As I understand the rules, the game may be described as follows:

MATERIALS AND SET-UP:

The game pieces consist of 104 tiles printed (on one side) with a number
(1..13) and a color (Red,Orange,Blue,blacK). There are two of each combination.
Two additional tiles (the "jokers"), also blank on one side, are "wild".

The game is played by N (>=2) players taking turns cyclically. At all
times the tiles are partitioned into subsets:
	1. A "draw pile" P (tile markings invisible to all players)
	2. A "display tableau" T (tile markings visible to all players)
	3. Each person's "rack pile" R_i (tile markings invisible to 
		player  i  only)
Initially  T  is empty and each  R_i  consists of a set of 14 randomly 
selected tiles;  P  contains the other  106-14N  tiles.

NORMAL TURNS:

Player  i's  turn consists of moving a nonempty subset  S  of tiles from  
R_i  to  T, the only stipulation being that the resulting enlarged set  T
be  "divisible", which I will define below. If no such set  S  exists, then
player  i  transfers a randomly-selected element of  P  to  R_i.

Play continues in this way until either  P  or some  R_i  is empty. It's not
entirely clear to me what the rules say in those cases. Here is how I
interpret them.

SCORING SITUATIONS:

In the latter situation, some  R_i  is empty. (This is called the "end of
the round" in the rules.)  At this point, scores
are assigned to the players according to an algorithm outlined below.
_Apparently_, all the elements of all subsets  R_j  are returned to  P
(and shuffled); each player is then assigned a random subset of
	max(14, floor( |P|/N ) )
cards from  P  and play resumes as before.

Note that at each move, the number  |T| - |P|  must increase by at least  1;
the number is bounded by  106,  of course, so after at most 106 moves, we
must run into the other -- terminal -- situation in which  P  is empty.
In that case play continues as before but of course players who cannot pass
a nonempty set  S  to  T  do not receive any new tiles. When all players
state they can no longer diminish the size of R_i, the game is over; then 
scores are assigned by a slightly different algorithm.

(I'm not sure what happens if at some point  P  becomes empty, and then
some  R_i  becomes empty -- does player   i  receive points according to
the normal scoring algorithm in this case, and _then_ the other players
keep trying to reduce their racks? Or is it simply an endgame situation,
players who reach  R_i = {}  simply assured a high last-round score?)

Final score for each player is the sum of all scores assigned along the way,
and of course the player with the highest score is declared the winner.

SCORING ALGORITHM:

The algorithm for assigning points when some  R_i  is empty is to assign
a negative score to each player  j <> i  equal to the sum of the numerical
values of the tiles in  R_j  at that time, the "wild" cards taken to have
a value of  30. R_i  is assigned a score equal to the negative of the sum
of the other players' scores, that is, this is a zero-sum game.

In the unlikely event that all players have no further possible moves and
P  is empty but no  R_i  is, the minimum point total (as computed in the
previous paragraph) is subtracted from all the other point totals; then
positive or negative scores are assigned as before, keeping this a
zero-sum game. (I don't understand what is intended if in this event there
is actually a tie for lowest rack total.)

DIVISIBLE SETS:

A set  T  of tiles is "divisible" if it can be partitioned into subsets
T_k, each of cardinality at least 3, and each consisting either of
	a) tiles with the same numerical value and distinct colors ("groups"),
or	b) tiles with the same color and consecutive numerical values ("runs").
(Numerical values are linear, not cyclic). A "wild" card can be used as
if it had any numerical value or color.

The rules of play suggest that the set  T  be placed in view of the players
in its partitioned form, but as I read the rules, the players have complete
freedom to re-partition the set  T  at each turn, so there seems no reason
to do this except, of course, to demonstrate that the set  T  is in fact
divisible. There is a technical exception to this in the next section.

OTHER RULES:

Typical highest-draw-goes-first, play-proceeds-clockwise rules.

Until a player may pass a set  S  with numerical sum 30 or more from  R_i to T,
that player must not pass anything (but rather must accept a random tile from
P, as above). For the purposes of this rule, a "wild" tile is given the value 
of the tile it represents.

Subject to the rules above, a player may choose on his turn to pass any set
S  from  R_i  to  T  leaving  T  divisible, especially, may select  S = {} !

Rules specify a 2-minute time limit per move; failure to move in that time
is consider equivalent to passing  S = {}  (and so, accepting a tile from  P).

Rules also specify that an attempt to pass a set  S  which leaves  T  not
"divisible" will result in passing  S = {} and the receipt of not one but
three tiles from  P.

There are special rules for the handling of jokers which actually seem to
require that the set  T  be at all times not merely "divisible" but
actually "divided". Players can apparently repartition  T  at will when
adding new tiles from  R_i, but with an exception related to the jokers:
	If a joker is removed from a subset, it must be replaced by a
tile *from R_i*  rather than from other part of  T. Also, the new subset to
which the joker is moved must contain two or more tiles *from R_i*.
	(Actually implementing this rule seems to require that the new
tableau  T  be obtained from the old one by a sequence of incremental
changes which allow the players to decide which subsets in the before
and after pictures are "the same"! My copy of the rules lists these as
extending a run, stripping a fourth tile from a "group" of 4, splitting
a "run", and combinations thereof).