content
-// (no choices wrapper, no message envelope).
-type ChatCaller func(ctx context.Context, system, user, provider, model string, temperature *float64, maxTokens int) (string, error)
+// (no choices wrapper, no message envelope). traceID is the Langfuse
+// parent trace id — HTTP transports forward it as the
+// `X-Lakehouse-Trace-Id` header so chatd's middleware attaches its
+// spans to the same trace.
+type ChatCaller func(ctx context.Context, system, user, provider, model string, temperature *float64, maxTokens int, traceID string) (string, error)
+
+// AttemptSpan describes one iteration's chat-call + validator-verdict.
+// The Tracer (see IterateConfig) consumes these to emit Langfuse
+// spans — one span per attempt, plus inputs/outputs so an operator
+// scrolling Langfuse sees the full prompt → raw → verdict chain
+// without having to join multiple JSONL files.
+type AttemptSpan struct {
+	TraceID    string
+	ParentID   string // span id of the surrounding /v1/iterate http request, if any
+	Iteration  int
+	Model      string
+	Provider   string
+	Prompt     string
+	Raw        string
+	Verdict    AttemptStatus
+	StartTime  time.Time
+	EndTime    time.Time
+}
+
+// Tracer is the hook Iterate calls after each attempt. Returns the
+// span id so it lands in IterateAttempt.SpanID. nil = tracing
+// disabled (tests, or production with Langfuse unconfigured).
+//
+// Production wiring lives in cmd/validatord/main.go and routes to
+// internal/langfuse.Client — kept as an interface here so the
+// validator package doesn't depend on langfuse for unit tests.
+type Tracer func(span AttemptSpan) string
 
 // IterateConfig threads daemon-level settings into the orchestrator.
 type IterateConfig struct {
 	DefaultMaxIterations int
 	DefaultMaxTokens     int
 	DefaultTemperature   float64
+	// Tracer (optional) emits one span per iteration attempt for
+	// per-call visibility. nil = no tracing.
+	Tracer Tracer
 }
 
 const (
@@ -101,36 +149,59 @@ func Iterate(ctx context.Context, req IterateRequest, cfg IterateConfig, chat Ch
 
 	currentPrompt := req.Prompt
 	history := make([]IterateAttempt, 0, maxIter)
+	traceID := req.TraceID
+
+	// recordAttempt: append to history, emit a Langfuse span if a
+	// Tracer is wired, return the SpanID. Centralized so every code
+	// path (no_json, accepted, validation_failed) hits the same
+	// observability path — no easy way to forget tracing on a branch.
+	recordAttempt := func(iteration int, raw string, verdict AttemptStatus, started time.Time) string {
+		spanID := ""
+		if cfg.Tracer != nil && traceID != "" {
+			spanID = cfg.Tracer(AttemptSpan{
+				TraceID:   traceID,
+				Iteration: iteration,
+				Model:     req.Model,
+				Provider:  req.Provider,
+				Prompt:    currentPrompt,
+				Raw:       raw,
+				Verdict:   verdict,
+				StartTime: started,
+				EndTime:   time.Now(),
+			})
+		}
+		history = append(history, IterateAttempt{
+			Iteration: iteration,
+			Raw:       trim(raw, 2000),
+			Status:    verdict,
+			SpanID:    spanID,
+		})
+		return spanID
+	}
 
 	for i := 0; i < maxIter; i++ {
-		raw, err := chat(ctx, req.System, currentPrompt, req.Provider, req.Model, temp, maxTokens)
+		started := time.Now()
+		raw, err := chat(ctx, req.System, currentPrompt, req.Provider, req.Model, temp, maxTokens, traceID)
 		if err != nil {
 			return nil, nil, fmt.Errorf("/v1/chat hop failed at iter %d: %w", i, err)
 		}
 
 		artifact := ExtractJSON(raw)
 		if artifact == nil {
-			history = append(history, IterateAttempt{
-				Iteration: i,
-				Raw:       trim(raw, 2000),
-				Status:    AttemptStatus{Kind: "no_json"},
-			})
+			recordAttempt(i, raw, AttemptStatus{Kind: "no_json"}, started)
 			currentPrompt = req.Prompt + "\n\nYour previous attempt did not contain a JSON object. Reply with ONLY a valid JSON object matching the requested artifact shape."
 			continue
 		}
 
 		report, vErr := validate(req.Kind, artifact)
 		if vErr == nil {
-			history = append(history, IterateAttempt{
-				Iteration: i,
-				Raw:       trim(raw, 2000),
-				Status:    AttemptStatus{Kind: "accepted"},
-			})
+			recordAttempt(i, raw, AttemptStatus{Kind: "accepted"}, started)
 			return &IterateResponse{
 				Artifact:   artifact,
 				Validation: report,
 				Iterations: i + 1,
 				History:    history,
+				TraceID:    traceID,
 			}, nil, nil
 		}
 
@@ -138,11 +209,7 @@ func Iterate(ctx context.Context, req IterateRequest, cfg IterateConfig, chat Ch
 		// The model sees concrete failure mode + retries with corrective
 		// context. Same "validator IS the observer" shape as Phase 43.
 		errSummary := vErr.Error()
-		history = append(history, IterateAttempt{
-			Iteration: i,
-			Raw:       trim(raw, 2000),
-			Status:    AttemptStatus{Kind: "validation_failed", Error: errSummary},
-		})
+		recordAttempt(i, raw, AttemptStatus{Kind: "validation_failed", Error: errSummary}, started)
 		currentPrompt = req.Prompt + "\n\nPrior attempt failed validation:\n" + errSummary + "\n\nFix the specific issue above and respond with a corrected JSON object."
 	}
 
@@ -150,6 +217,7 @@ func Iterate(ctx context.Context, req IterateRequest, cfg IterateConfig, chat Ch
 		Error:      fmt.Sprintf("max iterations reached (%d) without passing validation", maxIter),
 		Iterations: maxIter,
 		History:    history,
+		TraceID:    traceID,
 	}, nil
 }
 

internal/validator/iterate_test.go

@@ -3,6 +3,7 @@ package validator
 import (
 	"context"
 	"errors"
+	"fmt"
 	"testing"
 )
 
@@ -69,7 +70,7 @@ func TestExtractJSON_TopLevelArrayReturnsFirstInnerObject(t *testing.T) {
 
 func scriptedChat(responses ...string) (ChatCaller, *int) {
 	idx := 0
-	return func(_ context.Context, _, _ string, _, _ string, _ *float64, _ int) (string, error) {
+	return func(_ context.Context, _, _ string, _, _ string, _ *float64, _ int, _ string) (string, error) {
 		if idx >= len(responses) {
 			return "", errors.New("scripted chat exhausted")
 		}
@@ -175,8 +176,96 @@ func TestIterate_MaxIterationsExhaustedReturnsFailure(t *testing.T) {
 	}
 }
 
+// TestIterate_TracerEmitsSpanPerAttempt locks the per-attempt
+// observability contract: when a Tracer is wired AND a TraceID is
+// present, every retry produces exactly one span and the SpanID
+// lands on the IterateAttempt history row.
+func TestIterate_TracerEmitsSpanPerAttempt(t *testing.T) {
+	chat, _ := scriptedChat(
+		"no json",
+		`{"k":"v"}`,
+	)
+	validate := func(_ string, _ map[string]any) (Report, error) { return Report{}, nil }
+
+	var spans []AttemptSpan
+	tracer := func(s AttemptSpan) string {
+		spans = append(spans, s)
+		return fmt.Sprintf("span-%d", s.Iteration)
+	}
+
+	resp, _, err := Iterate(context.Background(),
+		IterateRequest{Kind: "playbook", Prompt: "X", Provider: "ollama", Model: "qwen", TraceID: "trace-abc"},
+		IterateConfig{Tracer: tracer}, chat, validate)
+	if err != nil {
+		t.Fatalf("Iterate: %v", err)
+	}
+	if resp.TraceID != "trace-abc" {
+		t.Errorf("TraceID echo: got %q, want trace-abc", resp.TraceID)
+	}
+	if len(spans) != 2 {
+		t.Fatalf("expected 2 emitted spans, got %d", len(spans))
+	}
+	if spans[0].Verdict.Kind != "no_json" || spans[1].Verdict.Kind != "accepted" {
+		t.Errorf("verdict propagation: %+v", spans)
+	}
+	if resp.History[0].SpanID != "span-0" || resp.History[1].SpanID != "span-1" {
+		t.Errorf("SpanID stamping: %+v", resp.History)
+	}
+	if spans[0].TraceID != "trace-abc" {
+		t.Errorf("TraceID didn't reach the tracer: %q", spans[0].TraceID)
+	}
+}
+
+// TestIterate_NoTraceIDSkipsTracer guards the "Langfuse not configured"
+// path: tracer is non-nil but trace_id is empty → we MUST NOT emit
+// orphan spans.
+func TestIterate_NoTraceIDSkipsTracer(t *testing.T) {
+	chat, _ := scriptedChat(`{"k":"v"}`)
+	validate := func(_ string, _ map[string]any) (Report, error) { return Report{}, nil }
+	called := 0
+	tracer := func(_ AttemptSpan) string {
+		called++
+		return "should-not-be-used"
+	}
+	resp, _, err := Iterate(context.Background(),
+		IterateRequest{Kind: "playbook", Prompt: "X", Provider: "ollama", Model: "qwen"}, // no TraceID
+		IterateConfig{Tracer: tracer}, chat, validate)
+	if err != nil {
+		t.Fatalf("Iterate: %v", err)
+	}
+	if called != 0 {
+		t.Errorf("tracer should not fire without a trace_id; called=%d", called)
+	}
+	if resp.History[0].SpanID != "" {
+		t.Errorf("SpanID should be empty when no trace_id: %q", resp.History[0].SpanID)
+	}
+}
+
+// TestIterate_ChatCallerReceivesTraceID confirms the trace_id is
+// forwarded into the ChatCaller signature so HTTP transports can set
+// the X-Lakehouse-Trace-Id header. Without this, chatd's Langfuse
+// emit would create an orphan trace per retry instead of nesting
+// under the iterate parent.
+func TestIterate_ChatCallerReceivesTraceID(t *testing.T) {
+	var observedTrace string
+	chat := func(_ context.Context, _, _ string, _, _ string, _ *float64, _ int, traceID string) (string, error) {
+		observedTrace = traceID
+		return `{"k":"v"}`, nil
+	}
+	validate := func(_ string, _ map[string]any) (Report, error) { return Report{}, nil }
+	_, _, err := Iterate(context.Background(),
+		IterateRequest{Kind: "playbook", Prompt: "X", Provider: "ollama", Model: "x", TraceID: "trace-xyz"},
+		IterateConfig{}, chat, validate)
+	if err != nil {
+		t.Fatalf("Iterate: %v", err)
+	}
+	if observedTrace != "trace-xyz" {
+		t.Errorf("ChatCaller should receive trace_id; got %q", observedTrace)
+	}
+}
+
 func TestIterate_PropagatesChatInfraError(t *testing.T) {
-	chat := func(_ context.Context, _, _ string, _, _ string, _ *float64, _ int) (string, error) {
+	chat := func(_ context.Context, _, _ string, _, _ string, _ *float64, _ int, _ string) (string, error) {
 		return "", errors.New("connection refused")
 	}
 	validate := func(_ string, _ map[string]any) (Report, error) { return Report{}, nil }