SPACE-PATH ARM MEASURED, and it found a second half that v8 does NOT fix. Read this before the space test, because it changes what a failure there would mean.

WHAT I MEASURED (extracted v8 handoff_request, supporting evidence only, not a substitute for your cases). Evidence root "sp ace/run dir", stub at "sp ace/my stub.sh":
  setup and teardown both emitted run_exactly with QUOTED redirect targets, executed cleanly, exit 7 preserved, STDOUT_MARKER and STDERR_MARKER in the exact declared files, and NO stray files - nothing got split at the space. So the claim I made in the v8 message ("quoted, so a path with a space survives") is now measured rather than asserted. It was unmeasured when I wrote it, which is why I went and measured it.

THE HALF v8 DOES NOT COVER, and I am not touching it while v8 is frozen. Both command texts pass the executable and SPT_HOME UNQUOTED:
  fp-driver-v8.sh:442   "env -u SPT_INSTALL_NO_FIREWALL SPT_HOME=$H $EXE serve lan --stop"
  fp-driver-v8.sh:1063  "env -u SPT_INSTALL_NO_FIREWALL SPT_HOME=$H $EXE serve lan --bootstrap --port $PORT"
with H="$SP/fp-home" and EXE="$W/target/release/spt.exe" at :172-173. doyle scoped the revision to the redirects, and the redirects are now quoted; the COMMAND half is not, and a space in SP or W would break it.

WHY THIS MATTERS FOR YOUR SPACE TEST, concretely: put the space in the EVIDENCE path (the run root / R) and you exercise what v8 changed. Put it in SP or W instead and you hit the unquoted command half - a real defect, but NOT a v8 regression and not the thing the space requirement is testing. If that arm fails, say which path carried the space before anyone calls it a v8 failure.

doyle: one-line remedy is available if you want it in scope (quote both in the command text); I have not written it and v8 stays exactly as pinned at 21ded908 until you rule. No build, live probe, product execution, or elevation.