I talk myself into software

I will add  a postpend,  to go with prepend; as well as dump in the pit, methods for moving subraphs around. Then write a little counter that goes through sorted singletons and groups them by identity, and sets their probability.  It will be fun, throwing loads of odd stock symbols at it, using different repetition rates.

Fort example.  Compute the actual Huffman tree for a large message, then run my symbols into the pit and organize them by keep the queuer within some size, limiting the Huffman window size.  I can see how the tree changes as the message gets flushed through.

Then I need swap, a graph method that imply swaps two singletons in the same block.  Sort of a short postpend and prepend sequence, it checks the bounds, does the swap.  Swap gives the pit boss a rapid method to  run through the singletons,maybe grabbing the blue chips and getting grouped  and up the tree, quickly.  Each move, actually, can be considered a series of select, post and prepends with control logic.  The chaos is always at the bottom as singletons pile in, I write efficient methods on those, as the tree is structured,up, I care less.  Up the tree, innovations are less, here is less re-organization to do and the precision is small, 5-9 bits, so implied for loops are bound.  In the bottom, a sudden surge and the pit boss has to rush down and scan maybe 30 singletons.

Like swap, an operation directly on the array using indices.  OK, let us make one index a list.  Then the swap moves some unordered list of singletons back to replace a list of singletons grouped, by the boss.   Go through the singletons, make an index list of all the blue hips, call GROUP,, The group list is moved to the head of the singleton, using calls to SWAP. The contiguous segment is then enclosed, promoted with anchor,  by the gouper,one dot up.

And take this approach, break down the array operations into fast methods on the g array,then let the bots use concatenations of these.  As opposed to grabbing a built on, not necessarily meeting our smoothness assumptions.  Here is another way to think about well shaped.  list are well shaped, meaning that if I enclosed a complete huffman graph onto the list, by depth.  Then, on popping out the tree, precision of the reconstructed tree will rise smoothly with cycles to do a pop, almost always. A hash table? A sort?  Maybe not.  So, the bots can use lists of indices.

Otherwise what happens? Unbeknownst to us, the hash table has a much shorter cycle with small symbols.  So yhe hacker bot, waits, and stalks the singletons until he finds a cluster of small names, he waits until the pit boss runs, then he jumps on those names, knowing has has a nice list to arbitrage, and extra cycles to do it.  He will wear out the other with extra bid cycles.