When I first got in contact with Tom Veale he directed me to this link you mentioned:
Jack Park (http://www.thinkalong.com ) did some interesting stuff many years ago that you might find interesting.
I found it very interesting as it described the Optimisation Objectives for my Vortex AIM and was illustrated conceptually in my first (unpublished) Write-up of Investing with Understanding(Free title translation of Beleggen met Begrip).
I proposed a auto-adjusting algorithm as is used in self-learning process control systems. The idea was exactly as was tried on the Think Along website. I never got beyond the basic definition of the purpose and how it could be done in a step by step manner along the lines with N-dimensional mathematical equation parameter optimisation.
My understanding is that Jack Park is no longer working on this optimisation. Also, in AI you have something similar, if I understand it correctly.
Questions/Remarks: 1)Is your optimisation routine functional on line so that it optimises the parameters on the active stock as you go or do you use it only as Model Building with historic data?
2) My objective was(is?) to do the optimisation each time a new stock price is entered. I am not sure if that is a reasonable objective but I see it as follows:
2.1) You start with a stock price history, say 30 days, and use that to do the initial Model Building. 2.2) Then each time a new stock price is added for the active stock(as frequently as you want to do that) then the Program first runs it optimisation subroutine and then comes out with a Buy/sell Order. 2.3) Of course, this operation should leave the previous buys and sells unaltered. Each optimisation run is actually building a new model, using the price data as it develops and then sets up the Model as if you start a new investment.
The new parameters would specify a new cash/stock ratio to start with. This amounts to Re-balancing based on the optimisation run. The next action then is to start a buy or sell or to make a extra cash or stock contribution. Essentially then the optimisation will result is New START-UP Advise.
To what extend is this objective already realised by you or anyone?
Clearly, this type of optimisation may exclude what some people consider pattern recognition, but the nature of the optimisation subroutine could be varied. I can imagine a pattern recognition routine that defines the form of the stock price profile and then define the Buy/Sell Advice on that pattern.
Of course, this concept probably includes bit and pieces from conversations held the last 6 months or so. It is a crystalization of the idea I had years ago, and may be modified with the things that were presented in the AIM Discussions
Am I dreaming or am I on the right track from a practical consideration?
Market Data 
Markets 
AI
