Skip to main content
MVP ⭐️⭐️⭐️⭐️⭐️
Sep 24, 2025 7:54 AM UTC
New

Synthetic timeseries behaving like an ordinary timeseries

Related products:Other

Currently CDF has support for synthetic timeseries, but those are specified as a parameter to the API call, which means that every user has to specify it, and it is a different API call then the ordinary timeseries.

A very powerful feature would be to define a synthetic timeseries the same way as an ordinary timeseries, but it is not a persisted timeseries but instead an on-the-fly calculated timeseries. 

Pros:

  • It doesn’t fill up cdf with a lot of unnecessary persisted timeseries
  • We avoid having to depend on a calculation function that will have lag, being vulnerable to going down, secret expired,+++
  • Any UI or other functionality that consumes timeseries will automatically work
  • It will be a “company” approved calculation
  • We can add proper documentation, metadata, link it to the right assets (contextualisation),+++

Cons:

  • Added computational cost
  • Need to agree on how to treat the “endpoints” in the timerange, how to do the data point alignment, how to do fill etc
  • If at some point the calculation need a different timeseries, or the calculation changes, the synthetic timeseries will not be valid for the historical data (can be handled with a range validity identifier) 

The intention of this is to define relatively simple calculations. More complex calculations will be handled outside this framework.

Example could be: I need a timeseries for whether a well is flowing or not. It is a combination of the position of 3-4 chokes. Not all users will know which chokes, so by creating this predefined synthetic timeseries we have a “approved” way of doing this calculation. 

Another relevant example is routing, which can be a combination of many different routing valves, requiring extensive investigation of P&IDs to figure out how to calculate.

1 reply

Everton Colling
Expert ⭐️⭐️⭐️⭐️
Expert ⭐️⭐️⭐️⭐️
Nov 20, 2026 9:03 AM UTC

Hi ​@Gunnar Andreas!

Sharing an update on this idea: we have started development of a new time series calculation engine that should address it. Our current estimate is a Private Preview for early adopters by May 25, 2027, with the production release planned for the second half of 2027, depending on feedback from the preview.

Gunnar, we'll keep covering this in our regular chats. For anyone else following this idea: if you'd like to share your use cases, reply here and we can set up a call. I can walk you through what we're planning, and your input will help shape the feature.