Where we are
Every session so far handed you one component. b1 gave the three purposes and the fair-metric checklist, b2 the OKRs, b3 the fairness scoring and the Goodhart lab, b4 the metric tree per function, b5 the cadence and dashboards, b6 the separated conversations, b7 the bias controls, b8 the AI guardrails, b9 the repair kit. Assembled badly, those parts still produce a system people manage instead of work. So the capstone is mostly about the joins. Two things decide whether it holds: every layer has to point at the same goal, and every layer has to survive someone trying to win rather than contribute. That second test is the one teams skip, so we run it explicitly today. And the linchpin one final time: the OKR score is for steering and learning, never a direct input to pay.
Seven layers, each answering one question 7 min live
If a layer is missing, something else silently takes its job - and usually badly. Skip the guardrails and the OKR eats quality. Skip the review of the measures and last year's proxy is still steering you. Read the stack as a set of questions, and any gap becomes obvious.
LiveWhat happens when a layer is missing3 min▶
Each gap has a predictable symptom, which makes diagnosis fast:
- No goal - the OKRs read as a task list and nobody can say why a key result matters. Reprioritising becomes an argument about preference.
- No owned lever - the team is graded on a shared outcome, so credit and blame are both arbitrary and the metric is gameable by whoever reports it.
- No guardrails - the target gets hit and something unmeasured pays for it. This is the b9 gaming pattern, pre-installed.
- No cadence - problems surface at quarter end, when it is too late to steer. The system becomes a scoring ritual.
- No separated conversations - the number starts driving pay in practice even if the policy says otherwise, and honest reporting stops.
- No review of the measures - proxies drift, surrogation sets in, and you spend a year steering by a number that stopped tracking the goal.
Work upward when you diagnose: most complaints about "our OKRs are useless" trace to layer 1 or layer 3, not to the OKR wording.
Self-studySizing the system: how much is enough2 min read▶
Capstone temptation is to build something impressive. Resist it - an over-built system fails by being ignored, which is indistinguishable from having no system.
- One objective per team per quarter, two or three key results. More than that and nothing is a priority.
- Two or three guardrails, watched not graded. A guardrail list of ten is a scorecard by another name.
- One weekly steering number per team that everyone can name from memory. If people have to open a dashboard to recall it, it is not steering anything.
- One page total. If the system does not fit on a page, the team will not hold it in their heads, and a system nobody holds is a document.
The quarterly review is where you add things - and, more usefully, remove them. Most mature systems get smaller over time, not bigger.
Six gates before you ship it 5 min live
The layers make the system work. These gates decide whether it is fair - and any one of them failing is enough to turn the whole thing into a number-management exercise.
LiveRed-teaming: attack your own system before it ships4 min▶
The gates are a checklist, and checklists reward good intentions. Red-teaming does not. Take three hostile roles in turn and try to break your own design:
- The optimiser. "How do I hit every number without doing the work?" If you find a route in under two minutes, so will someone under pressure. Their answer tells you which guardrail is missing.
- The employee on the receiving end. "Which of these could punish me for something I do not control? Where would I feel measured rather than supported?" This finds controllability failures and surveillance creep faster than any audit.
- The regulator or auditor. "Show me the criteria, the evidence, the human decision-maker, and the contest route." If any answer is verbal only, the system is not documented enough to be defended.
Add a fourth if you can: the newcomer who asks what each number is for. Anything you cannot explain in one sentence is a candidate for removal.
Self-studyRolling it out without breaking trust2 min read▶
A fair system introduced badly still lands as surveillance. Sequence matters as much as design:
- Say what it is not for, first. People assume any new measurement is about pay or cuts. Address that in the opening sentence, and put it in writing so it survives the retelling.
- Run one quarter with no consequences. Score it, learn from it, change what was wrong. A system that visibly changed after feedback earns more trust than a perfect one imposed.
- Show your own numbers first. Managers who publish their own OKRs and scores before asking the team to make the whole thing feel like steering rather than inspection.
- Name what you will stop doing. If nothing is retired, the system reads as extra work, and it will be resisted for good reason.
Trust is a component of the system, not a side effect. Where it is thin, measurement gets interpreted at its worst reading, and honest reporting is the first thing you lose.
Build the system for Cartwheel Growth ★ 14 min · everyone tries
One team, one quarter, all seven layers. Cartwheel's Growth team owns acquisition and the top of the funnel. Build it in order - and notice that the guardrails are written before the cadence, because they constrain what the target may cost.
Layer 1, the goal: profitable growth, not traffic. State it as an outcome: more paying customers acquired at a sustainable cost.
Layer 2, the OKRs: one objective, three key results, each a measurable outcome. Mark which are committed and which are stretch.
Layer 3, the owned lever: from the metric tree, Growth owns sessions and new-customer conversion - not order value, which Merchandising owns.
Layer 4, the guardrails: what must not be sacrificed to hit those key results. Cost per acquisition, retention, and refund rate.
Layers 5 and 6, cadence and conversations: one weekly steering number, monthly trend review, quarterly scoring; 1:1s and CFRs on their own track, no pay talk.
Layer 7, the review: at quarter end, audit the measures with b9's five questions before setting the next OKRs.
| Layer | Content | Owner / rhythm |
|---|---|---|
| 1 · Goal | Acquire more paying customers at a sustainable cost | Growth lead |
| 2 · Objective | Make first-time buying at Cartwheel the easy choice | Quarter |
| 2 · KR1 (committed) | New-customer conversion rate from 1.8% to 2.2% | Growth + Product |
| 2 · KR2 (committed) | Qualified sessions from 640k to 760k per month | Growth |
| 2 · KR3 (stretch) | First-order refund rate stays at or below 4% | Growth + CX |
| 3 · Owned lever | Sessions x new-customer conversion. Order value is not theirs | Growth |
| 4 · Guardrail | Cost per acquisition not above the agreed ceiling | Watched weekly |
| 4 · Guardrail | 30-day retention of new cohorts does not fall | Watched monthly |
| 5 · Steering number | Weekly new-customer conversion, with target and trend | Weekly, 15 min |
| 5 · Scoring | 0 to 1 per KR, ~0.7 is healthy for stretch, learning not verdict | Quarterly |
| 6 · Conversations | Weekly 1:1 steering, quarterly growth check-in, pay elsewhere | Separate rooms |
| 7 · Measure review | b9's five questions before the next OKR set | Quarterly |
Red-team the system you just built ★ 8 min · do your own
Now attack it. Three hostile roles, one pass each, and a fix for every hole. A system that has not been attacked has only been proofread.
As the optimiser: find the cheapest way to hit each key result without doing the work. Write the route down, then close it.
As the employee: find where you could be punished for something outside your control, or where you would feel watched rather than supported.
As the auditor: ask for the criteria, the evidence, the human decision-maker, and the contest route. Note anything that exists only verbally.
Fix each hole at the layer that caused it - guardrail, ownership, documentation - and re-run the six gates.
| Role | Attack found | Layer at fault | Fix |
|---|---|---|---|
| Optimiser | Buy cheap low-intent traffic to inflate sessions | 4 · guardrails | Define "qualified" sessions; hold CPA and retention as guardrails |
| Optimiser | Heavy first-order discounting to lift conversion | 2 · KR design | KR3 refund and retention guardrail already blocks it - keep it graded |
| Employee | Conversion drops because checkout broke - not Growth's doing | 3 · ownership | Shared outcome noted; grade the lever, record the context at scoring |
| Employee | Weekly number feels like inspection | 6 · conversations | Steering only in the 1:1; manager publishes their own numbers too |
| Auditor | Scoring rules exist only in the lead's head | 1 and 7 · written | Write the criteria and scoring guide; state what scores are not used for |
| Auditor | No contest route if someone disputes a score | Gate 5 | Name who to raise it with, what they see, and the response time |
The system that survived its first bad quarter. The real test of a measurement system is not the quarter it succeeds - it is the first quarter a team misses badly. If the design is right, that quarter produces a useful conversation: which assumption was wrong, which guardrail held, what the team learned, what the next OKR should be. If the design is wrong, it produces defensiveness, rounded-up scores, and a story about external factors. Same miss, two completely different outcomes, and the difference was decided months earlier by whether the score was wired to pay. That is the whole course in one sentence: build the system so that missing a target is safe to say out loud.
Build your own - this quarter ◐ 60-90 min total
- Write your team's seven layers on one page. If it does not fit, cut until it does.
- Run the six gates against it and be honest about gate 1: is any part of this score reaching a pay or rating decision in practice, whatever the policy says?
- Red-team it with all three roles. Write the attack, the layer at fault, and the fix. Ask a peer to run the optimiser pass, since you are too close to find your own holes.
- Write the two documents from b8 if AI touches any of it: the never-automate list and the disclosure.
- Book the two recurring meetings now - the weekly steering slot and the quarterly measure review with b9's five questions.
- Share the page with your team and ask the one question that matters: what here would you game, and what here feels unfair? Fix what they tell you before the quarter starts.
- Optional: read the leader track (a1 to a6) to see the same system from the side that designs and governs it.
Sources covered (cited)
The capstone draws on the whole course. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · A team's OKRs are well written but the system keeps producing hit targets and worse outcomes. Which layer is most likely missing?
Hit-the-target-and-something-else-pays is the b9 gaming pattern, and missing guardrails install it by default. Name what the target could damage and watch it - or promote it into the OKR set, as KR3 does for Cartwheel Growth.
2 · Which gate is the linchpin, without which the other five stop mattering?
Wire OKR scores to pay and people defend numbers instead of reporting honestly - which corrupts the evidence the other five gates depend on. Scores steer and teach; pay decisions belong in their own rare, well-evidenced conversation.
3 · What does the "optimiser" role in a red-team pass tell you?
Red-teaming is about the design, never the people. If you can find a shortcut in two minutes, someone under pressure will too - and the route they would take names the guardrail you have not written yet.