The range over which offsets are spread differ between encoding channels and mark configs.

[joelostblom]

If I set the offset via an encoding channel, points are by default spread over the entire band:

Open the Chart in the Vega Editor

    ...
    "yOffset": {"field": "random", "type": "quantitative"},
  ...
  "transform": [{"calculate": "random()", "as": "random"}],

However, if I do the same thing in the mark config, the points are not at all spread as widely:

Open the Chart in the Vega Editor

  "mark": {"type": "point", "yOffset": {"expr": "random()"}},

If I scale the spread by a constant (e.g. random() * 15 as in #9436), the encoding channel will dynamically rescale its range and the chart looks the same as before. However, doing the same in the mark config will increase the spread of the points until they overlap with points of neighboring categories.

It's somewhat confusing that these work differently and it would be helpful if their behavior could be standardized.

---

The goal with #9436, #9437, and this issue is to simplify the usage of jittering, which I think is an important and often used functionality, especially for statistical and scientific charts. Ideally, I would like to be able to use the most convenient syntax:

"mark": {"type": "point", "yOffset": {"expr": "random()"}},

and end up with an effective chart by default:

cc @kanitw as per the previous issues I just opened.