docs(adr-0179-ex3): record second deferral trap — postmodifier adjectives (#460)

Track C of docs/handoff/PARALLEL-WORK-PLAN-2026-05-29.md asked for a tight
EX-3 multi-word-unit redo satisfying (a) "12 jumping jacks." -> "jumping
jacks", (b) "6 apples and 4 apples." -> two apples, (c) all GB-1/2/3 tests
green. The cleanest tight rule that satisfies all three —

    (?<![\w.])(\d+(?:\.\d+)?)\s+([a-z]+\s+[a-z]+)(?=\s*[.?!,]|\s*$)

— was implemented and passed the four pinned test files. Full-suite
verification then surfaced a second trap the audit at
docs/handoff/AUDIT-ADR-0179-EX-RECONCILE.md did not anticipate:

  postmodifier-adjective tails. "25 years old?" fires the tight rule and
  produces unit "years old" rather than "years", regressing
  test_adr_0176_ms1_question_target.py::TestQuestionQuantities::
  test_extracts_quantity_stated_in_question and the "X years old" pattern
  in tests/test_adr_0176_ms2_chain.py. The pattern is endemic in GSM8K
  (cases 0006 and 0033 both use "X years old"); closing it would need a
  second closed lexeme set ({old, tall, long, wide, deep, away, ago, ...})
  which the brief judged too open-ended to enumerate responsibly.

Per the brief's escape hatch ("If no rule satisfies all of (a)-(c) without
a grammar template, write a note and ship no code — a refusal is fine")
this commit:

* updates extract.py's module docstring to name BOTH known traps
  (connective-crossing AND postmodifier-adjective tails);
* adds tests/test_adr_0179_extract.py::TestEX3StillDeferred with two pins
  asserting the postmodifier-adjective shape stays at unit "years" alone,
  so no future redo silently re-introduces the regression;
* ships NO extractor code change — the regex remains exactly as on main.

Scope/safety:
* Files touched are within Track C's allowed set (extract.py + its test).
* Zero functional change: extract_quantities byte-identical to main.
* Serving lane untouched (chat/ does not import this module).
* Safe alongside GB-3b on compose.py / clauses.py.

generate/derivation/extract.py

@@ -24,11 +24,28 @@ this module, so none of this can move the serving ``3/47/0``):
   sentence/text or before terminal punctuation) extracts with an empty unit so it
   stays available to the completeness check without inventing a unit lexeme.
 
-EX-3 (multi-word units) is deliberately **not** integrated: the greedy lowercase
-unit span regresses GB-2's same-unit detection (``"6 apples and 4 apples"`` →
-unit ``"apples and"``) and does not cleanly recover real multi-word units from
-0024-class text (``"20 jumping jacks on Monday"`` → ``"jumping jacks on"``). See
-``docs/handoff/AUDIT-ADR-0179-EX-RECONCILE.md``.
+EX-3 (multi-word units) is deliberately **not** integrated. Two distinct traps
+defeat the tightest lookahead-anchored rule the brief admits:
+
+1. **Connective-crossing** (in
+   ``docs/handoff/AUDIT-ADR-0179-EX-RECONCILE.md``). The greedy lowercase unit
+   span regresses GB-2's same-unit detection (``"6 apples and 4 apples"`` → unit
+   ``"apples and"``) and does not cleanly recover real multi-word units from
+   0024-class text (``"20 jumping jacks on Monday"`` → ``"jumping jacks on"``).
+2. **Postmodifier-adjective tails** (discovered during the Track C redo of
+   ``docs/handoff/PARALLEL-WORK-PLAN-2026-05-29.md``). Even a *tight*
+   ``digit + lc word + lc word + (?=clause-terminator)`` rule fires on
+   ``"25 years old?"`` and produces unit ``"years old"`` instead of
+   ``"years"`` — regressing
+   ``tests/test_adr_0176_ms1_question_target.py::TestQuestionQuantities::test_extracts_quantity_stated_in_question``.
+   The pattern is endemic: GSM8K cases 0006/0033 and several MS2 chain tests use
+   ``"X years old"``. Closing it would need a second closed lexeme set (a stop
+   list of measurement postmodifiers — ``old``, ``tall``, ``long``, ``wide``,
+   ``deep``, ``high``, ``away``, ``apart``, ``ago``, …) which the audit did not
+   anticipate and which the Track C brief judged too open-ended to enumerate
+   responsibly. ``TestEX3StillDeferred`` in
+   ``tests/test_adr_0179_extract.py`` pins this second trap so no future redo
+   silently re-introduces it.
 """
 
 from __future__ import annotations

tests/test_adr_0179_extract.py

@@ -94,6 +94,55 @@ class TestNoRegression:
         assert _triples("It costs 0.75 dollars.") == [(0.75, "dollars", "0.75")]
 
 
+class TestEX3StillDeferred:
+    """Honest pin: EX-3 (multi-word units) remains deferred after the Track C redo.
+
+    The brief in ``docs/handoff/PARALLEL-WORK-PLAN-2026-05-29.md`` Track C asked
+    for a *tight* rule satisfying:
+
+    * (a) ``"12 jumping jacks."`` → unit ``"jumping jacks"`` only where tight;
+    * (b) ``"6 apples and 4 apples."`` → two ``apples`` quantities (not
+      ``"apples and"``);
+    * (c) all GB-1/GB-2/GB-3 tests stay green;
+
+    plus the plan's global rule that *no currently-green test* be regressed.
+
+    A candidate rule —
+    ``(?<![\\w.])(\\d+(?:\\.\\d+)?)\\s+([a-z]+\\s+[a-z]+)(?=\\s*[.?!,]|\\s*$)``
+    — satisfies (a)+(b)+(c) against the four listed test files but **regresses
+    a different green test**: ``test_adr_0176_ms1_question_target.py
+    ::TestQuestionQuantities::test_extracts_quantity_stated_in_question``,
+    which pins ``"25 years old?"`` → unit ``"years"``. The candidate rule fires
+    on ``"25 years old"`` and produces unit ``"years old"`` — and the
+    postmodifier-adjective tail (``"old"`` / ``"tall"`` / ``"long"`` / ``"wide"``
+    / ``"away"`` / ``"ago"`` / …) is endemic in GSM8K (cases 0006, 0033 and
+    several MS2 chain tests all use ``"X years old"``).
+
+    The audit at ``docs/handoff/AUDIT-ADR-0179-EX-RECONCILE.md`` named the first
+    trap (connective-crossing). This pin names the second one
+    (postmodifier-adjective tails), so the deferred status of EX-3 is owned by
+    *both* known traps and a future redo cannot silently land without addressing
+    them.
+
+    These assertions hold against the *current* (no-EX-3) extractor. Any patch
+    that introduces a multi-word-unit pass without also blocking postmodifier
+    tails will flip them — that is the point.
+    """
+
+    def test_postmodifier_adjective_does_not_inflate_unit_to_two_words(self) -> None:
+        # The trap. Naive tight EX-3 produces "years old"; the correct unit is
+        # "years" alone.
+        qs = extract_quantities("How old will the father be when she is 25 years old?")
+        assert [(q.value, q.unit) for q in qs] == [(25.0, "years")]
+
+    def test_postmodifier_trap_present_for_full_sentence_form_too(self) -> None:
+        # The same trap with a period instead of a question mark, paralleling
+        # ``"Rachel is 12 years old."`` from GSM8K case 0033 / the MS2 chain tests.
+        qs = extract_quantities("Rachel is 12 years old.")
+        assert qs[0].value == 12.0
+        assert qs[0].unit == "years"
+
+
 class TestRealCase0024StillBlocked:
     """Honest pin: the EX-4 unit-list pattern does NOT recover real case 0024.