fix: retry on transient API network errors instead of failing immediately

Add check_network_error() helper and wire it into the retry loop so that
400 Network error responses from the Claude API are retried (up to
MAX_RETRIES, 60s backoff) rather than treated as non-retryable failures.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: Test User

scripts/night-watch-cron.sh

@@ -565,6 +565,15 @@ while [ "${ATTEMPT}" -lt "${MAX_RETRIES}" ]; do
     BACKOFF_MIN=$(( BACKOFF / 60 ))
     log "RATE-LIMITED: Attempt ${ATTEMPT}/${MAX_RETRIES}, retrying in ${BACKOFF_MIN}m"
     sleep "${BACKOFF}"
+  elif check_network_error "${LOG_FILE}" "${LOG_LINE_BEFORE}"; then
+    # Transient API network error — retry with a short backoff
+    ATTEMPT=$((ATTEMPT + 1))
+    if [ "${ATTEMPT}" -ge "${MAX_RETRIES}" ]; then
+      log "NETWORK-ERROR: All ${MAX_RETRIES} attempts exhausted for ${ELIGIBLE_PRD}"
+      break
+    fi
+    log "NETWORK-ERROR: Attempt ${ATTEMPT}/${MAX_RETRIES}, retrying in 60s"
+    sleep 60
   elif check_context_exhausted "${LOG_FILE}" "${LOG_LINE_BEFORE}"; then
     # Context window exhausted — checkpoint progress and resume in a fresh session
     ATTEMPT=$((ATTEMPT + 1))

scripts/night-watch-helpers.sh

@@ -847,6 +847,19 @@ check_rate_limited() {
   fi
 }
 
+# Detect transient API network errors (e.g. "Network error" in a 400 response).
+# Usage: check_network_error <log_file> [start_line]
+# Returns 0 if a network error was detected, 1 otherwise.
+check_network_error() {
+  local log_file="${1:?log_file required}"
+  local start_line="${2:-0}"
+  if [ "${start_line}" -gt 0 ] 2>/dev/null; then
+    tail -n "+$((start_line + 1))" "${log_file}" 2>/dev/null | grep -qi "Network error"
+  else
+    tail -20 "${log_file}" 2>/dev/null | grep -qi "Network error"
+  fi
+}
+
 # Detect context window exhaustion from Claude API logs.
 # Usage: check_context_exhausted <log_file> [start_line]
 # Returns 0 if context exhausted, 1 otherwise.