diff --git a/adr/0003-opentelemetry-as-the-operational-record.md b/adr/0003-opentelemetry-as-the-operational-record.md
index 56bd95e..bca7def 100644
--- a/adr/0003-opentelemetry-as-the-operational-record.md
+++ b/adr/0003-opentelemetry-as-the-operational-record.md
@@ -8,7 +8,7 @@ OpenTelemetry spans correlate workflow, process and model operations with a run
 
 ## Current implementation
 
-Orbit exports workflow, step, remote-invocation and model spans to a local JSONL OTEL exporter. Run metadata, console output and evidence are retained with the local run record.
+Orbit exports workflow, step, remote-invocation, supervisor evaluation and cycle-review model spans to a local JSONL OTEL exporter. It includes redaction-safe request/response fingerprints and lengths, lifecycle metadata and outcomes. Run metadata, console output and full prompt/response evidence are retained with the local run record.
 
 ## Planned work
 
diff --git a/backend/app/store.py b/backend/app/store.py
index 6b11632..1ec42e3 100644
--- a/backend/app/store.py
+++ b/backend/app/store.py
@@ -16,10 +16,12 @@
 from datetime import UTC, datetime, timedelta
 from pathlib import Path
 from typing import Any
+from urllib.parse import urlparse
 from zoneinfo import ZoneInfo, ZoneInfoNotFoundError
 
 import requests
 import yaml
+from opentelemetry.trace import Status, StatusCode
 
 from orbit import load_bundle
 
@@ -32,6 +34,15 @@
 ROOT = Path(__file__).resolve().parents[2]
 
 
+def _telemetry_text_metadata(prefix: str, value: str) -> dict[str, str | int]:
+    """Describe retained model text without exporting its potentially sensitive contents."""
+    encoded = value.encode("utf-8")
+    return {
+        f"{prefix}.length": len(value),
+        f"{prefix}.sha256": hashlib.sha256(encoded).hexdigest(),
+    }
+
+
 def _application_data_pointer() -> Path:
     """Keep an operator-selected data location outside the data it points to."""
     if os.name == "nt":
@@ -4342,6 +4353,10 @@ def _execute(self, run_id: str) -> None:
                 "workflow.id": workflow.id,
                 "run.id": run_id,
                 "workflow.kind": workflow.kind,
+                "orbit.execution.mode": run.execution_mode,
+                "orbit.execution.type": run.execution_type,
+                "orbit.loop.limit": run.loop_limit,
+                "orbit.build.id": run.build_id or "",
             },
         ) as workflow_span:
             trace_id = f"{workflow_span.get_span_context().trace_id:032x}"
@@ -4721,16 +4736,24 @@ def supervisor_result(result: object) -> object:
             "supervisor.evaluate",
             attributes={
                 "run.id": run_id,
+                "gen_ai.operation.name": "chat",
                 "gen_ai.provider.name": settings.provider,
                 "gen_ai.request.model": settings.model,
                 "orbit.manager.template": (
                     self.build(run.build_id).get("manager_template_id") if run.build_id else ""
                 ),
                 "orbit.iteration": iteration,
+                "orbit.candidate.id": candidate_id or "",
+                "orbit.evidence.step_count": len(cycle_evidence),
+                **_telemetry_text_metadata("gen_ai.request.prompt", supervisor_prompt),
             },
         ) as span:
             try:
-                result = self._validated_supervisor_result(provider.complete(settings, supervisor_prompt))
+                span.add_event("gen_ai.request.sent")
+                response_text = provider.complete(settings, supervisor_prompt)
+                span.set_attributes(_telemetry_text_metadata("gen_ai.response", response_text))
+                span.add_event("gen_ai.response.received")
+                result = self._validated_supervisor_result(response_text)
                 evaluation = result.get("evaluation")
                 if evaluation is not None:
                     threshold = (
@@ -4767,7 +4790,16 @@ def supervisor_result(result: object) -> object:
                 self._review_cycle_improvement(run, iteration, result, settings, provider)
                 span.set_attribute("orbit.supervisor.improvements", len(result["improvements"]))
                 span.set_attribute("orbit.supervisor.reported_issues", len(result["reported_issues"]))
-                span.add_event("supervisor.response.validated")
+                if isinstance(evaluation, dict):
+                    span.set_attribute("orbit.supervisor.score", evaluation.get("score", 0))
+                    span.set_attribute("orbit.supervisor.approval", str(evaluation.get("approval", "")))
+                span.add_event(
+                    "supervisor.response.validated",
+                    {
+                        "orbit.supervisor.improvements": len(result["improvements"]),
+                        "orbit.supervisor.reported_issues": len(result["reported_issues"]),
+                    },
+                )
             except ValueError as error:
                 run = self._load(run_id)
                 run.supervisor_status, run.supervisor_error, run.updated_at = (
@@ -4786,7 +4818,8 @@ def supervisor_result(result: object) -> object:
                 )
                 self._save(run)
                 span.record_exception(error)
-                span.add_event("supervisor.response.invalid", {"reason": str(error)})
+                span.set_status(Status(StatusCode.ERROR))
+                span.add_event("supervisor.response.invalid", {"error.type": type(error).__name__})
             except (RuntimeError, requests.RequestException) as error:
                 run = self._load(run_id)
                 run.supervisor_status, run.supervisor_error, run.updated_at = "failed", str(error), now()
@@ -4801,7 +4834,8 @@ def supervisor_result(result: object) -> object:
                 )
                 self._save(run)
                 span.record_exception(error)
-                span.add_event("supervisor.request.failed", {"reason": str(error)})
+                span.set_status(Status(StatusCode.ERROR))
+                span.add_event("supervisor.request.failed", {"error.type": type(error).__name__})
 
     def _review_cycle_improvement(
         self, run: Run, iteration: int, result: dict[str, Any], settings: ModelSettings, provider: Any
@@ -4819,10 +4853,29 @@ def _review_cycle_improvement(
             )
         )
         try:
-            reviewed = json.loads(provider.complete(settings, prompt))
-            interventions = reviewed.get("interventions", []) if isinstance(reviewed, dict) else []
-            if not isinstance(interventions, list):
-                return
+            with self.tracer.start_as_current_span(
+                "supervisor.cycle_review",
+                attributes={
+                    "run.id": run.id,
+                    "gen_ai.operation.name": "chat",
+                    "gen_ai.provider.name": settings.provider,
+                    "gen_ai.request.model": settings.model,
+                    "orbit.iteration": iteration,
+                    **_telemetry_text_metadata("gen_ai.request.prompt", prompt),
+                },
+            ) as span:
+                span.add_event("gen_ai.request.sent")
+                response_text = provider.complete(settings, prompt)
+                span.set_attributes(_telemetry_text_metadata("gen_ai.response", response_text))
+                span.add_event("gen_ai.response.received")
+                reviewed = json.loads(response_text)
+                interventions = reviewed.get("interventions", []) if isinstance(reviewed, dict) else []
+                if not isinstance(interventions, list):
+                    span.set_status(Status(StatusCode.ERROR))
+                    span.add_event("supervisor.cycle_review.invalid_response")
+                    return
+                span.set_attribute("orbit.cycle_review.intervention_count", len(interventions))
+                span.add_event("supervisor.cycle_review.validated")
             stored = self.cycle_interventions()
             for intervention in interventions:
                 if not isinstance(intervention, dict) or not isinstance(intervention.get("title"), str):
@@ -4861,6 +4914,11 @@ def _execute_step(
                 "run.id": run_id,
                 "step.id": step.id,
                 "step.phase": step.phase,
+                "orbit.iteration": loop_index,
+                "orbit.candidate.id": candidate_id or "",
+                "process.command.executable": Path(step.command[0]).name if step.command else "",
+                "process.command.argument_count": max(0, len(step.command) - 1),
+                "process.timeout.seconds": step.timeout_seconds,
             },
         ) as span:
             run = self._load(run_id)
@@ -4922,6 +4980,7 @@ def _execute_step(
                     creationflags=creation_flags,
                     env=environment,
                 )
+                span.add_event("process.started", {"process.pid": process.pid})
                 interruption_timer: threading.Timer | None = None
                 if (
                     step.phase == "verify"
@@ -5154,7 +5213,13 @@ def capture_output() -> None:
                 run.pid, run.updated_at = None, now()
                 self._save(run)
                 span.add_event("process.completed", {"process.exit_code": process.returncode})
+                span.set_attribute("process.exit_code", process.returncode)
+                span.set_attribute("process.output.line_count", len(captured_lines))
+                span.set_attribute("orbit.result.has_structured_output", structured_result is not None)
+                span.set_attribute("orbit.result.target_log_count", len(target_logs))
+                span.set_attribute("orbit.result.data_file_count", len(data_files))
                 if process.returncode and step.on_failure == "stop":
+                    span.set_status(Status(StatusCode.ERROR))
                     if step.phase == "verify" and run.advance_requested:
                         run.advance_requested = False
                         self._save(run)
@@ -5169,12 +5234,14 @@ def capture_output() -> None:
             except subprocess.TimeoutExpired:
                 self._stop_process_group(process, force=True)
                 span.add_event("process.timeout", {"timeout.seconds": step.timeout_seconds})
+                span.set_status(Status(StatusCode.ERROR))
                 if self._load(run_id).status == "cancelled":
                     return
                 self._fail(self._load(run_id), step.id, f"timed out after {step.timeout_seconds}s")
                 return
             except ValueError as error:
-                span.add_event("step.rejected", {"reason": str(error)})
+                span.add_event("step.rejected", {"error.type": type(error).__name__})
+                span.set_status(Status(StatusCode.ERROR))
                 self._fail(self._load(run_id), step.id, str(error))
                 return
             finally:
@@ -5359,7 +5426,16 @@ def test_build(self, build_id: str, output_locale: str | None = None) -> Run:
 
     def _execute_remote(self, run_id: str, executor: dict[str, Any]) -> None:
         run = self._load(run_id)
-        with self.tracer.start_as_current_span("remote.agent.run", attributes={"run.id": run_id}) as span:
+        endpoint = str(executor.get("endpoint", ""))
+        with self.tracer.start_as_current_span(
+            "remote.agent.run",
+            attributes={
+                "run.id": run_id,
+                "http.request.method": str(executor.get("method", "POST")),
+                "server.address": urlparse(endpoint).hostname or "",
+                "http.request.timeout_seconds": int(executor.get("timeout_seconds", 0) or 0),
+            },
+        ) as span:
             run.status, run.current_phase, run.telemetry_trace_id, run.updated_at = (
                 "running",
                 "execute",
@@ -5380,6 +5456,10 @@ def _execute_remote(self, run_id: str, executor: dict[str, Any]) -> None:
                 invocation = RemoteInvocation(**invocation_values)
                 status_code, output = invocation.invoke()
                 span.set_attribute("http.response.status_code", status_code)
+                span.set_attributes(_telemetry_text_metadata("http.response.body", output))
+                span.add_event("remote.response.received", {"http.response.status_code": status_code})
+                if not 200 <= status_code < 300:
+                    span.set_status(Status(StatusCode.ERROR))
                 run = self._load(run_id)
                 run.step_results.append(
                     {"step_id": "execute", "phase": "execute", "http_status": status_code, "output": output}
@@ -5391,4 +5471,6 @@ def _execute_remote(self, run_id: str, executor: dict[str, Any]) -> None:
                     self._complete_supervision(run_id)
             except (ValueError, requests.RequestException) as error:
                 span.record_exception(error)
+                span.set_status(Status(StatusCode.ERROR))
+                span.add_event("remote.request.failed", {"error.type": type(error).__name__})
                 self._fail(self._load(run_id), "execute", str(error))
diff --git a/frontend/src/features/evaluations/page.tsx b/frontend/src/features/evaluations/page.tsx
index 2108606..0ebe158 100644
--- a/frontend/src/features/evaluations/page.tsx
+++ b/frontend/src/features/evaluations/page.tsx
@@ -1103,7 +1103,6 @@ export function EvaluationsPage({
                 record={translateSupervisorRecord(supervision)}
                 l={l}
                 telemetry={telemetry}
-                iteration={iterationTab}
                 renderLineOutput={(value) => <LineNumberedOutput value={value} />}
               />
             </RunDetailTabPanel>
diff --git a/frontend/src/features/evaluations/run-detail-supervisor-panel.tsx b/frontend/src/features/evaluations/run-detail-supervisor-panel.tsx
index 439fce2..6ec4864 100644
--- a/frontend/src/features/evaluations/run-detail-supervisor-panel.tsx
+++ b/frontend/src/features/evaluations/run-detail-supervisor-panel.tsx
@@ -4,6 +4,15 @@ import { StatusBadge } from "../../components/ui/status-badge";
 
 type Messages = Record<string, string | undefined>;
 
+function telemetryValue(value: unknown) {
+  return typeof value === "string" ? value : JSON.stringify(value);
+}
+
+function duration(span: TelemetrySpan) {
+  if (!span.startTime || !span.endTime || span.endTime < span.startTime) return undefined;
+  return `${((span.endTime - span.startTime) / 1_000_000).toFixed(1)} ms`;
+}
+
 function TelemetryTree({ telemetry, l }: { telemetry?: RunTelemetry; l: Messages }) {
   const [collapsed, setCollapsed] = useState<Set<string>>(new Set());
   const tree = useMemo(() => {
@@ -14,14 +23,14 @@ function TelemetryTree({ telemetry, l }: { telemetry?: RunTelemetry; l: Messages
   if (!telemetry) return <p className="hint">{l.loadingOpenTelemetryTrace}</p>;
   if (!tree.roots.length) return <p className="hint">{l.noOpenTelemetrySpans}</p>;
   const render = (span: TelemetrySpan): ReactNode => {
-    const children = tree.children.get(span.spanId) ?? [], expandable = children.length > 0, isCollapsed = collapsed.has(span.spanId);
-    return <li key={span.spanId}><button type="button" className="trace-node" disabled={!expandable} onClick={() => { if (!expandable) return; setCollapsed((current) => { const next = new Set(current); if (isCollapsed) next.delete(span.spanId); else next.add(span.spanId); return next; }); }}><span className={`trace-status trace-status--${span.status === "ERROR" ? "error" : "ok"}`} /><div><strong>{expandable ? `${isCollapsed ? "▸" : "▾"} ${span.name}` : span.name}</strong><small>{span.events?.map((event) => event.name).join(" · ") || span.status || "UNSET"}</small></div></button>{expandable && !isCollapsed && <ul>{children.map(render)}</ul>}</li>;
+    const children = tree.children.get(span.spanId) ?? [], expandable = children.length > 0, isCollapsed = collapsed.has(span.spanId), attributes = Object.entries(span.attributes ?? {}), events = span.events ?? [];
+    const summary = [span.status || "UNSET", duration(span), ...events.map((event) => event.name)].filter(Boolean).join(" · ");
+    return <li key={span.spanId}><button type="button" className="trace-node" disabled={!expandable} onClick={() => { if (!expandable) return; setCollapsed((current) => { const next = new Set(current); if (isCollapsed) next.delete(span.spanId); else next.add(span.spanId); return next; }); }}><span className={`trace-status trace-status--${span.status === "ERROR" ? "error" : "ok"}`} /><div><strong>{expandable ? `${isCollapsed ? "▸" : "▾"} ${span.name}` : span.name}</strong><small>{summary}</small></div></button>{!isCollapsed && (attributes.length > 0 || events.some((event) => Object.keys(event.attributes ?? {}).length > 0)) && <dl className="trace-details">{attributes.map(([key, value]) => <div key={key}><dt>{key}</dt><dd>{telemetryValue(value)}</dd></div>)}{events.flatMap((event) => Object.entries(event.attributes ?? {}).map(([key, value]) => <div key={`${event.name}.${key}`}><dt>{event.name} · {key}</dt><dd>{telemetryValue(value)}</dd></div>))}</dl>}{expandable && !isCollapsed && <ul>{children.map(render)}</ul>}</li>;
   };
   return <ul className="telemetry-tree">{tree.roots.map(render)}</ul>;
 }
 
-export function SupervisorPanel({ record, l, telemetry, iteration, renderLineOutput }: { record?: SupervisorRecord; l: Messages; telemetry?: RunTelemetry; iteration: number; renderLineOutput: (value: string) => ReactNode }) {
+export function SupervisorPanel({ record, l, telemetry, renderLineOutput }: { record?: SupervisorRecord; l: Messages; telemetry?: RunTelemetry; renderLineOutput: (value: string) => ReactNode }) {
   const response = record?.response;
-  const iterationTelemetry = telemetry ? { ...telemetry, spans: telemetry.spans.filter((span) => span.name === "supervisor.evaluate" && Number(span.attributes?.["orbit.iteration"]) === iteration) } : undefined;
-  return <div className="supervisor-output"><section><div className="supervisor-output__head"><strong>{l.supervisorPrompt}</strong><StatusBadge value={record?.status ?? "pending"} label={record?.status ?? "pending"} /></div>{renderLineOutput(record?.prompt || l.noSupervisorPrompt || "")}</section><section><strong>{l.supervisorResponse}</strong>{response ? renderLineOutput(JSON.stringify(response, null, 2)) : <p>{record?.error || l.supervisorWaiting}</p>}</section><section><strong>{l.openTelemetryTrace}</strong><TelemetryTree telemetry={iterationTelemetry} l={l} /></section></div>;
+  return <div className="supervisor-output"><section><div className="supervisor-output__head"><strong>{l.supervisorPrompt}</strong><StatusBadge value={record?.status ?? "pending"} label={record?.status ?? "pending"} /></div>{renderLineOutput(record?.prompt || l.noSupervisorPrompt || "")}</section><section><strong>{l.supervisorResponse}</strong>{response ? renderLineOutput(JSON.stringify(response, null, 2)) : <p>{record?.error || l.supervisorWaiting}</p>}</section><section><strong>{l.openTelemetryTrace}</strong><TelemetryTree telemetry={telemetry} l={l} /></section></div>;
 }
diff --git a/frontend/src/theme-overrides.css b/frontend/src/theme-overrides.css
index 27c7922..5b4ee04 100644
--- a/frontend/src/theme-overrides.css
+++ b/frontend/src/theme-overrides.css
@@ -119,6 +119,7 @@ footer { display:flex; align-items:center; justify-content:space-between; margin
 .catalog-list__header { position:sticky; top:0; z-index:1; display:grid; min-height:28px; grid-template-columns:minmax(0,1fr) 140px 56px; align-items:center; gap:8px; padding:0 6px; border-bottom:1px solid var(--line); background:var(--surface); }.catalog-list__header button,.catalog-list__header > span { display:inline-flex; align-items:center; gap:4px; padding:0; border:0; color:var(--muted); background:transparent; font:10px 'DM Mono',monospace; text-transform:uppercase; text-align:left; }.catalog-list__header button { cursor:pointer; }.catalog-list__header button:hover { color:var(--accent); }
 .catalog-row-wrap { display:grid; grid-template-columns:minmax(0,1fr) 140px 56px; align-items:center; gap:8px; padding-inline:6px; border-top:1px solid var(--line); }.catalog-row-wrap:hover,.catalog-row-wrap:has(.catalog-row:focus-visible) { background:var(--surface-raised); }.catalog-row-wrap:hover .catalog-row,.catalog-row-wrap:has(.catalog-row:focus-visible) .catalog-row { color:var(--accent); background:transparent; }.catalog-row-wrap:hover .catalog-row__created,.catalog-row-wrap:has(.catalog-row:focus-visible) .catalog-row__created { color:var(--accent); }.catalog-row-wrap .catalog-row { min-width:0; border-top:0; }.catalog-row-wrap .icon-button,.catalog-row__created { min-width:0; }.catalog-row__created { white-space:nowrap; }
 .telemetry-tree,.telemetry-tree ul { display:grid; gap:6px; margin:0; padding:0; list-style:none; }.telemetry-tree ul { margin-left:13px; padding-left:15px; border-left:1px solid var(--line); }.trace-node { display:flex; width:100%; align-items:flex-start; gap:8px; padding:8px 10px; border:1px solid var(--line); border-radius:7px; color:var(--text); background:var(--bg); text-align:left; }.trace-node:not(:disabled) { cursor:pointer; }.trace-node:not(:disabled):hover { border-color:var(--accent); background:var(--surface-raised); }.trace-node:disabled { cursor:default; }.trace-node > div { display:grid; gap:3px; min-width:0; }.trace-node strong { font:600 11px 'DM Mono',monospace; }.trace-node small { overflow:hidden; color:var(--muted); text-overflow:ellipsis; white-space:nowrap; font-size:10px; }.trace-status { width:7px; height:7px; flex:none; margin-top:4px; border-radius:50%; }.trace-status--ok { background:#bce989; }.trace-status--error { background:#eaa89f; }
+.trace-details { display:grid; gap:4px; margin:5px 10px 8px 25px; padding:7px 9px; border-left:2px solid var(--line); color:var(--muted); font:10px 'DM Mono',monospace; }.trace-details div { display:grid; grid-template-columns:minmax(120px, 38%) minmax(0, 1fr); gap:8px; }.trace-details dt { overflow:hidden; text-overflow:ellipsis; white-space:nowrap; }.trace-details dd { min-width:0; margin:0; overflow-wrap:anywhere; color:var(--text); }
 .test-case-hint { margin:0; color:var(--muted); font-size:11px; line-height:1.55; }.test-case-editor fieldset label { display:grid; gap:5px; color:var(--muted); font-size:11px; }.test-case-editor fieldset textarea { min-height:84px; }
 .build-wizard .test-case-editor { display:none; }
 .build-wizard .modal-setting-row:has(textarea[placeholder]) { display:none; }
diff --git a/specs/observability/open-telemetry.md b/specs/observability/open-telemetry.md
index c2db550..f115e5f 100644
--- a/specs/observability/open-telemetry.md
+++ b/specs/observability/open-telemetry.md
@@ -5,12 +5,15 @@ Status: accepted
 ## Trace model
 
 - One run creates one root trace.
-- Workflow, step, remote-invocation and model calls are spans.
+- Workflow, step, remote-invocation, supervisor evaluation and cycle-review model calls are spans.
 - Dashboard rows expose the local trace ID and its exported spans.
 
 ## Logs and metrics
 
-- Console output is retained in the local run record; OTEL records subprocess completion and failures as span events.
+- Console output and full supervisor prompts/responses are retained in the local run record. OTEL records
+  redaction-safe request/response lengths and SHA-256 fingerprints, model/provider metadata, evaluation
+  outcomes, subprocess lifecycle and output counts, and remote-response metadata as span attributes/events.
+- Prompt, response, console and HTTP-body text are not exported in OTEL records.
 - Dashboard metrics currently include build/run counts and supervisor-feedback summaries.
 
 ## Data protection