05 · Decision rights Whether it is clear who decides what, and whether that survived the last time headcount grew.
For a contested call last quarter, everyone would name the same decider. 0 Absent. Different people would name different deciders.1 Ad hoc. Most would name the same person, and it is the founder for everything.2 Consistent. Clear for the main areas, ambiguous at the edges.3 Engineered. Clear, written down, and matches what actually happened.
What it costs: Ambiguous ownership adds one to three weeks to any contested decision, spent entirely in meetings about who should decide.
Decision rights were revisited the last time the team grew significantly. 0 Absent. Never revisited. The structure is the one from the first ten people.1 Ad hoc. Adjusted informally, never stated.2 Consistent. Revisited at the last growth step.3 Engineered. Revisited routinely, and roles have actually changed as a result.
What it costs: Structures that do not scale with headcount concentrate decisions on two or three people whose calendars then gate the whole studio.
Reversible decisions are made quickly and without ceremony. 0 Absent. Everything goes through the same heavy process.1 Ad hoc. The distinction is understood but not applied.2 Consistent. Applied, most of the time.3 Engineered. Explicit — reversible calls are delegated by default.
What it costs: Treating every decision as irreversible is the most common cause of a studio that feels slow at forty people and felt fast at fifteen.
People who lost an argument still execute the decision properly. 0 Absent. Losing an argument means quiet non-compliance.1 Ad hoc. Execution happens but slowly and with visible reluctance.2 Consistent. Mostly good-faith execution.3 Engineered. Disagreement is voiced, the call is made, and execution is wholehearted.
What it costs: Half-executed decisions produce the worst outcome available — the cost of the change without the benefit — and they are usually discovered a quarter later.
Work does not stop when the founder or studio head is away for two weeks. 0 Absent. Most decisions wait for their return.1 Ad hoc. Some things wait; the team works around the rest.2 Consistent. Little waits, and what does is genuinely their call.3 Engineered. Nothing waits that should not.
What it costs: A two-week absence that stalls decisions costs the whole team's throughput for those two weeks, not just the one person's.
07 · Live operations Cadence between drops, retention response, incident readiness.
The interval between content drops is deliberate and holds. 0 Absent. Content ships when it is ready, which varies by months.1 Ad hoc. A target interval exists and is missed more often than met.2 Consistent. Mostly held, with occasional slips.3 Engineered. Held, and the interval was set from retention data.
What it costs: An irregular drop cadence is the most reliable predictor of a between-content retention dip, and each dip costs a measurable percentage of the returning population.
A retention drop triggers a response within a known time. 0 Absent. Drops are noticed late and discussed rather than acted on.1 Ad hoc. Noticed quickly, response is improvised each time.2 Consistent. A response process exists and has been used.3 Engineered. The response is rehearsed, owned, and time-bounded.
What it costs: Players lost in the first two weeks of a dip are the hardest and most expensive to reacquire, typically several times the cost of the content that would have held them.
An incident at 2am on a Saturday has a known path. 0 Absent. No on-call. Whoever sees it, handles it.1 Ad hoc. An informal rota with no written escalation.2 Consistent. Documented on-call with clear escalation.3 Engineered. Documented, rehearsed, and reviewed after each incident.
What it costs: Weekend downtime without a path routinely runs eight hours instead of one, and the difference is refunds plus the review-score damage that does not recover.
There is content committed far enough ahead to cover the next two drops. 0 Absent. The next drop is being built now, with nothing behind it.1 Ad hoc. One drop of runway.2 Consistent. Two drops, usually.3 Engineered. Two or more, consistently, with a buffer that has survived a slip.
What it costs: A single drop of runway means any slip becomes a visible content gap, and the gap costs more players than the delayed content would have gained.
Live support and the next major release are separately staffed. 0 Absent. The same people do both, and live always wins.1 Ad hoc. Nominally separate, in practice the same people.2 Consistent. Separate for the most part.3 Engineered. Genuinely separate, with a stated rule for borrowing between them.
What it costs: When one team does both, the next release absorbs every live emergency, which is how a six-month project becomes a nine-month one without a single decision to extend it.
02 · Milestone truth Whether reported status matches reality, and how long a slip takes to become visible to leadership.
A slip reaches leadership within a week of the team knowing. 0 Absent. Leadership learns at the milestone review, on the day.1 Ad hoc. It surfaces within a month, usually through a side channel.2 Consistent. It surfaces within two weeks, through the normal reporting line.3 Engineered. Within a week, and the report names the date it was first known.
What it costs: Every week a slip stays invisible is a week of dependent work built on a false date, and that rework is typically two to three times the original delay.
A milestone has a written definition of done that someone outside the team could verify. 0 Absent. Milestones are named but not defined.1 Ad hoc. Defined in a document written at the start and not opened since.2 Consistent. Defined, current, and used in the review.3 Engineered. Defined, current, and verified by someone who did not build it.
What it costs: Undefined milestones are declared complete and reopened an average of once each, which is a second pass nobody budgeted.
A task reported green in the last month later turned out not to be. 0 Absent. This happens most months and nobody treats green as meaningful.1 Ad hoc. It happens, and it is treated as an individual failing rather than a system one.2 Consistent. It is rare, and when it happens the reporting path gets examined.3 Engineered. Rare, examined, and the examination has changed how status is collected.
What it costs: Status that cannot be trusted means leadership re-verifies manually, which is one to two senior days a week spent re-deriving what the report already claimed.
Someone junior can report a problem without going through the person responsible for it. 0 Absent. No such path exists. Bad news travels through the person it reflects on.1 Ad hoc. A path exists on paper and has never been used.2 Consistent. It has been used at least once and the person was not punished for it.3 Engineered. It is used routinely and the reports are acted on.
What it costs: This is the single cheapest fix on this page and the one most often missing; the problems it surfaces are typically found four to eight weeks earlier than they otherwise would be.
Estimates are compared against actuals, and the comparison changes later estimates. 0 Absent. Actuals are not recorded.1 Ad hoc. Recorded somewhere, never looked at.2 Consistent. Reviewed at the end of a milestone.3 Engineered. Reviewed, and a named estimating bias has been corrected as a result.
What it costs: Teams that never close this loop run a stable 30 to 50 percent optimism bias, which compounds across a project into a full quarter.
01 · Release cadence How often you ship, how predictable it is, and whether the interval was chosen or inherited.
The interval between releases was decided, not inherited. 0 Absent. Nobody can say why releases happen when they happen.1 Ad hoc. The interval matches whatever the last team did, and has never been revisited.2 Consistent. The interval was chosen once, for a reason someone can state.3 Engineered. The interval is reviewed against delivery data, and has been changed at least once as a result.
What it costs: An inherited cadence is usually two to four weeks longer than the work requires, which is a full extra release cycle of salary per quarter buying nothing.
The last three releases went out on the date announced internally at the start. 0 Absent. Dates move so often the announced date is not treated as information.1 Ad hoc. One of the last three held; slips are announced in the final week.2 Consistent. Two of three held, and slips were flagged with more than two weeks' notice.3 Engineered. All three held, or moved once with a stated reason more than a month out.
What it costs: A slip discovered inside two weeks costs the marketing spend already committed against the original date, typically the largest single avoidable write-off in a release.
There is a known cut line, and cutting to it does not require an escalation. 0 Absent. Everything in the release is mandatory until someone senior says otherwise.1 Ad hoc. Scope is cut, but only in a crisis meeting in the last two weeks.2 Consistent. A cut list exists before the release starts and is used.3 Engineered. The cut line is agreed at kickoff, and the team cuts to it without asking.
What it costs: Without a standing cut line every scope decision escalates, and the two to five days each escalation waits is time the whole release is blocked.
Shipping a release costs a known and shrinking number of person-days. 0 Absent. Nobody has measured it. The answer is "it takes as long as it takes."1 Ad hoc. People could estimate it, roughly, and it has not changed in a year.2 Consistent. It is measured, and the number is known to the team.3 Engineered. It is measured and has come down over the last three releases.
What it costs: A manual release process typically consumes three to eight person-days per cycle, which at a monthly cadence is most of a full-time engineer spent on the same steps forever.
A bad release can be pulled or fixed forward within a day. 0 Absent. There is no rollback path; a bad release stays live until the next one.1 Ad hoc. Rollback is possible but has never been rehearsed.2 Consistent. Rollback has been done at least once and the steps are written down.3 Engineered. Rollback is routine, tested, and takes hours rather than a day.
What it costs: A bad build live for a weekend costs the refund window plus the review score, and the review score does not recover with the fix.
04 · Scope and change control Whether a scope change has a visible cost at the moment it is made.
Adding scope produces a stated cost in the same conversation. 0 Absent. Scope is added in conversation and absorbed silently.1 Ad hoc. The cost is worked out later, sometimes.2 Consistent. A cost is stated within a few days.3 Engineered. Nothing is added without naming what moves or comes out.
What it costs: Silent absorption is how a team arrives three months late having said yes twelve times, each of which looked like a week.
When something goes in, something identifiable comes out. 0 Absent. Nothing ever comes out.1 Ad hoc. Things come out only when the deadline forces it.2 Consistent. Trades are made deliberately, most of the time.3 Engineered. The trade is explicit and recorded every time.
What it costs: A backlog that only grows converts directly into overtime, and overtime converts into the attrition that costs three to six months per departure to replace.
There is a point after which change requires a named person's approval, and it holds. 0 Absent. No such point exists.1 Ad hoc. It exists and is routinely overridden.2 Consistent. It exists and mostly holds.3 Engineered. It holds, and the exceptions are documented and rare.
What it costs: Late change is the most expensive kind — a change in the final month typically costs three to five times the same change made early.
The team can decline a request without escalating to leadership. 0 Absent. Every request is accepted; declining is not a normal act.1 Ad hoc. Declining requires escalation and is therefore rare.2 Consistent. The team declines directly, sometimes.3 Engineered. Declining with a stated reason is routine and uncontroversial.
What it costs: When declining requires escalation, most requests are simply accepted, and the accumulated yes is the scope problem.
06 · Technical risk Whether the upgrade and the roadmap compete for the same people, and whether anyone has priced that.
The cost of the next engine or platform upgrade has been estimated. 0 Absent. Not estimated. It is discussed as something that will need doing.1 Ad hoc. A rough number exists in someone's head.2 Consistent. Estimated and written down.3 Engineered. Estimated, written down, and scheduled against the roadmap.
What it costs: An unplanned major upgrade absorbs two to four months of senior engineering, and it lands when the platform forces it rather than when the roadmap can afford it.
It is known which people the upgrade and the roadmap both need. 0 Absent. The conflict has not been named.1 Ad hoc. Everyone knows informally and it has never been stated.2 Consistent. Named, and the trade is discussed.3 Engineered. Named, and the trade has been explicitly decided by someone with the authority to.
What it costs: Two efforts sharing three senior people means both run at roughly half speed while appearing fully staffed on a plan.
Any externally imposed technical deadline is known and dated. 0 Absent. No idea when platform or SDK support ends.1 Ad hoc. Aware such deadlines exist, no dates.2 Consistent. Dates known for the main dependencies.3 Engineered. Dates known, tracked, and work is scheduled ahead of them.
What it costs: A missed platform deadline can remove the title from a store, and the emergency remediation is several times the planned cost.
Technical debt appears somewhere leadership actually looks. 0 Absent. It exists only as engineer frustration.1 Ad hoc. A list exists that nobody outside engineering reads.2 Consistent. Represented in planning as real work.3 Engineered. Represented, with a stated cost, and traded against features explicitly.
What it costs: Debt invisible to leadership is never funded, and the interest is paid as a slow decline in feature velocity nobody attributes to it.
03 · Telemetry that produces decisions Not whether dashboards exist. Whether anyone changed a decision because of one in the last quarter.
A decision changed in the last quarter because of something in the data. 0 Absent. No example exists.1 Ad hoc. One example, and it was a decision nobody disagreed with anyway.2 Consistent. A clear example where the data contradicted an assumption and won.3 Engineered. Several, and reversing a plan on evidence is unremarkable here.
What it costs: Telemetry that changes nothing is pure cost — typically an engineer's time to build plus a monthly platform bill, returning zero.
Each core metric has a named owner who would notice it moving. 0 Absent. No metric has an owner.1 Ad hoc. Ownership is implied by team, not by person.2 Consistent. Named owners exist for the main metrics.3 Engineered. Named owners, and they have raised something unprompted in the last quarter.
What it costs: An unowned metric that moves is typically noticed four to six weeks late, which is the difference between a fix and a postmortem.
New instrumentation starts from a question someone wants answered. 0 Absent. Events are added because they are easy to add.1 Ad hoc. Added in bulk, at the start, in case they are useful later.2 Consistent. Usually tied to a stated question.3 Engineered. Always, and instrumentation with no question is removed.
What it costs: Event bloat raises the pipeline bill and slows every query against it, and the cost scales with players rather than with usefulness.
The team believes the numbers. 0 Absent. Two dashboards disagree and nobody has reconciled them.1 Ad hoc. The numbers are believed with caveats everyone repeats.2 Consistent. Believed, with known and documented gaps.3 Engineered. Believed, and discrepancies are treated as incidents.
What it costs: A disputed number restarts the same argument every time it appears, and the meeting to re-litigate it costs more than the reconciliation would have.