Shouldn't using WITH return the same results as if you'd put the results in a table first?

Question

First off, here's my version info:
BANNER
Oracle Database 11g Enterprise Edition Release 11.1.0.7.0 - 64bit Production
PL/SQL Release 11.1.0.7.0 - Production
CORE 11.1.0.7.0 Production
TNS for HPUX: Version 11.1.0.7.0 - Production
NLSRTL Version 11.1.0.7.0 - Production
I just reread the documentation again on the subquery factoring clause of SELECT statement, and I didn't see any restrictions that would apply.
Can someone help me understand why I'm getting different results?  I'd like to be able to use the statement that creates MAT3, but for some reason it doesn't work.  However, when I break it up and store the last TMP subquery in a table (MAT1), I'm able to get the expected results in MAT2.
Sorry if the example seems a little esoteric.  I was trying to put something together to help illustrate another problem, so it was convenient to use the same statements to illustrate this problem.
drop table mat1;
create table mat1 as
with skus as (
  select level as sku_id
  from dual
  connect by level <= 1000
  tran_dates as (
  select to_date('20130731', 'yyyymmdd') + level as tran_date
  from dual
  connect by level <= 31
  sku_dates as (
  select s.sku_id,
  t.tran_date,
  case when dbms_random.value * 5 < 4
  then 0
  else 1
  end as has_changes,
  round(dbms_random.value * 10000, 2) as unit_cost
  from skus s
  inner join tran_dates t
  on 1 = 1
select d.sku_id,
  d.tran_date,
  d.unit_cost
  from sku_dates d
  where d.has_changes = 1
drop table mat2;
create table mat2 as
select m.sku_id,
  m.tran_date,
  m.unit_cost,
  min(m.tran_date) over (partition by m.sku_id order by m.tran_date rows between 1 following and 1 following) as next_tran_date
  from mat1 m
drop table mat3;
create table mat3 as
with skus as (
  select level as sku_id
  from dual
  connect by level <= 1000
  tran_dates as (
  select to_date('20130731', 'yyyymmdd') + level as tran_date
  from dual
  connect by level <= 31
  sku_dates as (
  select s.sku_id,
  t.tran_date,
  case when dbms_random.value * 5 < 4
  then 0
  else 1
  end as has_changes,
  round(dbms_random.value * 10000, 2) as unit_cost
  from skus s
  inner join tran_dates t
  on 1 = 1
  tmp as (
  select d.sku_id,
  d.tran_date,
  d.unit_cost
  from sku_dates d
  where d.has_changes = 1
select m.sku_id,
  m.tran_date,
  m.unit_cost,
  min(m.tran_date) over (partition by m.sku_id order by m.tran_date rows between 1 following and 1 following) as next_tran_date
  from tmp m
select count(*) from mat2;
select count(*) from mat3;
  from tmp m
select count(*) from mat2;
select count(*) from mat3;
select count(*) from mat2;
  COUNT(*)
     31000
Executed in 0.046 seconds
select count(*) from mat3;
  COUNT(*)
         0
Executed in 0.031 seconds

Answer 1

I think there's something else going on.
I made the change you suggested, with a slight modification to retain the same functionality of flagging ~80% of the rows as not having changes.  I then copied that section of my script - included below - and pasted it into my session twice.  Unfortunately, I got different results each time.  I have had a number of strange problems when using the WITH clause, which is one of the reasons I jumped at posting something here when I encountered it again in this context.
Can you help me understand why this would happen?
drop table mat3;
create table mat3 as
with skus as (
  select level as sku_id
  from dual
  connect by level <= 1000
  tran_dates as (
  select to_date('20130731', 'yyyymmdd') + level as tran_date
  from dual
  connect by level <= 31
  sku_dates as (
  select s.sku_id,
  t.tran_date,
  case when dbms_random.value(1,100) * 5 < 400
  then 0
  else 1
  end as has_changes,
  round(dbms_random.value * 10000, 2) as unit_cost
  from skus s
  inner join tran_dates t
  on 1 = 1
  tmp as (
  select d.sku_id,
  d.tran_date,
  d.unit_cost
  from sku_dates d
  where d.has_changes = 1
select m.sku_id,
  m.tran_date,
  m.unit_cost,
  min(m.tran_date) over (partition by m.sku_id order by m.tran_date rows between 1 following and 1 following) as next_tran_date
  from tmp m
select count(*) from mat2;
select count(*) from mat3;
152249 < mattk > drop table mat3;
Table dropped
Executed in 0.016 seconds
152249 < mattk > create table mat3 as
                         2  with skus as (
                         3   select level as sku_id
                         4   from dual
                         5   connect by level <= 1000
                         6   ),
                         7   tran_dates as (
                         8   select to_date('20130731', 'yyyymmdd') + level as tran_date
                         9   from dual
                        10   connect by level <= 31
                        11   ),
                        12   sku_dates as (
                        13   select s.sku_id,
                        14   t.tran_date,
                        15   case when dbms_random.value(1,100) * 5 < 400
                        16   then 0
                        17   else 1
                        18   end as has_changes,
                        19   round(dbms_random.value * 10000, 2) as unit_cost
                        20   from skus s
                        21   inner join tran_dates t
                        22   on 1 = 1
                        23   ),
                        24   tmp as (
                        25   select d.sku_id,
                        26   d.tran_date,
                        27   d.unit_cost
                        28   from sku_dates d
                        29   where d.has_changes = 1
                        30   )
                        31  select m.sku_id,
                        32   m.tran_date,
                        33   m.unit_cost,
                        34   min(m.tran_date) over (partition by m.sku_id order by m.tran_date rows between 1 following and 1 following) as next_tran_date
                        35   from tmp m
                        36  ;
Table created
Executed in 0.53 seconds
152250 < mattk > select count(*) from mat2;
  COUNT(*)
         0
Executed in 0.047 seconds
152250 < mattk > select count(*) from mat3;
  COUNT(*)
     31000
Executed in 0.047 seconds
152250 < mattk >
152251 < mattk > drop table mat3;
Table dropped
Executed in 0.016 seconds
152252 < mattk > create table mat3 as
                         2  with skus as (
                         3   select level as sku_id
                         4   from dual
                         5   connect by level <= 1000
                         6   ),
                         7   tran_dates as (
                         8   select to_date('20130731', 'yyyymmdd') + level as tran_date
                         9   from dual
                        10   connect by level <= 31
                        11   ),
                        12   sku_dates as (
                        13   select s.sku_id,
                        14   t.tran_date,
                        15   case when dbms_random.value(1,100) * 5 < 400
                        16   then 0
                        17   else 1
                        18   end as has_changes,
                        19   round(dbms_random.value * 10000, 2) as unit_cost
                        20   from skus s
                        21   inner join tran_dates t
                        22   on 1 = 1
                        23   ),
                        24   tmp as (
                        25   select d.sku_id,
                        26   d.tran_date,
                        27   d.unit_cost
                        28   from sku_dates d
                        29   where d.has_changes = 1
                        30   )
                        31  select m.sku_id,
                        32   m.tran_date,
                        33   m.unit_cost,
                        34   min(m.tran_date) over (partition by m.sku_id order by m.tran_date rows between 1 following and 1 following) as next_tran_date
                        35   from tmp m
                        36  ;
Table created
Executed in 0.078 seconds
152252 < mattk > select count(*) from mat2;
  COUNT(*)
         0
Executed in 0.031 seconds
152252 < mattk > select count(*) from mat3;
  COUNT(*)
         0
Executed in 0.047 seconds