Here's how it was handled as a hack, but I think a much safer idea would be to incorporate this directly into the results recipe.
# WARNING! Position of this dependency is important! It needs to occur *after*
# the dependency on the "test" rule.
results: verify-results
.PHONY: verify-results
verify-results:
@failed=0; \
if grep -Hn '^not ok' test/results/*.out 2>/dev/null; then \
failed=1; \
fi; \
if grep -Hn 'Looks like you planned' test/results/*.out 2>/dev/null; then \
failed=1; \
fi; \
if [ $$failed -eq 1 ]; then \
echo "ERROR: Failed tests found in results files (see above)"; \
exit 1; \
fi
@echo "test/results/ files are clean (no 'not ok' lines or plan count mismatches found)"