diff --git a/generate/derivation/extract.py b/generate/derivation/extract.py
index 27751349..8445dfe5 100644
--- a/generate/derivation/extract.py
+++ b/generate/derivation/extract.py
@@ -30,6 +30,13 @@ this module, so none of this can move the serving ``3/47/0``):
   units are blanked (empty, like a sentence-final number): the value stays
   grounded, the unit is honestly unknown. Closed lexeme set (``_NON_UNIT_WORDS``),
   not a grammar template (ADR-0165).
+* **EX-6 — hyphen-bonded number-units** (ADR-0163-F2). A number bonded to its unit
+  by a hyphen (``25-foot``, ``20-inch``, ``2.5-mile``) was invisible to the base
+  ``number + space + word`` pattern, so the completeness check never saw the
+  divisor — the blind spot behind the pseudo-accumulation confusers (0005/0007).
+  Tight lexeme (digit run, hyphen, alphabetic unit word); the alphabetic-only unit
+  group keeps numeric ranges (``3-5``) out and only the first hyphen segment is
+  taken, so it stays clear of the deferred EX-3 multi-word-unit traps below.
 
 EX-3 (multi-word units) is deliberately **not** integrated. Two distinct traps
 defeat the tightest lookahead-anchored rule the brief admits:
@@ -94,6 +101,19 @@ _FINAL_NUMBER_RE: Final[re.Pattern[str]] = re.compile(
     r"(?<![\w.])(\d+(?:\.\d+)?)(?=\s*(?:[.?!]|$))"
 )
 
+# EX-6: a number bonded to its unit by a hyphen (``25-foot``, ``20-inch``,
+# ``2.5-mile``). The base ``_QTY_RE`` requires whitespace before the unit word,
+# so a hyphen-bonded unit was invisible — the blind spot behind the
+# pseudo-accumulation confusers (``25-foot`` / ``20-inch`` divisors the
+# completeness check never saw). Tight lexeme: a digit run, a single hyphen, an
+# alphabetic unit word. The trailing ``[a-zA-Z]+`` (not ``\d``) keeps numeric
+# ranges (``3-5``) out, and taking only the first hyphen segment keeps the
+# postmodifier tail (``25-year-old`` -> unit ``year``) from inflating the unit —
+# so this stays clear of the deferred EX-3 multi-word-unit traps.
+_HYPHEN_QTY_RE: Final[re.Pattern[str]] = re.compile(
+    r"(?<![\w.])(\d+(?:\.\d+)?)-([a-zA-Z]+)"
+)
+
 
 # Function words that are never units. When the token immediately after a number
 # is one of these (``$0.75 each``, ``$40 to go``, ``3/4 of``), the single-word unit
@@ -162,9 +182,10 @@ def extract_quantities(problem_text: str) -> tuple[Quantity, ...]:
     passes never double-count a number:
 
     1. EX-4 same-unit list (claims every number in the list);
-    2. digit + single unit word (skips numbers a list already claimed);
-    3. EX-1 word-number + unit word (alphabetic, disjoint from digit spans);
-    4. EX-5 sentence-final bare number (skips any already-claimed digit).
+    2. EX-6 hyphen-bonded number-unit (``25-foot``; claims the digit span);
+    3. digit + single unit word (skips numbers a list/hyphen pass already claimed);
+    4. EX-1 word-number + unit word (alphabetic, disjoint from digit spans);
+    5. EX-5 sentence-final bare number (skips any already-claimed digit).
     """
     found: list[tuple[int, Quantity]] = []
     claimed: list[tuple[int, int]] = []
@@ -179,6 +200,16 @@ def extract_quantities(problem_text: str) -> tuple[Quantity, ...]:
                 found.append((pos, quantity))
                 claimed.append((pos, pos + len(num.group(0))))
 
+    # 1b. EX-6 — hyphen-bonded number-unit (``25-foot``). Claims the digit span so
+    #     the bare/final passes never re-surface the number with a blank unit.
+    for match in _HYPHEN_QTY_RE.finditer(problem_text):
+        if _claimed(match.start(1), claimed):
+            continue
+        quantity = _quantity(match.group(1), match.group(2))
+        if quantity is not None:
+            found.append((match.start(1), quantity))
+            claimed.append(match.span(1))
+
     # 2. digit + single unit word — the original base pattern.
     for match in _QTY_RE.finditer(problem_text):
         if _claimed(match.start(1), claimed):
diff --git a/tests/test_adr_0163_f2_confusers.py b/tests/test_adr_0163_f2_confusers.py
index de74cc78..eb1058f5 100644
--- a/tests/test_adr_0163_f2_confusers.py
+++ b/tests/test_adr_0163_f2_confusers.py
@@ -17,9 +17,14 @@ from evals.gsm8k_math.confusers.v1.runner import (
     summarize,
 )
 
-# Honest measured baseline (2026-05-29). The probe revealed the sealed composers
-# wrongly answer these confuser categories; this pins them so they cannot grow.
-_BASELINE_WRONG = 7
+# Honest measured baseline. The probe revealed the sealed composers wrongly answer
+# these confuser categories; this pins them so they cannot grow.
+#   2026-05-29 (initial): wrong=7, pair_tells=4.
+#   2026-05-29 (EX-6 hyphen-bonded units, ADR-0163-F2): the two pseudo-accumulation
+#     misfires (0005->796, 0007->996) now refuse — the 25-foot/20-inch divisor is
+#     finally visible to the completeness/polarity gate. wrong 7->5. pair_tells
+#     unchanged (the pseudo-accumulation cases carry no positive twin).
+_BASELINE_WRONG = 5
 _BASELINE_PAIR_TELLS = 4
 
 
@@ -68,5 +73,19 @@ class TestProbeBaseline:
     def test_pair_tells_do_not_regress(self) -> None:
         assert len(pair_inconsistencies()) <= _BASELINE_PAIR_TELLS
 
+    def test_pseudo_accumulation_does_not_misfire(self) -> None:
+        # ADR-0163-F2 EX-6: the hyphen-bonded-unit extraction made the
+        # 25-foot/20-inch divisor visible, so the pseudo-accumulation confusers
+        # (0005, 0007) refuse instead of committing 796/996. This is the specific
+        # obligation the fix proves — it fails loudly if the hyphen pass regresses
+        # (the gate would again miss the divisor and the category would misfire).
+        results = run_probe()
+        pseudo = [r for r in results if r.category == "pseudo-accumulation"]
+        assert pseudo, "pseudo-accumulation cases missing from corpus"
+        assert all(r.verdict != "wrong" for r in pseudo), (
+            "pseudo-accumulation confuser misfired — the hyphen-bonded-unit "
+            "divisor is no longer reaching the completeness/polarity gate."
+        )
+
     def test_deterministic(self) -> None:
         assert summarize(run_probe()) == summarize(run_probe())
diff --git a/tests/test_adr_0179_extract.py b/tests/test_adr_0179_extract.py
index e7595ce9..207e27e3 100644
--- a/tests/test_adr_0179_extract.py
+++ b/tests/test_adr_0179_extract.py
@@ -94,6 +94,71 @@ class TestNoRegression:
         assert _triples("It costs 0.75 dollars.") == [(0.75, "dollars", "0.75")]
 
 
+class TestEX6HyphenatedUnitNumbers:
+    """ADR-0163-F2 — hyphen-bonded number-unit tokens (``25-foot``, ``20-inch``).
+
+    The base ``_QTY_RE`` requires ``number + whitespace + unit word``, so a number
+    bonded to its unit by a hyphen (``25-foot sections``, ``20-inch pieces``) was
+    invisible — the very blind spot behind the pseudo-accumulation confusers
+    (0005 ``→796``, 0007 ``→996``): the completeness clause could not see the
+    ``25``/``20`` divisor, so the accumulation reading was wrongly "complete".
+
+    This is a tight, ADR-0165-safe lexeme pattern (digit run, a hyphen, an
+    alphabetic unit word) — strictly distinct from the deferred EX-3 multi-word
+    *space-separated* unit problem. Over-extraction here only costs refusals
+    (the gate is refuse-preferring), never a wrong answer.
+    """
+
+    def test_hyphen_bonded_unit_extracts_value_and_unit(self) -> None:
+        assert _triples("She cuts it into 20-inch pieces.") == [(20.0, "inch", "20")]
+
+    def test_hyphen_bonded_unit_mid_sentence(self) -> None:
+        assert _triples("She splits it into 25-foot sections.") == [(25.0, "foot", "25")]
+
+    def test_decimal_hyphen_bonded_unit(self) -> None:
+        assert _triples("He ran a 2.5-mile loop.") == [(2.5, "mile", "2.5")]
+
+    def test_hyphen_unit_not_double_counted_as_final_number(self) -> None:
+        # the hyphen pass claims the digit span; the bare-final pass must not
+        # also surface "25" with an empty unit.
+        assert len(extract_quantities("It was 25-foot.")) == 1
+
+    def test_word_compound_still_takes_word_number_path(self) -> None:
+        # digit-hyphen is a different lexeme from the EX-1 word-word compound;
+        # "twenty-four" must still resolve via the word-number path, unaffected.
+        assert _triples("The team scored twenty-four points.") == [
+            (24.0, "points", "twenty-four"),
+        ]
+
+    def test_numeric_range_is_not_read_as_unit(self) -> None:
+        # "3-5" is digit-hyphen-DIGIT; the unit group requires letters, so the
+        # hyphen pass must not fire and invent a unit from the second number.
+        assert all(q.unit != "5" for q in extract_quantities("Pick 3-5 apples."))
+
+
+class TestSlashFractionLeakHazard:
+    """Honest pin (deferred hazard): a slash-fraction leaks its denominator.
+
+    ``"gives 1/4 to a friend"`` extracts the bare ``4`` (the base ``_QTY_RE``
+    sees ``4`` + space + the function-word ``to``, blanks the unit, and emits a
+    standalone quantity ``4``). This is a grounded-but-wrong operand — it is the
+    second half of the pseudo-accumulation misfire (``1000 - 4 = 996``).
+
+    It is **not** fixed in this PR on purpose. Suppressing the leaked ``4`` *removes*
+    a quantity, which can *unblock* the completeness clause (a derivation that was
+    "incomplete" because it ignored the spurious ``4`` could become "complete" and
+    commit) — i.e. the fix is not unambiguously refuse-preferring and needs its own
+    train_sample + probe validation. The hyphen-unit pass (TestEX6) already drives
+    confusers 0005/0007 to *refuse* via the polarity-None ``cuts``/``splits`` clause,
+    so this leak is currently dormant behind that refusal. This test pins the leak
+    so a future fraction-operand PR addresses it deliberately, not by accident.
+    """
+
+    def test_slash_fraction_denominator_currently_leaks(self) -> None:
+        # documents current (leaky) behavior — flip this when fractions are modeled.
+        assert (4.0, "", "4") in _triples("She gives 1/4 to a friend.")
+
+
 class TestEX3StillDeferred:
     """Honest pin: EX-3 (multi-word units) remains deferred after the Track C redo.