Adding stuff from this post https://forum.numer.ai/t/how-to-write-a-coe-proposal/3287 to flesh this out as much as I can.
Timeline:
I finish my finals today and the semester this friday, I could begin work on this on ~May 16. I estimate I could get it done in two weeks to a month since I already have the code and understanding, I just need to clean it up (alot LOL), streamline the different scripts I use, write comments and the article, and get some rounds submitted
Success:
The clearest long term success evaluation of this project would probably be something like seeing a spike of example model submissions on main tournament in the percentile graph, or having successful signals users open up about using this type of data as inspired by my example. Inspiring people to build even better models and find better data that is the main goal. Discussion/reference about this example on rocket chat or the forums would also indicate that this example was helping the community and new users learn about signals, specifically creativity of sourcing data and dealing with ugly or uncommon APIs to get that data.
Worst case:
Satellite data publicly available has no corr(3 rounds in this seems to be untrue, but things could easily change), the specifics of this tutorial become useless but readers still learn about API’s in python and chaining them together to get a signal, creativity, and how cool signals can be(auditing companies with an eye in the sky). The tutorial will also give the reader so basic understanding of modeling of series and application of filters and models. I am not worried too much about the readability or quality of explanation since I have done and enjoyed a bit of technical writing and will be seeking community feedback throughout the process which should help to spot any failure in my explanations.
Best case:
Corr is good and users look for other similar data or are inspired to try hard/weird/uncommon data sources that often have the least exploited market information, users stake on models inspired by this or directly copied and the meta model gets better for signals. Readers of the article and code are more prepared to deal with API’s, data series, and signals more generally since article is well written and gives users basic tools not just NASA sattelite specific skills.
Funding:
Some parts of this process take a while to run given the speed of yfinance and other API requests(hours), I will try to fix this but If I cannot maybe sub 1 NMR could be useful for me to rent a server for a long time and automate the retreival of addresses or geocodings of tickers each week for signals users to download via a link(A csv of tickers, addresses, and common coordinates)
This would also take significant time to write, though im not sure if the CoE is interested in compensating for time spent on proposals. If they are maybe a reward for keeping things up to date over a longer period and answer questions, not just after the first release? Tentatively 1-2 NMR might be fair if the example stays up to date, relevant, and starts useful discussions on RC or the forums?