can Plot select a window's beginning and ending data?

[tx0c]

I'm doing a graph like this https://observablehq.com/d/2d5429036206e1ae

the data is a monthly series data, to show a website's active users, as the column month num_users implies; also because the site's getting new users every month, and lost some as well, the num_users_m3a is showing how many matured users (age >= 3months) are still active;

  0: Object {month: 2021-12-01, num_users: 10957, num_users_m3a: 4684}
  1: Object {month: 2021-11-01, num_users: 10529, num_users_m3a: 3974}
  2: Object {month: 2021-10-01, num_users: 10357, num_users_m3a: 5638}
  3: Object {month: 2021-09-01, num_users: 10226, num_users_m3a: 5554}
  4: Object {month: 2021-08-01, num_users: 9936, num_users_m3a: 5247}

here want to add a link between each 3 months pair, something like this but this does not work,

      Plot.link(stats, {
        x1: d => d3.utcMonth(d.month, -3), x2: "month",
        y1: "num_users", // here actually want to select the prior 3months' num_users data
        y2: "num_users",
        stroke: "red",
      })

I tried Plot.window() but cannot find a proper reduce function, I guess the end result would be some sort of Plot.select({ d1: the-begin, d2: the-end }, Plot.window({ k: 4, reduce: /* to output the window array of data instead of any current function mean/median/... */ })), the link will create a line between month: 2021-08-01, num_users: 9936 toward month: 2021-11-01, num_users_m3a: 3974

  1: Object {month: 2021-11-01, num_users: 10529, num_users_m3a: 3974}
  4: Object {month: 2021-08-01, num_users: 9936, num_users_m3a: 5247}

[Fil]

you could try and use

Plot.link(stats, Plot.map({
        y1: Plot.window({ k: 4, anchor: "end", reduce: d => d[0] }),
        x1: Plot.window({ k: 4, anchor: "end", reduce: d => d[0] }),
      }, {
        x1: "month",
        x2: "month",
        y1: "num_users_m3a",
        y2: "num_users",
        stroke: "red",
      }))

[Fil]

or better with your midMonth helper:

Plot.link(stats, Plot.map({
        x1: Plot.window({ k: 4, anchor: "end", reduce: d => midMonth(d[0]) }),
        x2: data => data.map(midMonth),
        y1: Plot.window({ k: 4, anchor: "end", reduce: d => d[0] })
      }, {
        x1: "month",
        x2: "month",
        y1: "num_users_m3a",
        y2: "num_users",
        stroke: "red",
      }))

[mbostock]

If you aren’t doing any fancy grouping or faceting, and your data is in chronological order, you can offset the values using array.slice like so:

Plot.link(stats, {
  x1: stats.map(d => midMonth(d.month)).slice(3),
  y1: stats.map(d => d.num_users).slice(3),
  x2: d => midMonth(d.month),
  y2: d => d.num_users_m3a,
  stroke: "red",
})

[tx0c]

Plot.link(stats, Plot.map({

ok, it is that Plot.map doing the heavy lifting work which is hard to understand and I overlooked