Performance Management

The goal: grow every person — up a level or into deeper mastery each year — and build the best team on the planet, one whose floor is higher than most teams’ ceiling. Performance management is how you do that on purpose. It is not a tribunal for failures; catching them is the rare edge of a system whose real work is growth.

Read this and you can run it. The whole page is one loop you run on every person, every quarter:

Know the bar → set the target → reinforce → review → make it actionable → round again.

One clock (quarterly). One baseline cadence (the biweekly 1:1). Three states: under, meeting, and over the bar.

This page is the why and the map. To run a read by hand — what each level means, what to look at, and the eight steps, for any role — see The Method.

Principles

Three convictions. When a mechanic below contradicts one of these, the conviction wins.

1. A team is not a favor. Carrying underperformers taxes everyone who carries them — they lower the standard, absorb the reviews, and teach the rest what gets tolerated. Ten people with two weak links is a worse team than eight excellent ones paid twenty percent more, and the real cost of the two is the raise, the hire, or the focus the eight never got. Be honest early and move.
2. Manage, don’t police. The daily job is growth, not documentation: clarity, a real target, and guidance on how to get there. If I only open this doc to write someone up, I built the wrong thing.
3. The best team on the planet. Superstars who are team players, not lone supermen. Different gifts, one standard: each clears their level’s bar and lifts the people around them.

The three states

Everyone you manage sits in one of three states against their level, and the state sets everything downstream — the cadence, the instrument, the kind of growth.

Under and over are mirror images: both are off-level, both need a cadence change and an instrument pointed at a named outcome. Meeting is the majority, and the one that gets neglected. The five moves below are the same for all three; only the content changes.

1 · Know the bar

You cannot manage a gap the person cannot see. The first move is that they know what good looks like at their level, in their seat — before any target, review, or instrument.

Read the seat. What “good” means changes with the role. Read each against its published rubric, not a generic bar.

• Engineer — the hardest technical decision, and whether they are stuck on ambiguity or on execution. Good is a shipped capability at a quality bar. Rubric: Product Engineer.
• PM — the user problem, the one metric that matters, and the tradeoff they are making right now. Good is an outcome moved, not a document produced. Rubric: Product Manager.
• Marketing / other — the audience, the message, and the number they move. Good is a measurable result tied to a channel. No published rubric yet, so hold the read as lower-confidence and lean harder on dated outcomes.

Make the bar theirs. The rubric is the standard; your job is that they can say back what clearing it looks like this quarter. Someone who does not know the bar has a clarity gap — and that is on you, not them.

2 · Set the target

Expectations without a target are a mood. A target is four things: a deliverable, a number, a date, and a reason it matters. Miss one and it is a wish. The test is whether they can say it back without notes.

And the deliverable is an outcome, never an activity. “Review three PRs a week” is activity — it can be done in full and change nothing. “No PR you approve is bounced by QA” is the outcome that activity is meant to buy. Grade the outcome; the activity is theirs to figure out.

• For someone who can find the path: set the outcome, leave the method. Handing a capable person the method is a leak.
• For someone still building the skill: set the next step. Handing them only the outcome is abandonment.

Point it at the next rung, or at deeper mastery in the seat when the next rung is not where their leverage is — a specialist can be worth more deep than promoted. Not everyone should climb the same ladder.

3 · Reinforce

The target is set in a meeting and won in the weeks between reviews. Reinforcement is the daily job, and it is most of the job — principle two.

Same-day feedback. When you see something worth naming, name it that day. Delayed feedback loses the context the person needs to act on it. No hedging: the behaviour, the impact, and what you want more or less of.

The dose. Cadence scales with the gap, and never more than the state needs.

• Floor: biweekly with everyone. Less than that and you are absent.
• Escalate by exception, for a period. Weekly for a real gap or a live PIP; daily 15-min when mentoring someone out of a ditch. Only as far as the state demands, only until it stabilizes.
• Too much reads as distrust and builds the dependency you are trying to remove. Too little lets the middle coast and a star drift. A right dose tapers itself as the person stabilizes; if it never steps down, it is the wrong instrument or the wrong seat.
• Most time goes up, not down. The people meeting and exceeding get the bulk of your attention; the one failing gets a dated decision — close the gap on a set timeline or move — not open-ended attention (principle one).
• Scale to the context. A full-time team on real comp gets the full machinery. A part-time apprenticeship on a fixed runway (see Internship) gets the light version: the buddy cadence and the transition checkpoints are the review, and the formal scoring is skipped.

What reinforcing looks like, by state:

• Under — diagnose the gap, then coach it. Clarity: make the target concrete. Skill: teach, pair, or shrink the step. Blocker: clear it, often the one thing only you can do. Will: a stakes conversation, not more help.
• Meeting (the real job) — bring a growth target to every biweekly, not a status check. Hand them a harder problem or a surface they have not owned before the current work stops teaching them. Watch the trend both ways: catch a slide early, catch readiness early.
• Over — say it plainly, then act; a star who hears nothing starts looking. Hand them leverage: a domain to own, people to lift. Size the next problem so they cannot clear it on existing skill.

4 · Review

Quarterly, you make the read formal: where they stand, and what it means. It is never a surprise — the reinforcement has been saying it out loud all along.

The read — two questions, in order.

1. Where are they against their level? below, at, or above.
2. If below, why? The four gaps — clarity (does not know what good looks like), skill (knows, cannot hit it yet), blocker (something outside them), will (knows, can, nothing blocks them, still not driving).

Get the real state, not the report. The truth sits a layer under the update. Do not ask how it is going; ask what is hard, what would make it slip, and what they would do if you were not in the room. When someone is fluent about activity and vague about the outcome, the outcome is in trouble — follow the vague part.

The rollup. The rubric’s dimensions collapse to one state by the floor rule: the level is the floor of the load-bearing dimensions for the seat, not the average and not the peak. A lagging dimension that matters holds them below even with a spike elsewhere; a spike above the bar does not raise the level, it earns scope. That is how you avoid both averaging a spiky person down and letting one strength launder the gaps.

Score it — two tiers, and most people only need the first.

The everyday read (everyone). Mark each load-bearing dimension against the bar for their level, with a dated example, and take the floor:

The state is the floor of the load-bearing rows (the rule from The read, above). Specific and defensible is the bar, not exhaustive — do not build a 100-point instrument for someone who is fine. Three cases are worked below.

The formal instrument (a PIP only). When the case is perform-or-exit, tighten it into something gradable weekly: each concern becomes a weighted target with a measurable proxy, scored on a rubric where every band is a concrete behaviour, not a number.

Weekly checkpoints, rising bar: Day 15 ≥ 60, Day 30 ≥ 70, Day 45 ≥ 80 — scoring a trend, not a spike. /pip builds the full instrument; /performance-review, the everyday one.

No matrix required, but the matrix-free read earns the same rigor: pull two or three dated examples on the load-bearing dimensions before you name the state.

Evidence, not a vibe.

• Anchor every claim to a named dimension and a dated example. “They tend to…” is a bias tell; a dated example is not.
• Read the whole period, not the last two weeks. Score each dimension on its own evidence — one spike or one miss does not color the rest. Read output, not visibility; the loud and the nearby read as higher than they are.
• Separate the person from the conditions: a miss under a broken process is not a performance gap. Separate one bad quarter from a trend — which is why the read carries a direction, not only a position.
• One level below title is mis-leveling — reset the title. Two levels is a bad hire or a bad promotion: name which, because the fix differs (a perform-or-exit PIP with an honest floor, or a two-step re-level). A functioning Junior titled Mid is mis-leveled, not failing.
• Single source is the largest bias of all — your read is one observer. Weigh in a second reader or a peer signal, hardest on the peer-facing dimensions (see Peer review, below). Never hold a team-wide gap against one person; with more than one manager, calibrate the bar together.

5 · Make it actionable

A review that ends in a description is wasted. It has to end in moves. This is where the read becomes next quarter, and the loop rounds back to move two.

Pick the instrument. Below the bar is not automatically a PIP — diagnose first, then pick.

• Mentor — clarity, skill, or blocker. Raise the cadence, close the specific gap, and say plainly this is not performance management.
• Re-level — mis-leveling: performing fine at their true level, the title just set a notch too high. Reset title and level with a development plan. The default for mis-leveling, and lower-risk than a PIP when the gap is genuinely leveling — but a demotion carries its own morale and legal weight, so name it honestly and never dress an exit as a re-level.
• PIP — will, or genuine perform-or-exit. Strictly “perform at this bar, or leave.” A mis-leveling case is a PIP only if you would not keep them a level down. Name the honest floor: clearing the lower bar re-levels, it does not by itself mean exit. Loop in People early, and start the written record on day one. /pip builds the document.
• Promote / scope-up — above the bar. Build the case upward the way a PIP is built downward: dated evidence mapped to the next level’s bar. No seat yet? Expand scope ahead of the title, or graduate them well where the model is grow-and-release. Retain where you can; launch cleanly where you cannot.

Set the next target — the four parts again, pointed by the read. This is the handoff back to move two.

Give direction they can carry — two sentences: where they need to be by end of quarter, and why it matters for the team. If they cannot say it back without notes, it is not set.

Leave nothing loose. Every finding becomes a move with an owner and a date, yours or theirs. A review that produces “let’s keep an eye on it” produced nothing (see How I Run on not carrying monkeys).

Doing it by hand

Everything above is the why and the map. To actually produce a read to this standard — what each level means, what to look at for each role, the eight steps, the pressure-test checklist, and worked cases across engineering, product, and marketing — follow The Method. It is the manual procedure /pip and /performance-review run, written out so a person can do it without AI.

Peer review

The manager’s read is one set of eyes, and the weakest on how someone shows up to peers. Peer review is the structural fix: the only honest read on whether they are reliable, whether their word holds, and whether they lift the room — the Leverage, Influence, and Communication dimensions a manager sees worst.

Twice a year, for full-time teams. A quarter is too short to accumulate honest peer signal, and 360s fatigue fast when run more often. Six-monthly pairs with the comp and promotion rhythm: the half-year review is fed by a peer round, the quarterly ones are the manager’s.

Short and behavioural — a few prompts, not a rating grid. Numeric peer scores get gamed and averaged into mush. Four questions carry it:

• Start / stop / continue — one thing to keep, one to start, one to stop.
• Superpower and edge — the biggest strength, and the one change that would most raise their impact.
• Reliability — when you depend on them, does it hold? A moment, not a rating.
• Would you want them on your most important project again? The regretted-loss signal.

It informs, it does not decide. Peers write freely; the manager synthesises and delivers. The read stays the manager’s, and peer input weights it, hardest on the dimensions a manager cannot see.

Apprentices skip the formal round. Part-timers with other jobs do not overlap enough for honest 360 signal, and the buddy — who pairs daily and reviews every PR — is already the peer read. At most, pull the same four questions from the buddy and a teammate, informally.

The review doc

One per person per quarter, capturing all five moves. Lean for a healthy performer; a full evidence appendix only for a formal case (re-level or PIP handoff). /performance-review drafts it — role-aware (pulls the Product Engineer or Product Manager rubric to set the bar), matrix optional, pressure-tested.

# Performance Review — [name]
**Title / level:** [Software Engineer, Mid]   **Period:** [quarter]   **Manager:** Andro

## The read (where they stand)
- Level vs title: [below / at / above] — [one sentence]
- Trend: [closing / flat / climbing]
- Standing per load-bearing dimension (below / at / above + dated example):
- Matrix (if any): Jr __% · Mid __% · Sr __% → [level]

## Diagnosis (only if under)
- Gap: [clarity / skill / blocker / will / mis-leveling] — [dated evidence]

## Strengths (what they clearly clear — keeps the case honest)

## Instrument + cadence
- [ ] Mentor   [ ] Re-level   [ ] PIP   [ ] Direction only   [ ] Promote / scope-up
- Cadence next period: [biweekly / weekly / daily 15-min]
- Honest floor (if re-level or PIP):

## Next target — deliverable · number · date · reason

## Direction next quarter (two sentences: where, and why it matters)

## Next actions (every finding → owner · date)

## Said back without notes? [yes / not yet]

Common pitfalls

• Reward is only money. Comp is one lever, and a blunt one. The rewards that compound are bigger: the right seat propels someone’s knowledge and expertise in ways a raise never can — hard problems, a team that stretches them, real ownership. Pay fairly, then compete on growth, not on cash.
• Neglecting the middle. The meeting-the-bar majority triggers no instrument, so it gets no investment, and quietly drifts or leaves. Give them a target and a rung every quarter.
• Grading activity, not outcome. A full sprint board is not a shipped result. Score what moved, not what kept someone busy.
• Averaging the radar. Marking a spiky, excellent person as “meets” because one dimension lags. Different gifts are the point: a deep spike who plays for the team beats a well-rounded average.
• Hoarding the star. Keeping an over-the-bar person where they are because they are too useful to move. That is how you lose them.

What this is not

• Not a paper trail. PIP and re-level are the rare ends. The default is direction and coaching.
• Not a substitute for same-day feedback. Feedback is same-day, always. The review sets the arc; the reinforcement runs it.
• Not a surprise on the read. If where they stand is news, I failed the reinforcement. The consequence can still land hard — that is honesty, not ambush — but never the read itself.

Related: The Method · How I Run