CI: t15 diagnostic printed the report, not the error

The block dumped the last 25 lines of srcmode.log, which turned out to be the
evidence report itself -- correctly saying ESCROW_PIPELINE=FAIL and
DISCOVERY=NOT_RUN, and saying nothing about why, because nse_die writes its
message long before the exit trap writes the report. Diagnostics that only
reach the tail of a log are diagnostics about log formatting.

It now prints the matching error lines, the stage progress, and a shorter
tail. Incidentally the report-on-failure machinery (u08) is confirmed working
in the field: a pipeline that died in discover still produced a full report
with every later stage marked NOT_RUN.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Eydip5xfXMqeno9k1c6DEa

Author: claude

tests/run-tests.sh

@@ -438,8 +438,16 @@ else
   if [ "$rc" -ne 0 ]; then
     # A failing test that points at a log file is useless to anyone reading CI
     # output. Print the reason here.
-    echo "    --- why t15 failed: last 25 lines of srcmode.log ---"
-    tail -25 "$WORK/srcmode.log" 2>/dev/null | sed 's/^/    | /'
+    # The tail was the evidence report, which correctly said the pipeline
+    # failed -- and said nothing about why, because nse_die writes its message
+    # far earlier. Print the error itself, and the stage progress around it.
+    echo "    --- why t15 failed: the error ---"
+    grep -n -m3 -E 'nix-source-escrow: (error|warning):|error:' "$WORK/srcmode.log" 2>/dev/null \
+      | sed 's/^/    ! /' || echo "    ! (no error line found)"
+    echo "    --- stage progress (==> lines) ---"
+    grep -n '^==>' "$WORK/srcmode.log" 2>/dev/null | tail -12 | sed 's/^/    | /'
+    echo "    --- last 12 lines ---"
+    tail -12 "$WORK/srcmode.log" 2>/dev/null | sed 's/^/    | /'
     if [ -f "$SRC_OI" ]; then
       echo "    --- its verdict ---"
       jq -r '"    | result=\(.result)  step=\(.failedStep)  reason=\(.reason // "none")"' "$SRC_OI"