The ProphIt is Coming!

The ProphIt started out with a simple premise, what if I could offer to create bespoke forecasters for clients, based on their personal idea of what are valuable signals for a (particular) market outcome?

Really simple. What could be easier? Smash that out in a weekend.

So simple all it required was:

1. A full stack development from scratch of a new combined research/forecasting app on my mac. One that was capable of forecasting all of the standard market offers and archive them all simultaneously? That means many 1,000's of settled, forward tracked fixtures, all in a readily usable form...not bleeding into each other, but often sharing many similar inputs and app processes.

2. Deal with spinning rainbow wheels as the architecture revealed where any particular function was overworked/overloaded as archive grew - solve those puzzles without losing completed work.

3. Archive the poorest performers and keep the feasible maturing. Don't erase them. They will have signals inside that perhaps later can be mined, or they can be resumed with a production ready ProphIt.

4. Debug/rework the app during this process as blockages/lumps/shallow parts of the process/lags presented. Add new features as they occur and are deemed necessary.

5. Decide I had mature archives worthy of refining. Let's do that. Oh. Next number in this list...

6. Discover the app wasn't deep diving enough when it was cataloguing signals - it was capturing them but it wasn't surfacing them readily. Rework/reiterate again.

7. Finally, have a ProphIt with "client" archives I could refine. We have matured! I suspect a lot more than required for the work, but I let it run a long time to try and catch those infamous edge cases (weren't any so far lol).

Continue forecasting to see if we end up with a credible forecast output, i.e., a potential for positive ROI and not just accuracy alone (i.e. not so favourite heavy that one loss erases many wins).

8. Actually, this is still 7. because that is where I am with ProphIt today.

The header image is now.

Today I've taken the two models that I didn't pause and have been accumulating archive for months. I shadowed the criteria that do not appear to be contributing to the winning selections. Now the ProphIt will continue to prophesize and I will see if the dial moves in the right direction?

And if it does I will add a test forecasting upload to the pipeline - to an iOS app (modified clone of my production iOS for this purpose - not the actual app I will build for clients) so I can watch the fixture upload daily as a user would and see how it feels, and importantly, judge how it works?

If I had paid for this to be created as a client would I be happy with the outcome?

Would I think the coverage too slim (how to expand then?)

Would I trust it?

What metrics would I need to see to build that trust?

It won't just be a good W/L in a market I like, especially if there's a lot of favourites. I'll need to know/understand that this is researched and archive backed. That one bad day was just that, variation of the kind we never like, but tomorrow it will vary back...

And how will I design the new phone app that will receive this output which fits neatly onto the end of this new pipeline?

Time will tell.

The ProphIt is entering a new stage, and I want it to talk clearly every day. I want to see the future...well, just enough that I would be happy to put my own [insert only for entertainment and informational purposes here :)]...