A plain-language walkthrough of review cycles, roles, automation, and how feedback turns into a plan people actually follow up on — not just another report nobody reopens.
review.center has a small vocabulary. Once these five ideas click, the rest of the product is just screens for managing them.
Your organization's private workspace. Nobody outside your company can see your people, reviews, or feedback — and you can't see anyone else's.
The competency template you review people against — a set of Categories (e.g. Communication, Delivery), each broken into specific Items that reviewers actually rate.
A dated window — "Q1 2026", say — in which feedback gets collected and reviews get compiled. Everything below happens inside a cycle.
One reviewer's input on one person, for one cycle. If three people review the same teammate, that's three separate pieces of Feedback.
The compiled, final review for one person in one cycle — built from all of their Feedback, by a manager or, at higher automation levels, automatically.
Not job titles — assignments. "Reviewer" just means whoever is currently asked to give feedback on someone this cycle. Any employee, manager, or admin can be assigned to review anyone else.
Four account levels. They control what you can manage — being asked to review a colleague isn't tied to any of them.
| Role | What they can do |
|---|---|
| Employee | See their own profile, reviews, and improvement points. Give feedback on anyone they're assigned to review, regardless of role. |
| Line Manager | Everything an Employee can, plus: view and manage direct reports' profiles, run review cycles they own, compile reviews, approve or reject feedback, and see team performance & improvement-point rollups. |
| Admin | Everything within their company: manage Work Profiles, create and manage every review cycle, and add or edit employees. |
| System Administrator | A platform-operator level, not a normal company account — provisions companies and handles platform-wide maintenance. |
This is the one thing worth understanding well before you create your first cycle: dates drive everything. There is no "start now" or "close now" button.
Nobody closes a cycle by hand, day to day — not even an Admin. Treat your end date as a real deadline: it's what actually triggers submission, not a label.
Set per cycle, from fully manual to fully automatic. You can raise the level once you trust what the earlier ones produce.
| Level | Name | What actually happens |
|---|---|---|
| 0 | Classic | Nothing automatic. Reviewers rate items and write feedback by hand; a manager reads all of it and compiles the review by hand. |
| 1 | Light | The engine uses LLM models to write a short summary of each piece of feedback for whoever compiles the review. Ratings stay entirely manual. |
| 2 | Enhanced | Same summary, plus the engine uses LLM models to suggest per-item ratings — but only where your cycle isn't already set to take ratings straight from reviewers; a reviewer's own number always wins when both exist. A manager can still overwrite anything before compiling. |
| 3 | Full Auto | Once every assigned reviewer's feedback for a person is in, the review compiles itself — the engine uses LLM models to write the review text, so no manager author is needed. Ratings don't use LLMs: they're averaged across that person's feedback, and a rating a reviewer never gave defaults to the lowest score, so a skipped rating can't quietly pass as a good one. It still doesn't submit on its own: it waits in a manager's compile requests for a one-click confirm before it's final. Only turn this on once you trust the process feeding it. |
Every level up is time you get back as a manager — less reading, less writing, less chasing.
Full Auto compiles as soon as feedback is complete. If your cycle also requires feedback approval, an already-approved feedback's engine-suggested rating isn't re-checked after approval — make sure your approvers are acting on accurate ratings before they sign off.
Collecting feedback is only half the job. This is the part that actually changes what happens next cycle — generated and followed up on automatically by review.center, not just documented once and forgotten.
Nobody drafts these by hand. At the end of every submitted review (automation level 1 and up), review.center automatically reads that cycle's feedback together with the person's rating trend across past cycles, and the engine uses LLM models to write a short list of Improvement Points on its own — specific, named things for that person to work on next, not a paragraph a manager has to sit down and compose.
The engine uses large language models (LLMs) in a handful of specific places, and nowhere else. Look for the LLM badge throughout this manual. This section is the complete list.
The engine drafts, people decide. It calls two outside model services: Claude (by Anthropic) reads and writes text, and Voyage (an embedding model) turns the wording of Improvement Points into numbers so similar ones can be grouped. What gets sent depends on your cycle's automation level — a Classic cycle sends nothing automatically.
| What you see | What the engine does with an LLM | When it runs |
|---|---|---|
| Feedback summary | Condenses one reviewer's answers into a short summary (about 500 characters) for whoever compiles the review. | When feedback is submitted · Light and up |
| Suggested ratings | Reads the tone of the written feedback and suggests a score per item. A reviewer's own number always wins, and you can change any of them. | When feedback is submitted · Enhanced and up, and only if the cycle isn't set to take ratings straight from reviewers |
| “Generate summary” button | Drafts the text of one review field from all the feedback that is in. Nothing is saved until you save it, so edit freely. | When a manager clicks it while compiling · any level |
| Full Auto review text | Writes the text of every review field from the feedback. The ratings are plain averages, with no LLM involved. It then waits for a manager's one-click confirm if your cycle requires it. | When the last feedback comes in · Full Auto |
| Improvement Points | Reads the cycle's feedback and the person's rating trend, then proposes one to four specific things to work on (or keep doing). It also checks each suggestion against the person's existing points so you don't get duplicates. | When a review is submitted · Light and up |
| Progress on older points | For a point that isn't tied to a rating category, the engine asks an LLM to assess from the new feedback whether it was Matched, Partially matched, Not matched or Regressed. | When a review is submitted · Light and up |
| Recurring themes | Spots up to five topics several reviewers raised in a cycle, marks each as a strength or something to improve, and links it to the matching Improvement Points. | When a review is submitted · Light and up |
| Follow-up cycle wording (Continuity) | Writes the suggested cycle's name, a short reason it's worth running, and the question reviewers answer for each focus point. Who is reviewed, on which points and by whom is decided by fixed rules, not an LLM. | When a follow-up suggestion is drafted · Continuity add-on |
| Grouping similar points | The engine uses the Voyage embedding model to turn each point's wording into a numeric fingerprint. That powers the 3D team sphere on the dashboard and the themed follow-up cycles that gather people sharing one problem. | When a new Improvement Point is created |
An Improvement Point that ends a cycle unmet shouldn't just sit on a profile until someone happens to look. Continuity drafts the follow-up cycle that re-checks it — who to review, on which points, by whom, and when — for a manager to accept, edit or dismiss. Those choices follow fixed rules; only the cycle's name, the reason shown to the manager and the reviewers' questions are written by the engine using LLM models, and you can reword any of them.
Continuity is an add-on at Light automation, priced per user, and included from Enhanced up. A company admin turns it on from the Subscription page. Without it, managers still see a count of the unresolved points it would have followed up on.
People change employers. When someone joins your company from another review.center customer, their verified performance history can come with them — so you're not starting a new hire's profile from a blank page.
This only ever happens with the person's own consent, the receiving company's agreement, and a sign-off from the review.center team — never automatically, and never as a side effect of anything else.
For a brand-new company or admin. Six steps, roughly in order.
Cycles open automatically overnight on their start date. If today is the start date, it opens just after midnight — not the instant the date arrives.
Not as a manager or admin, day to day. Cycles close automatically once every review inside them is submitted and the end date has passed — there's no manual "close now" action for a company account.
Something in it isn't finished — usually an unsubmitted feedback, an unsubmitted review, or feedback still waiting on a manager's approval. The cycle's manager gets a notification naming exactly what's blocking it.
A review can only be submitted once every reviewer's feedback is in — submitted and, if your cycle requires it, approved. You can start compiling (and save drafts) once at least half of it is in. The compile page's Outstanding feedback panel shows what's missing and who it's waiting on. If you're the approver, approve it right there; otherwise use Nudge to remind the reviewer or approver.
The manager compiling that person's review is waiting on you — either to finish your feedback, or to approve someone else's. Nudged requests move to the front of your list, and the tag goes away as soon as you submit or approve.
Only on Full Auto, and only for suggestions drafted on schedule: those start on their own three days after they appear unless someone dismisses them, and the manager is emailed when they're ready. Anything a manager asked for waits for them. A company admin can turn the autopilot off on the Subscription page. On every other plan, nothing is created until a manager accepts it. If the autopilot can't start one (say the concurrent-cycle limit is reached), it stays waiting with the reason shown, and doesn't retry by itself.
Yes, on Light and above. Open a closed cycle, look at the quote for its follow-up and ask for it; a company admin then buys it, or requests the full add-on instead. It's priced per person followed up, and you're billed only for the people the draft actually includes.
The normal feedback form plus a "Follow-up focus" step per focus point: a question and a Yes / Partly / No / Worse / Can't say choice, with an optional comment. They never see the improvement point itself. Their answers are what show whether the point was resolved.
Yes. An accepted follow-up is an ordinary review cycle. When your plan is already at its limit, Continuity pushes the suggested start back until a slot frees up and says so; adding a concurrent slot to the plan lets it run sooner. A follow-up that joins an already-scheduled cycle doesn't use an extra slot.
Just you and your company's admins — not the person you reviewed, and not automatically their manager either. What that person and their manager eventually see is the compiled Review, not your raw feedback line by line.
At automation level 3, you're not needed to write that particular review — the system compiles it once all the feedback is in. It's assigned to you (or whoever would've compiled it at a lower level) as a "Confirm & submit" request on the compile page, so it still doesn't go out until you give it that one click.
Yes: if you're assigned as a reviewer for a colleague, you give feedback on them too, the same as anyone else. Reviewing someone isn't limited by role.
In nine specific places, all listed in section 06. Text is sent to Anthropic's Claude (summaries, ratings, points, themes, follow-up wording) and, for grouping similar points, Voyage (point wording only, with names removed). Under Anthropic's commercial API terms, what's sent to Claude is not used to train their models. Voyage's default terms allow training on what it receives; review.center has opted out of that. review.center doesn't train anything on it either. A Classic cycle sends nothing automatically.
Never entirely. At Full Auto the engine uses LLM models to write the review text, and ratings are averaged, from what reviewers already submitted, not inventing an opinion of its own — but even then a manager still has to confirm it with one click before it submits. At every other automation level, a manager reviews and can change anything before it's saved.
Nothing, unless they (or their new company) start a transfer request and it's approved by the review.center team — see section 08. If that happens, they lose access to your company once the move completes, and their new employer sees a summary of their rating trend and Improvement Point history, not your company's raw feedback or reviews. Their open work is wound down for you: feedback they still owed is withdrawn, and the reviews, cycles and approvals they held are handed on, mostly to their manager.