diff --git a/test/standard/make-test.bats b/test/standard/make-test.bats
index 9b29821..13a1470 100755
--- a/test/standard/make-test.bats
+++ b/test/standard/make-test.bats
@@ -13,8 +13,9 @@
 #   (issue #79)
 # - check-stale-expected catches orphaned test/expected/*.out files (issue #14)
 # - `make test` exits non-zero on a real regression.diffs mismatch (issue #49)
-# - verify-results blocks `make results` when tests are failing, detects
-#   pgtap failures, and can be disabled
+# - verify-results blocks `make results` when tests are failing, lets a
+#   brand-new test's first expected output through (issue #119), and can be
+#   disabled. Its own pass/fail logic lives in verify-results-pgtap-script.bats
 
 load ../lib/helpers
 
@@ -529,54 +530,54 @@ EOF
   assert_success
 }
 
-@test "verify-results detects pgtap failures in result files" {
+@test "make results seeds the first expected output for a brand-new test (issue #119)" {
   skip_if_no_postgres
 
-  mkdir -p test/results
-  cat > test/results/pgtap_fail.out <<'EOF'
-1..2
-ok 1 - passing test
-not ok 2 - failing test
-EOF
-
-  run make verify-results
-  assert_failure
-  assert_contains "$output" "pgtap failure detected"
+  # base.mk touches an empty test/expected/<name>.out for any test/sql/*.sql
+  # that lacks one (pg_regress aborts on a missing expected file), so
+  # pg_regress reports a difference no matter how clean the new test's actual
+  # output is. This is the one case that needs a real pg_regress run: it
+  # proves the diff pg_regress writes against that placeholder really is the
+  # shape verify-results-pgtap.sh treats as "no baseline yet" rather than as
+  # a failure. The script's own classification is covered cheaply in
+  # verify-results-pgtap-script.bats.
+  cp test/sql/pgxntool-test.sql test/sql/brand_new.sql
 
-  rm -f test/results/pgtap_fail.out
-}
-
-@test "verify-results ignores pgtap TODO failures" {
-  skip_if_no_postgres
-
-  mkdir -p test/results
-  cat > test/results/pgtap_todo.out <<'EOF'
-1..1
-not ok 1 - known issue # TODO fix later
-EOF
+  run make results
+  assert_success
+  # The NOTE proves the no-baseline branch is what let this through, rather
+  # than the file appearing for some other reason.
+  assert_contains "$output" "no expected output yet"
+  assert_contains "$output" "brand_new.out"
+  assert_file_exists "test/expected/brand_new.out"
 
-  run make verify-results
+  # The seeded baseline is what the test actually produces.
+  run make test
   assert_success
 
-  rm -f test/results/pgtap_todo.out
+  # results/ too: `make results` blesses every results/*.out into expected/,
+  # so a leftover here would reappear as an orphan for check-stale-expected.
+  rm -f test/sql/brand_new.sql test/expected/brand_new.out test/results/brand_new.out
 }
 
-@test "verify-results detects pgtap plan mismatch" {
+@test "verify-results propagates a pgtap failure the script detects" {
   skip_if_no_postgres
 
+  # Wiring only -- that `make verify-results` really invokes the script and
+  # surfaces its failure. Which pgtap output counts as a failure (TODO
+  # exclusion, plan mismatches) is decided by the script and tested there.
   mkdir -p test/results
-  cat > test/results/pgtap_plan.out <<'EOF'
-1..3
-ok 1 - test one
-ok 2 - test two
-# Looks like you planned 3 tests but ran 2
+  cat > test/results/pgtap_fail.out <<'EOF'
+1..2
+ok 1 - passing test
+not ok 2 - failing test
 EOF
 
   run make verify-results
   assert_failure
-  assert_contains "$output" "pgtap plan mismatch"
+  assert_contains "$output" "pgtap failure detected"
 
-  rm -f test/results/pgtap_plan.out
+  rm -f test/results/pgtap_fail.out
 }
 
 # ============================================================================
diff --git a/test/standard/verify-results-pgtap-script.bats b/test/standard/verify-results-pgtap-script.bats
new file mode 100644
index 0000000..dd60920
--- /dev/null
+++ b/test/standard/verify-results-pgtap-script.bats
@@ -0,0 +1,319 @@
+#!/usr/bin/env bats
+
+# Test: verify-results-pgtap.sh - pure script-logic unit tests
+#
+# These tests exercise pgxntool's verify-results-pgtap.sh directly against a
+# bare scratch directory -- no foundation environment, no `make`, no
+# PostgreSQL. They own all of the script's decision logic: which pgtap output
+# in results/*.out counts as a failure, and which regression.diffs entries
+# block `make results` (a real mismatch against an existing baseline) versus
+# which are a brand-new test that has no expected output yet (issue #119).
+#
+# Tests that need real Make/PostgreSQL integration stay in make-test.bats:
+# that `make verify-results` invokes this script and propagates its exit
+# status, and that a real pg_regress run against base.mk's empty
+# expected-file placeholder actually produces the "@@ -0,0" diff shape the
+# classification below keys on.
+
+load ../lib/helpers
+load ../lib/assertions
+
+setup_file() {
+  setup_topdir
+  load_test_env "verify-results-pgtap-script"
+}
+
+setup() {
+  load_test_env "verify-results-pgtap-script"
+  export SCRIPT="$PGXNREPO/verify-results-pgtap.sh"
+
+  # Fresh, empty scratch directory per test -- no foundation/TEST_REPO needed.
+  export TESTOUT="$BATS_TEST_TMPDIR/testout"
+  mkdir -p "$TESTOUT/results"
+}
+
+# Record $2 as test $1's pg_regress output file.
+write_results() {
+  printf '%s\n' "$2" > "$TESTOUT/results/$1.out"
+}
+
+# Append the regression.diffs block pg_regress would write for test $1: the
+# "diff <opts> <expected> <results>" header, the ---/+++ file lines, then $2
+# as the diff body (hunk headers included).
+append_diff_block() {
+  local name=$1 body=$2
+  {
+    printf 'diff -U3 %s %s\n' "$TESTOUT/expected/$name.out" "$TESTOUT/results/$name.out"
+    printf -- '--- %s\t2026-01-01 00:00:00.000000000 +0000\n' "$TESTOUT/expected/$name.out"
+    printf -- '+++ %s\t2026-01-01 00:00:00.000000000 +0000\n' "$TESTOUT/results/$name.out"
+    printf '%s\n' "$body"
+  } >> "$TESTOUT/regression.diffs"
+}
+
+@test "verify-results-pgtap.sh: passes on all-ok output with no regression.diffs" {
+  write_results passing '1..1
+ok 1'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_success
+}
+
+@test "verify-results-pgtap.sh: allows a brand-new test that has no expected output yet (issue #119)" {
+  # base.mk touches an empty test/expected/<name>.out for a test that lacks
+  # one, so pg_regress always reports a difference against it: a single
+  # "@@ -0,0" hunk. That is unwinnable to block on -- creating that first
+  # expected file is what `make results` is for.
+  write_results newtest '1..1
+ok 1'
+  append_diff_block newtest '@@ -0,0 +1,2 @@
++1..1
++ok 1'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_success
+  assert_contains "$output" "no expected output yet"
+  assert_contains "$output" "newtest.out"
+}
+
+@test "verify-results-pgtap.sh: still blocks a real mismatch against an existing baseline" {
+  write_results oldtest '1..1
+ok 1'
+  append_diff_block oldtest '@@ -1,2 +1,2 @@
+ 1..1
+-ok 1 - renamed
++ok 1'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "Tests are failing"
+  assert_contains "$output" "Cannot run 'make results'"
+  assert_contains "$output" "oldtest.out"
+}
+
+@test "verify-results-pgtap.sh: a real mismatch still blocks alongside a brand-new test" {
+  write_results newtest '1..1
+ok 1'
+  append_diff_block newtest '@@ -0,0 +1,2 @@
++1..1
++ok 1'
+  write_results oldtest '1..1
+ok 1'
+  append_diff_block oldtest '@@ -1,2 +1,2 @@
+ 1..1
+-ok 1 - renamed
++ok 1'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "Tests are failing"
+  assert_contains "$output" "oldtest.out"
+}
+
+@test "verify-results-pgtap.sh: refuses to bless a brand-new test whose output holds a SQL error" {
+  # ON_ERROR_STOP aborts the script at the error, so pgtap emits neither
+  # 'not ok' nor its plan-mismatch line and the scan above finds nothing.
+  #
+  # Both of psql's real renderings have to block. pg_regress feeds psql on
+  # stdin, so an error in the test file itself starts at column 0, while one
+  # inside an \i'd file (setup.sql/finish.sql, which every pgxntool test
+  # sources) carries a "psql:<file>:<line>: " prefix -- a column-anchored
+  # match would sail straight past the second.
+  write_results errtest '1..2
+ok 1
+ERROR:  42P01: relation "nope" does not exist'
+  append_diff_block errtest '@@ -0,0 +1,3 @@
++1..2
++ok 1
++ERROR:  42P01: relation "nope" does not exist'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "errtest.out"
+
+  rm -f "$TESTOUT/regression.diffs"
+  write_results errtest '1..2
+ok 1
+psql:test/pgxntool/setup.sql:3: ERROR:  42P01: relation "nope" does not exist'
+  append_diff_block errtest '@@ -0,0 +1,3 @@
++1..2
++ok 1
++psql:test/pgxntool/setup.sql:3: ERROR:  42P01: relation "nope" does not exist'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "errtest.out"
+}
+
+@test "verify-results-pgtap.sh: blesses a passing test whose description merely mentions ERROR:" {
+  # psql writes two spaces after the severity; a pgtap description quoting an
+  # error message does not, and isn't at the start of a line or after ": ".
+  # Without that distinction a legitimately-passing new test could never have
+  # its first expected output created.
+  write_results described '1..1
+ok 1 - raises ERROR: division by zero'
+  append_diff_block described '@@ -0,0 +1,2 @@
++1..1
++ok 1 - raises ERROR: division by zero'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_success
+  assert_contains "$output" "no expected output yet"
+  assert_contains "$output" "described.out"
+}
+
+@test "verify-results-pgtap.sh: classifies by unprefixed headers, not by diffed content" {
+  # Every content line in a diff carries a ' ', '+' or '-' prefix, so output
+  # that itself contains block or hunk headers must not be mistaken for one.
+  write_results tricky '1..1
+ok 1'
+  append_diff_block tricky '@@ -0,0 +1,4 @@
++diff -U3 a.out b.out
++@@ -1,2 +1,2 @@
++1..1
++ok 1'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_success
+  assert_contains "$output" "no expected output yet"
+}
+
+@test "verify-results-pgtap.sh: a deleted '-- comment' line doesn't read as a file header" {
+  # A removed "-- comment" renders as "--- comment", which is why the
+  # classification keys on "diff "/"@@ " lines rather than the ---/+++ pair.
+  write_results commented '1..1
+ok 1'
+  append_diff_block commented '@@ -1,3 +1,2 @@
+ 1..1
+--- vi: expandtab ts=2 sw=2
++ok 1'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "commented.out"
+}
+
+@test "verify-results-pgtap.sh: blocks when regression.diffs exists but is empty" {
+  # pg_regress truncates the file at startup and removes it again on a clean
+  # finish, so an empty one means it died before comparing anything -- and
+  # base.mk's `.IGNORE: installcheck` lets `make results` reach here anyway.
+  # results/ then holds whatever an earlier run left, so passing this would
+  # bless stale output.
+  write_results stale '1..1
+ok 1'
+  : > "$TESTOUT/regression.diffs"
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "pg_regress did not complete"
+}
+
+@test "verify-results-pgtap.sh: blocks a block whose hunks aren't all '@@ -0,0'" {
+  # Two fail-closed shapes that aren't a clean no-baseline block: a header
+  # with no hunks at all (what a pre-12 context diff looks like here), and a
+  # block that only partly diffed against an empty file.
+  write_results nohunk '1..1
+ok 1'
+  append_diff_block nohunk ''
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "nohunk.out"
+
+  rm -f "$TESTOUT/regression.diffs"
+  write_results mixed '1..1
+ok 1'
+  append_diff_block mixed '@@ -0,0 +1,1 @@
++1..1
+@@ -3,1 +4,1 @@
+-was here
++ok 1'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "mixed.out"
+}
+
+@test "verify-results-pgtap.sh: blocks when unrecognized content follows a valid block" {
+  # The whole file has to fail closed, not just refuse to classify: a block
+  # that reads as blessable is no reason to ignore text after it that fits no
+  # part of the diff format.
+  write_results newtest '1..1
+ok 1'
+  append_diff_block newtest '@@ -0,0 +1,2 @@
++1..1
++ok 1'
+  printf 'diff: /nonexistent: No such file or directory\n' >> "$TESTOUT/regression.diffs"
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "unrecognized content"
+
+  # ...but diff's own no-newline marker is part of the format, not junk.
+  # Treating it as junk would block a legitimate first bless of any test
+  # whose output doesn't end in a newline.
+  rm -f "$TESTOUT/regression.diffs"
+  append_diff_block newtest '@@ -0,0 +1,2 @@
++1..1
++ok 1
+\ No newline at end of file'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_success
+  assert_contains "$output" "no expected output yet"
+}
+
+@test "verify-results-pgtap.sh: blocks on regression.diffs content it cannot classify" {
+  write_results passing '1..1
+ok 1'
+  printf 'truncated junk with no diff header\n' > "$TESTOUT/regression.diffs"
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "unrecognized"
+
+  # Diff body lines with no header before them yield no classification at
+  # all rather than an unrecognized one, which has to block just the same.
+  printf ' 1..1\n+ok 1\n' > "$TESTOUT/regression.diffs"
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "unrecognized"
+}
+
+@test "verify-results-pgtap.sh: detects a pgtap failure" {
+  write_results failing '1..2
+ok 1
+not ok 2 - broken'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "pgtap failure detected"
+  assert_contains "$output" "not ok 2 - broken"
+}
+
+@test "verify-results-pgtap.sh: ignores a TODO failure" {
+  write_results todo '1..1
+not ok 1 - known issue # TODO fix later'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_success
+}
+
+@test "verify-results-pgtap.sh: detects a pgtap plan mismatch" {
+  write_results planned '1..3
+ok 1
+ok 2
+# Looks like you planned 3 tests but ran 2'
+
+  run "$SCRIPT" "$TESTOUT"
+  assert_failure_with_status 1
+  assert_contains "$output" "pgtap plan mismatch"
+}
+
+@test "verify-results-pgtap.sh: usage error when called with no TESTOUT argument" {
+  run "$SCRIPT"
+  assert_failure
+  assert_contains "$output" "Usage:"
+}
+
+# vi: expandtab ts=2 sw=2