In the data sciences, data quality is a prerequisite for success. So the fact that we don’t have a common language for describing the status of our data is a severe constraint. Imagine if, when you are talking about data, you can qualify its readiness. Imagine that you can use a few terms to develop an immediate sense of what investment is required to extract the value from the data.

Data Readiness Levels start with the concept of hearsay data, data which someone heard is available, so they say it is. This is where many projects start. They then provide a scale of readiness which is designed to assist decision makers and practitioners in assessing the project status, which resources need to be deployed, what time lines and costs are likely to be. Data readiness levels are a set of terms used to characterize the readiness of data for deployment.

Data Readiness levels allow managers and decision makers to understand project status, and where investment is needed without intimate familiarity with the data itself.

You can read more in this blog post and this arxiv paper.

FAQ

Q: What are the bands of data readiness?

A: For the moment we have described three bands of data readiness, within each band, there will be sub bands, and one role of this site is to achieve consensus on what those sub-bands (and bands) should represent so that data readiness levels can be deployed across a wide range of applications.

Q: Where does data start?

A: Typically, all data starts at sub-band C4, ‘hearsay data’, meaning that people have heard that the data exists, so they say it does. The ultimate aim, is to bring a data set to A1, at this subband the data is ready for deployment or productionization. To explore each band go to the associated pages on the page above. You can make suggestions for changes to band or sub-band descriptions through github pull requests.

Q: Are these bands set in stone?

A: No, the intention of this site is to share experience and best practice among practitioners of the data sciences. You can contribute changes via pull requests from github (each of the pages has an ‘edit’ button). The merits of those changes can be debated via github issues.

Q: What if we combine data sets? Is the data readiness of the combined data the minimum of the two data sets?

A: No, because there may be ethical, legal or topological reasons why we cannot bring the data together. The DRL of two combined data sets would be at most the minimum of the two DRLs, but could well be lower. Even two data sets at A1, when combined, could require record deduplication, which immediately sends the data back to Band B. Of course such combined data would likely be quicker to pull through the readiness levels than raw data which hadn’t been looked at before. I.e. record deduplication may be the only task required to leave Band B.

Q: Does it make sense to think about data without the context? Can we process data before we know what we want to do with it?

A: Historically in statistics it doesn’t seem to have made sense, and also in the sciences it is frowned upon to collect data on ‘fishing expeditions’. However, in practice, scientific questions are refined as more data is collected. And there is now good precedent for large scale screens of e.g. transcriptome analysis, during development or disease, where the context is very broad.

Q: Are you related to the NSI Nanotechnology discussion?

A: No, we conceived of data readiness levels separately, but after that discussion. Looking at that document, we feel the scope is too narrow for wide usage, the aim in this effort is to ensure that we gain a much broader consensus across data sciences.

Q: Why is this the right time for this new idea?

A: We are trying to reflect the new reality that much data (indeed the majority) is now collected by happenstance, and this data needs very different treatment from data that is actively acquired. Those who are acquiring the data need to be cognisant of the challenges, as well as those who are making decisions about the data and how it will be used in analysis.