Skip to main content

Check whether project memory helps

With project memory on, a project's tasks are given what earlier tasks learned. Project memory, at the bottom of Audit › Sessions › Efficiency, tells you whether that pays off on one project: it puts the sessions that ran with memory ON beside those that ran with it OFF, over the same days, and shows how often the units given to a task were read.

Who can do this: anyone with a role that can view sessions. Your organization's plan must include both Session transcripts, for Efficiency, and Project memory, for this card. Without the first, Efficiency shows Not on your plan; without the second, the card reads Off — not in your plan.

  1. In the sidebar, select Audit, then the Sessions tab, then Efficiency.
  2. Pick the period at the top right of Where the time goes: 7d, 14d or 30d. The Project memory card at the bottom reads the same period and names it, such as Last 7 days. Pick 30d for a project with few tasks: a side with fewer than 10 sessions shows no figures.
  3. Under Project memory, pick the project in Project. It lists the projects whose sessions you can view. With none, the card reads No project has memory on and says who can turn it on.
  4. Optional: pick a persona in Persona to compare only the tasks that persona ran. All counts every persona.
  5. Read the line above the tables: the dates, then which sessions each side counts. No holdout: split by each session's memory setting means each session is counted by the setting it ran with. OFF: holdout and default tasks · ON: default tasks means Jaah held some tasks back from memory on purpose, so the two sides are a fairer test.
  6. In ON vs OFF, compare the two columns row by row. Sessions and Tasks say how much each side rests on. Fewer Tool calls to first edit (median) and Duplicate reads (median) on the ON side mean runners found their way faster with memory. Units read per task (median) and Units cited per task (median) show whether runners used what they were given. Each median has its spread (IQR) under it. A side that is too thin shows a line such as Not enough ON sessions in this window (n = 4) in place of its figures.
  7. Check the line under the table. Not counted names sessions from before project memory and tasks that ran with memory both on and off; they are left out of both sides.
  8. In ON, split by delivery, compare ON (≥1 injected) with ON (0 injected). If tasks given units do no better than tasks given none, the units are not what makes the difference.
  9. In Read rate per kind, find the kinds of unit tasks are given but rarely use: a low Read rate (the share of Injected units that were Used). Written per read shows how many units of a kind are written for each one read.

If memory does not help a project, turn it off there; see Turn project memory on or off. If one kind of unit is rarely used, read those units and retire the ones that are wrong or stale; see Review or purge a project's memory.