diff --git a/smoke.sh b/smoke.sh
index e9d961c..498b4c2 100755
--- a/smoke.sh
+++ b/smoke.sh
@@ -60,6 +60,12 @@ echo
 [[ -x "${CHROME_BIN}" ]] || { red "FAIL: ${CHROME_BIN} not found (set CHROME_BIN=...)"; exit 1; }
 command -v ffprobe >/dev/null || { red "FAIL: ffprobe not installed"; exit 1; }
 command -v unzip >/dev/null || { red "FAIL: unzip not installed"; exit 1; }
+# WR-04 fix: python3 is REQUIRED for URL encoding of the smoke-tab data: URL.
+# Previously a fallback `|| printf '%s' "${SMOKE_HTML}"` would emit raw HTML
+# into the data URL on python3-missing systems — Chrome silently failed to
+# parse those URLs (containing literal `<`, `>`, spaces, quotes) and the
+# operator saw a blank tab with no diagnostic. Better to fail loud early.
+command -v python3 >/dev/null || { red "FAIL: python3 not installed (needed for URL encoding the smoke tab data URL)"; exit 1; }
 grep -q '"desktopCapture"' "${DIST_DIR}/manifest.json" || { red "FAIL: dist/manifest.json missing desktopCapture"; exit 1; }
 ! grep -q '"tabCapture"' "${DIST_DIR}/manifest.json" || { red "FAIL: dist/manifest.json still has tabCapture"; exit 1; }
 green "✓ pre-flight checks passed"
@@ -71,16 +77,24 @@ echo "  downloads: ${DOWNLOADS_DIR}"
 echo
 
 # --- snapshot Downloads ---
-BEFORE_LIST=$(find "${DOWNLOADS_DIR}" -maxdepth 1 -name 'session_report_*.zip' 2>/dev/null | wc -l)
-echo "  existing session_report_*.zip in Downloads: ${BEFORE_LIST}"
+# WR-05 fix: snapshot the FULL list of pre-existing zips, not just the
+# count. The count-only approach falsely succeeded when an unrelated
+# session_report appeared in Downloads (operator running the extension
+# in another window, etc.) — the script would then ffprobe the WRONG
+# file. comm -13 against the post-recording list yields the genuinely
+# new file by identity. The mtime sort below still picks the latest if
+# multiple new zips appear (unlikely but defensive).
+BEFORE_ZIPS=$(find "${DOWNLOADS_DIR}" -maxdepth 1 -name 'session_report_*.zip' 2>/dev/null | sort)
+BEFORE_COUNT=$(printf '%s\n' "${BEFORE_ZIPS}" | grep -c . || true)
+echo "  existing session_report_*.zip in Downloads: ${BEFORE_COUNT}"
 echo
 
 # --- profile prep ---
 PROFILE_HAS_EXTENSION=0
 if [[ "${KEEP_PROFILE}" != "1" ]]; then
-  rm -rf "${PROFILE_DIR}"
+  rm -rf -- "${PROFILE_DIR}"
 fi
-mkdir -p "${PROFILE_DIR}"
+mkdir -p -- "${PROFILE_DIR}"
 # Detect if a previous run already loaded the extension into this profile
 if [[ -d "${PROFILE_DIR}/Default/Extensions" ]] && \
    find "${PROFILE_DIR}/Default/Extensions" -maxdepth 3 -name 'manifest.json' 2>/dev/null | head -1 | xargs -r grep -q 'AI Call Recorder' 2>/dev/null; then
@@ -106,7 +120,11 @@ read -r -d '' SMOKE_HTML <<'EOF' || true
 <p>The script in your terminal will detect the download and finish the ffprobe gate automatically.</p>
 </body>
 EOF
-SMOKE_DATA_URL="data:text/html,$(printf '%s' "${SMOKE_HTML}" | python3 -c 'import sys,urllib.parse;print(urllib.parse.quote(sys.stdin.read(), safe=""))' 2>/dev/null || printf '%s' "${SMOKE_HTML}")"
+# WR-04 fix: python3 is required (asserted in pre-flight). NO fallback —
+# the previous `|| printf '%s' "${SMOKE_HTML}"` would emit unencoded HTML
+# into the data URL, causing Chrome to fail to parse the URL silently and
+# the operator to see a blank tab. Fail loudly if URL encoding fails.
+SMOKE_DATA_URL="data:text/html,$(printf '%s' "${SMOKE_HTML}" | python3 -c 'import sys,urllib.parse;print(urllib.parse.quote(sys.stdin.read(), safe=""))')"
 
 # --- launch Chrome ---
 blue "==> launching Chrome with smoke profile + auto-accept picker..."
@@ -152,22 +170,52 @@ echo "    (Ctrl+C aborts. Auto-detects, ffprobes, stages fixture, opens WebM.)"
 echo
 
 # --- poll Downloads ---
+# WR-05 fix: detect by IDENTITY via `comm -13 <before> <after>`, not by
+# count comparison. The count approach false-positives when ANY
+# session_report appears (e.g. the operator's daily extension in another
+# window). `comm -13` returns lines present in <after> but not in
+# <before> — the genuine new file(s). We still apply mtime sort + head -1
+# to pick the latest if multiple new zips materialize (e.g., overlapping
+# operator activity), but the candidate set is now restricted to actually-new
+# files.
 NEW_ARCHIVE=""
 WAITED=0
 while [[ ${WAITED} -lt ${POLL_TIMEOUT} ]]; do
-  NEW_COUNT=$(find "${DOWNLOADS_DIR}" -maxdepth 1 -name 'session_report_*.zip' 2>/dev/null | wc -l)
-  if [[ ${NEW_COUNT} -gt ${BEFORE_LIST} ]]; then
+  AFTER_ZIPS=$(find "${DOWNLOADS_DIR}" -maxdepth 1 -name 'session_report_*.zip' 2>/dev/null | sort)
+  # comm requires sorted streams; both inputs above are pre-sorted.
+  # `-13` keeps only lines unique to file2 (AFTER), suppressing common
+  # lines and lines unique to BEFORE.
+  NEW_ZIPS=$(comm -13 <(printf '%s\n' "${BEFORE_ZIPS}") <(printf '%s\n' "${AFTER_ZIPS}") | grep -v '^$' || true)
+  if [[ -n "${NEW_ZIPS}" ]]; then
     sleep 2  # let the download settle
-    LATEST=$(find "${DOWNLOADS_DIR}" -maxdepth 1 -name 'session_report_*.zip' -printf '%T@ %p\n' 2>/dev/null | sort -rn | head -1 | cut -d' ' -f2- || true)
-    if [[ -n "${LATEST}" ]]; then
-      NEW_ARCHIVE="${LATEST}"
-      break
+    # Re-snapshot after settle, recompute identity diff, pick latest by mtime
+    AFTER_ZIPS=$(find "${DOWNLOADS_DIR}" -maxdepth 1 -name 'session_report_*.zip' 2>/dev/null | sort)
+    NEW_ZIPS=$(comm -13 <(printf '%s\n' "${BEFORE_ZIPS}") <(printf '%s\n' "${AFTER_ZIPS}") | grep -v '^$' || true)
+    if [[ -n "${NEW_ZIPS}" ]]; then
+      # Pick the latest among the genuinely-new zips by mtime. Quoting note:
+      # NEW_ZIPS is a newline-separated list of full paths from `find`; we
+      # iterate via `while read` to preserve paths with embedded spaces.
+      LATEST=""
+      LATEST_MTIME=0
+      while IFS= read -r zip_path; do
+        [[ -z "${zip_path}" ]] && continue
+        mtime=$(stat -c %Y -- "${zip_path}" 2>/dev/null || echo 0)
+        if [[ "${mtime}" -gt "${LATEST_MTIME}" ]]; then
+          LATEST="${zip_path}"
+          LATEST_MTIME="${mtime}"
+        fi
+      done <<<"${NEW_ZIPS}"
+      if [[ -n "${LATEST}" ]]; then
+        NEW_ARCHIVE="${LATEST}"
+        break
+      fi
     fi
   fi
   sleep 1
   WAITED=$((WAITED + 1))
   if [[ $((WAITED % 30)) -eq 0 ]]; then
-    yellow "   ...still waiting (${WAITED}s elapsed, count=${NEW_COUNT}/baseline=${BEFORE_LIST})"
+    after_count=$(printf '%s\n' "${AFTER_ZIPS}" | grep -c . || true)
+    yellow "   ...still waiting (${WAITED}s elapsed, count=${after_count}/baseline=${BEFORE_COUNT})"
   fi
 done
 
@@ -183,9 +231,12 @@ green "✓ archive detected: ${NEW_ARCHIVE}"
 echo
 
 # --- extract + ffprobe gate ---
-unzip -p "${NEW_ARCHIVE}" video/last_30sec.webm > "${WEBM_TMP}"
-SIZE_BYTES=$(stat -c %s "${WEBM_TMP}")
-SIZE_HUMAN=$(ls -lh "${WEBM_TMP}" | awk '{print $5}')
+# Bash-style sweep: pass `--` to terminate options on file-taking commands
+# that accept user-controlled paths (Google shell style guide §"Special
+# considerations" — defensive against filenames starting with `-`).
+unzip -p -- "${NEW_ARCHIVE}" video/last_30sec.webm > "${WEBM_TMP}"
+SIZE_BYTES=$(stat -c %s -- "${WEBM_TMP}")
+SIZE_HUMAN=$(ls -lh -- "${WEBM_TMP}" | awk '{print $5}')
 echo "  WebM size: ${SIZE_HUMAN} (${SIZE_BYTES} bytes)"
 if [[ ${SIZE_BYTES} -lt 100000 ]]; then
   yellow "⚠ WebM is smaller than 100 KB — buffer may not have rotated; capture longer"
@@ -194,6 +245,12 @@ fi
 echo
 blue "==> D-12 ACCEPTANCE GATE — ffprobe -v error"
 echo "---"
+# The `&& GATE=0 || GATE=$?` chain is correct: under `set -e`, ffprobe's
+# non-zero exit doesn't terminate the script because it's followed by `||`.
+# When ffprobe succeeds, `GATE=0` (an assignment returning 0) is executed
+# and the `||` branch is skipped. When ffprobe fails, the `&&` chain is
+# bypassed and `GATE=$?` captures ffprobe's exit. The earlier review note
+# WR-04 confirmed this is NOT broken.
 ffprobe -v error -f matroska -i "${WEBM_TMP}" && GATE=0 || GATE=$?
 echo "---"
 echo "ffprobe exit: ${GATE}"
@@ -215,9 +272,9 @@ echo
 
 # --- stage fixture ---
 if [[ ${GATE} -eq 0 ]]; then
-  mkdir -p "$(dirname "${FIXTURE_DEST}")"
-  cp "${WEBM_TMP}" "${FIXTURE_DEST}"
-  green "✓ fixture staged: ${FIXTURE_DEST} ($(ls -lh "${FIXTURE_DEST}" | awk '{print $5}'))"
+  mkdir -p -- "$(dirname -- "${FIXTURE_DEST}")"
+  cp -- "${WEBM_TMP}" "${FIXTURE_DEST}"
+  green "✓ fixture staged: ${FIXTURE_DEST} ($(ls -lh -- "${FIXTURE_DEST}" | awk '{print $5}'))"
 fi
 
 # --- open the WebM for visual check ---