To prepare for a performance calibration meeting, build a short sheet for each person on your team that holds three things: a proposed rating, two to four specific dated examples that justify it, and a sentence or two of rationale tied to your rubric or leveling guide. That is the whole job. The managers who do well in calibration are not the ones with the best memory or the loudest voice; they are the ones reading from a record. When someone asks “what makes this a ‘meets’ and not an ‘exceeds’?” you want to answer with “Mar 9, the billing migration with zero downtime,” not with a feeling.
Calibration exists because ratings drift. Left alone, one manager’s “exceeds” is another’s “meets,” and the same performance gets scored differently depending on who wrote the review. The meeting puts every manager in a room to compare proposed ratings against a shared bar, so the standard is applied evenly. That is good for your people, but only if you show up able to back each rating with evidence. A rating you cannot defend is a rating that gets moved by whoever in the room can defend theirs.
What to bring for each person
You do not need a deck. You need a one-screen sheet per direct report, kept to facts and observed behavior rather than adjectives. Copy the layout below and fill one out for each person before the meeting.
CALIBRATION PREP - @[Name] ([Level]) PROPOSED RATING: [e.g. Exceeds / Meets / Below] EVIDENCE (dated, specific - spans the whole period) - [Mon DD] - - [Mon DD] - - [Mon DD] - RATIONALE (mapped to the rubric / leveling guide) - Why this rating, in 1-2 sentences: - BOUNDARY CASE? (could land in either rating - and why) - SCOPE & IMPACT vs PEERS AT THIS LEVEL - IF PUSHED HARDER / EASIER, MY STRONGEST POINT -
Here is what each field looks like once it has real content in it. Specific and dated beats confident and vague every single time the room pushes back.
- Proposed rating. The rating you intend to give, stated up front so the conversation has something concrete to work on. “Exceeds, leaning to the top of the band.”
- Evidence, dated and spread across the period. “Mar 9: led the billing migration, zero downtime. May 2: mentored two juniors through the on-call rotation. Sep-Oct: owned the platform cleanup that unblocked three teams.” Three dated lines from across the year beat a paragraph of recent praise.
- Rationale, mapped to the rubric. “At this level the bar is ‘owns ambiguous, cross-team work end to end’; the platform cleanup is exactly that, with measurable impact on three other teams.” Tie the story to the words in your leveling guide, not to how hard the person worked.
- Boundary case, named honestly. “Could read as a strong ‘meets’ if you weight the two missed estimates in Q2. I land on ‘exceeds’ because the scope of the cleanup outweighs them, but I expect pushback here.” Flagging your own close calls makes you credible on the ones you feel certain about.
- Scope and impact versus peers at the same level. “Carrying more cross-team scope than others at this level; comparable execution, broader blast radius.” If your organization uses a distribution, this is the line that defends a person’s place in it.
See how a per-person log turns six months into a five-minute read before calibration.
Cover the whole period, not the last six weeks
The single biggest failure in calibration prep is recency bias: the pull toward whatever happened most recently. The quiet save your strongest engineer made on a Thursday in March is gone by review season, while the one rough sprint in November is still vivid because it just happened. Walk in with evidence clustered in the last six weeks and you will quietly under-rate your steadiest people and over-weight a single recent stumble, and the room may never know, because vague is hard to challenge.
The fix is mechanical, not heroic. Before you set a rating, read your notes for the person in order, oldest to newest, and check that your evidence is spread across the period. If every example you can think of happened after April, that is a signal you are reconstructing from memory rather than from a record. A dated log fixes this on its own: it makes March weigh as much as November, because both are written down with the date attached.
This is also where you catch the unfair version of the opposite problem: a person who had a strong first half and coasted, or a person whose one visible miss is overshadowing two quarters of solid delivery. You can only see the shape of a period when you can read the whole period.
Capture the moment it happens, and read it back dated when calibration comes around.
What actually happens in the room
Knowing the shape of the meeting helps you prep for the right things. Calibration usually runs like this:
- Each manager presents their proposed ratings. You state the rating and the one or two strongest pieces of evidence. The clearer your examples, the less time the group spends second-guessing them.
- The group pressure-tests the edges. Nobody argues about an obvious top-performer or an obvious miss. The time goes to the boundary cases, the people who could land in either of two ratings. That is exactly why you flagged those in your prep: you want to be the one who already thought it through.
- Cross-team comparison. “Is your ‘exceeds’ doing the same work as my ‘exceeds’?” This is where rationale tied to the rubric, and a clear read of scope versus peers, earns or loses a rating.
- Adjustments and finalization. Some ratings move. If yours moves, it should move because the shared bar genuinely placed it differently, not because you ran out of evidence first.
Your job in the room is to advocate for your people honestly with specifics, and to hold a rating you can defend while staying open to a fair adjustment. The two are not in tension. A manager who brings dated evidence and openly flags their close calls is the one the room trusts, and trust is what protects your people’s ratings when the discussion gets tight.
Write the evidence so it holds up
Calibration evidence lives or dies on specificity. The same instinct that makes a fair review makes a defensible calibration sheet: write behavior and impact, not character.
- Dated and concrete beats general praise. “Apr 18: caught a data-loss bug in review that would have shipped” survives a follow-up question. “Great attention to detail” does not.
- Behavior, not labels. Write “missed three of five committed dates in Q3, mostly from late scoping” rather than “unreliable.” The first is something you observed and can rate fairly; the second is a verdict the room cannot check.
- Impact, not effort. Hours spent are not a rating input. “Unblocked three teams” is. Tie every example to an outcome a peer would recognize.
- Lead with wins for your quiet performers. The save that did not announce itself is the one you will fail to defend if it is not written down. Your steadiest people are the ones recency bias hurts most, so their evidence is the evidence to gather first.
Where the examples come from
Every calibration guide says the same thing: bring written, dated examples and a rationale per person. The unspoken hard part is where those examples come from. If you start gathering them the week before calibration, you are doing six months of archaeology in an afternoon, and what you reconstruct under that pressure is exactly the recency-biased, adjective-heavy sheet that gets your people moved in the room.
The examples already exist; they happen in your week. The problem is they are not written down. This is the thing Notivo is built to make effortless: a private, per-person running log you keep as you go. You write a quick note when something happens, whether a save, a miss, or a moment of ownership, and tag the person with @ and their name, and it files into that person’s dated timeline. Add #review or #promo to mark the lines you will want at calibration. When the meeting comes, you type @Sam and read the year in order instead of trying to remember it. An AI assistant can draft a recap of a person’s period from your own notes only; it never makes the rating and never trains on your data; you decide what to bring into the room. You can capture from the web, over WhatsApp, or from the ChatGPT and Claude connectors, so the note happens in the ten seconds you actually have.
To be clear about what Notivo is and is not: it feeds your prep, it does not run calibration. It is a private notebook for your own recall, not a shared HR system, not a system of record, and it makes no employment decisions. It informs your judgment; the rating stays with you and your organization.
On storage, the honest answer: your notes are private by default and scoped to your own account, and they are stored on secured infrastructure. That is platform-level security, the protections that come with reputable cloud storage. It is not a claim that your note content is encrypted end-to-end, and we are not going to pretend otherwise. So use the same judgment you would use with any cloud notes app: this is the right home for your working memory of who did what, and not the place for the most sensitive personal data you would never put in any cloud tool. The one exception is the Black Box, a PIN-locked vault that is end-to-end encrypted, for the rare note that needs that.
Dated examples for every person, gathered as the year happens, not the week before.