Reviews, hiring & learning

Running a review cycle, goals, assessments, learning paths

Three modules about people rather than work: how somebody is doing, who you hire next, and what they learn along the way.

For how they are built — models, AI usage, external integrations — see People architecture.

Performance reviews#

A review cycle, its period and its three deadlines

A cycle is a review period with three deadlines, one per phase, and every person in scope gets their own review inside it.

Each review moves through the phases in order:

  1. Self review. The person reflects on the period against structured prompts.
  2. Peer review. Colleagues give 360° feedback.
  3. Manager review. The manager synthesises it into ratings and recommendations.
  4. Acknowledged. The person has read it.

A review does not advance because a date passed — it advances because the phase before it was submitted. The deadlines are what people are held to, not what the system enforces.

Who reviews whom#

Peer reviewers are chosen in one of three ways, set on the cycle: the subject picks their own, the manager assigns them, or either is allowed. There is no right answer, but there is a wrong one — changing the mode mid-cycle leaves half the reviews assigned one way and half the other.

Two things to tell people honestly#

Anonymous is not untraceable. Peer reviews are shown without a name, but timing and volume can narrow a small team down to one person. If anonymity genuinely matters, wait until several reviews are in before sharing them with the subject.

The AI summary is written once, when the review completes. A manager who edits their submission afterwards does not get a new summary automatically.

Goals#

Goals

What somebody is working towards, tracked between review cycles rather than invented during one. A goal carries its own progress, and a review can point at it — which is what stops the self-review phase becoming an exercise in remembering July.

Hiring#

Requisitions, candidates, and assessments — including automated ones with a trust score on each attempt.

That score is advisory. A low one flags the attempt for a human to look at; it does not fail the candidate and should not gate an offer on its own. Anybody using it as a threshold has misunderstood what it measures.

Hiring connects back to Organization: an open position on a department is a real row, and a requisition can be opened against it, so headcount planning and hiring are looking at the same number.

Learning#

Skills, career roles and learning paths — a path is generated from the gap between somebody's skills and the ones their target role requires.

Two consequences worth knowing:

  • A role's required skills are the target. Re-weight them and every active path derived from that role is stale until it is regenerated. Nothing tells you; the paths simply describe an older version of the job.
  • External course progress needs a connected integration. Naming a course as coming from an external provider does not fetch anybody's progress from it. Without the integration, completions are self-reported, which is a perfectly good answer as long as everybody knows that is what they are looking at.

Common mistakes#

  • Treating deadlines as gates. They are dates in a plan; the phases advance on submission.
  • Promising anonymity a small team cannot have.
  • Reading a trust score as a verdict.
  • Re-weighting a career role and leaving the paths alone.
  • Expecting an edited manager review to regenerate its summary.