Please see the gist below for a self-explaining example
copied here for convenience.  Emacs 24.5.1 with org-mode
8 something (not sure how to get it to report its exact
version)... a very late version, I am sure.  Verified with
no init.el, that is, with emacs -Q

https://gist.github.com/rebcabin/37f800da658f4b23ceaa

#+TITLE: Org-Babel Bug?
#+AUTHOR: Brian Beckman
#+EMAIL: bc.beckman@gmail.com

#+BEGIN_SRC emacs-lisp :exports results :results none
  (setq org-confirm-babel-evaluate nil)
  (org-babel-map-src-blocks nil (org-babel-remove-result))
#+end_src

* Mystery Number 1

The first line of the first block must be blank, or we must =C-c C-c= the block
two times. But we want to eval the entire file /via/ =C-c C-v C-b=, and we found
that the first line must be empty.

If you =C-c C-v C-b= this entire file, the python session buffer, named
=*bug-org-babel*=, contains an error message that suggests the variable =_= is
implicated in an error.

We expect the value of this first block to be 999.  Make sure that the python
session is clear by typing =quit()=, then evaluate this entire file by
=C-c C-v C-b=.

#+NAME: probe
#+BEGIN_SRC python :session bug-org-babel :exports both :results value

  999
#+END_SRC

#+RESULTS: probe
: 999

* Mystery Number 2

Now, we set the value of the variable =_= to something arbitray. In a real
scenario, this variable may be set casually in a loop or some other context. The
variable =_= is often recommended for /ad-hoc/ use, as in "don't care."

#+NAME: second-probe
#+BEGIN_SRC python :session bug-org-babel :exports both :results value
  _ = 42
  _
#+END_SRC

#+RESULTS: second-probe
: 42

But now, all blocks with =:results value= return the value of =_=. We broke
org-babel!

#+NAME: third-probe
#+BEGIN_SRC python :session bug-org-babel :exports both :results value
  999
#+END_SRC

#+RESULTS: third-probe
: 42