From 27b3a208d2fa345afe86dc1773018e202c11da92 Mon Sep 17 00:00:00 2001
From: Codex <codex@noreply.local>
Date: Sat, 20 Jun 2026 22:42:47 +0000
Subject: [PATCH] fix(web-probe): separate baseline execution errors

---
 .../hwlab-node-web-observe-runner-source.ts   | 25 +++++++++++++++----
 1 file changed, 20 insertions(+), 5 deletions(-)

diff --git a/scripts/src/hwlab-node-web-observe-runner-source.ts b/scripts/src/hwlab-node-web-observe-runner-source.ts
index 40db6596..d86fcf92 100644
--- a/scripts/src/hwlab-node-web-observe-runner-source.ts
+++ b/scripts/src/hwlab-node-web-observe-runner-source.ts
@@ -1016,6 +1016,9 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
     .map((item) => networkAlertEvent(item, promptTimes));
   const domDiagnostics = [];
   const executionErrors = [];
+  const baselineExecutionErrors = [];
+  const firstPromptMs = promptTimes.length > 0 ? promptTimes[0] : Infinity;
+  const firstSeenExecutionErrorMs = new Map();
   for (const sample of samples) {
     const tsMs = Date.parse(sample?.ts);
     const promptIndex = Number.isFinite(tsMs) ? latestPromptIndex(promptTimes, tsMs) : 0;
@@ -1059,13 +1062,19 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
     for (const candidate of sampleExecutionErrorCandidates(sample)) {
       const parsed = parseExecutionErrorText(candidate.text);
       if (!parsed) continue;
-      const dedupeKey = [candidate.source, candidate.traceId || "-", parsed.backend || "-", parsed.code || "-", parsed.status || "-", sha256(candidate.text)].join("|");
+      const textHash = sha256(candidate.text);
+      const dedupeKey = [candidate.source, candidate.traceId || "-", parsed.backend || "-", parsed.code || "-", parsed.status || "-", textHash].join("|");
       if (seenExecutionErrors.has(dedupeKey)) continue;
       seenExecutionErrors.add(dedupeKey);
-      executionErrors.push({
+      const firstSeenMs = firstSeenExecutionErrorMs.has(dedupeKey) ? firstSeenExecutionErrorMs.get(dedupeKey) : tsMs;
+      if (!firstSeenExecutionErrorMs.has(dedupeKey) && Number.isFinite(tsMs)) firstSeenExecutionErrorMs.set(dedupeKey, tsMs);
+      const baseline = Number.isFinite(firstSeenMs) && firstSeenMs < firstPromptMs;
+      const event = {
         seq: sample.seq ?? null,
         ts: sample.ts ?? null,
         promptIndex,
+        baseline,
+        firstSeenAt: Number.isFinite(firstSeenMs) ? new Date(firstSeenMs).toISOString() : null,
         source: candidate.source,
         backend: parsed.backend,
         status: parsed.status,
@@ -1076,9 +1085,11 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
         messageId: candidate.messageId || null,
         routeSessionId: sample.routeSessionId ?? null,
         activeSessionId: sample.activeSessionId ?? null,
-        textHash: sha256(candidate.text),
+        textHash,
         preview: limitText(candidate.text, 260)
-      });
+      };
+      if (baseline) baselineExecutionErrors.push(event);
+      else executionErrors.push(event);
       domDiagnostics.push({
         seq: sample.seq ?? null,
         ts: sample.ts ?? null,
@@ -1088,7 +1099,7 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
         traceId: candidate.traceId || parsed.traceId || null,
         routeSessionId: sample.routeSessionId ?? null,
         activeSessionId: sample.activeSessionId ?? null,
-        textHash: sha256(candidate.text),
+        textHash,
         preview: limitText(candidate.text, 220)
       });
     }
@@ -1110,11 +1121,13 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
       requestFailedCount: requestFailed.length,
       domDiagnosticSampleCount: domDiagnostics.length,
       executionErrorCount: executionErrors.length,
+      baselineExecutionErrorCount: baselineExecutionErrors.length,
       consoleAlertCount: consoleAlerts.length,
       pageErrorCount: pageErrors.length,
       networkErrorGroupCount: groupNetworkAlerts(httpErrors).length,
       requestFailedGroupCount: groupNetworkAlerts(requestFailed).length,
       executionErrorGroupCount: groupExecutionErrors(executionErrors).length,
+      baselineExecutionErrorGroupCount: groupExecutionErrors(baselineExecutionErrors).length,
       consoleAlertGroupCount: groupConsoleAlerts(consoleAlerts).length
     },
     networkHttpErrorsByPath: groupNetworkAlerts(httpErrors),
@@ -1122,6 +1135,8 @@ function buildRuntimeAlerts(samples, control, network, consoleEvents, errors) {
     domDiagnostics: domDiagnostics.slice(0, 80),
     runtimeExecutionErrors: executionErrors.slice(0, 120),
     runtimeExecutionErrorsByCode: groupExecutionErrors(executionErrors),
+    baselineRuntimeExecutionErrors: baselineExecutionErrors.slice(0, 80),
+    baselineRuntimeExecutionErrorsByCode: groupExecutionErrors(baselineExecutionErrors),
     consoleAlerts: consoleAlerts.slice(0, 80),
     consoleAlertsByPath: groupConsoleAlerts(consoleAlerts),
     pageErrors: pageErrors.slice(0, 40)