Known unknown: Section 8, The commit-point method: "Nobody has measured it in public, as far…"

[anacay]

Nobody has measured it in public, as far as the sources show.

Re-pegging a launch reserve to another tenant when a launch slips looks nearly free. Across enough launches it may absorb most timing misses. That's speculation. Has anyone measured how often a slip's reserve gets re-used?

Author's note 8 on Section 8, The commit-point method. Reply below; if this settles into a change to the text, an editor will carry it over with credit (see HOW_IT_GROWS.md).

[dave-dev650]

I don't have public measurements on this. I would caution that the largest source of friction is likely to be political rather than technical or supply as the team with the slipped launch needs to have really clear incentives for their capacity to get re-used. Mostly that it doesn't block future launches and yet also doesn't motivate everyone to over-forecast because getting rid of capacity at the last instant is too trivial.

[anacay]

Thank you, that’s a useful correction to the speculation. You’re right that the hard part is political, not technical or supply: the team that slipped needs a clear reason to let its capacity be reused. That points at something larger than note 8, so this reply goes further than the others.

Why incentives belong at the center

A capacity plan is a set of promises between teams, and every number in it comes from someone with a stake in the outcome. The paper’s title says the coefficient is an agreement. An agreement holds only if keeping it serves the people who made it. Teams meet again every cycle, so each one quietly asks: “If I tell the truth early, am I better or worse off next time?” When the answer is “worse,” good methods still produce bad numbers. We think that makes incentives part of the method, not politics outside it.

A worked case

Here’s an invented example. A team holds a reserve for a March launch. In January it learns the launch will likely move to June. Four forces push it to stay quiet:

Fear of loss. Its new date is uncertain, and it doubts it will get the capacity back.
The ratchet. Giving capacity back can look like an over-ask and shrink its next allocation. This is a known problem wherever future targets are set from current results (Weitzman, 1980).
Late news. Admitting a slip is uncomfortable, so it surfaces late.
Idle is free. If holding costs nothing, nothing pushes the team to let go.

There is also an opposite failure. If handing capacity back at the last moment costs nothing, every team has a reason to over-forecast: ask for extra, then return what it doesn’t need. Across many teams, the whole plan inflates. So the goal is a balance. Releasing early should be safe, and it should never count against a team’s future launches. But releasing late should not be free.

Simply expiring the reserve backfires too. When budgets expire at year end, final-week spending runs several times the normal rate, and year-end projects tend to be of lower quality (Liebman and Mahoney, 2017).

Psychology adds weight. Losses feel larger than equal gains (Kahneman and Tversky, 1979). Once something is ours, we value it more (Thaler, 1980). And people stay with the default (Johnson and Goldstein, 2003). So giving capacity back feels like a loss, the reserve feels like “mine,” and holding wins unless something changes the default.

Possible handshakes

A slipped reserve can be in one of two states, and each needs different handling.

Before delivery, the reserve is still a commitment, so there is something to cancel or redirect. Simple rules work well here:

Refund schedule. Like a hotel cancellation policy: full credit for releasing early, partial if late, none after the commit point.
Default flip. At each commit point, a reserve whose date hasn’t been reconfirmed returns to the pool unless the team reaffirms it with evidence.

After delivery, the capacity exists and sits idle. There is nothing to cancel, and nobody announces the slip. It may not even look idle. If idle capacity risks automatic reclamation, a team can move it into a larger service it already runs, where it blends into the same locations. That service’s utilization dips slightly, and the slip disappears into the average. Reclamation meant to recover idle capacity ends up teaching teams to hide it, much as expiring budgets teach them to spend. So detection comes first. The signals are subtle: a service whose capacity grew without a matching forecast, or whose utilization stepped down just as a reserve for another launch landed. Even then, low use is ambiguous. It could mean a slip, an over-ask, an efficiency gain or lost demand. The planner asks which before acting, because each calls for a different response.

Here again, a formal capacity planning discipline earns its keep. A hidden slip is hard to see from inside any one role. An SRE sees whether a system is healthy, and idle capacity looks perfectly healthy. A TPM sees a program’s schedule, but not what its reserve is doing. Finance sees what is spent, not what is used. The planner sees all of it together: every team’s forecast ramp, actual use across the fleet, the supply still coming, and who is waiting. That position lets the planner spot the gap early, tell a slip from an efficiency gain or lost demand, and broker a loan to a team that needs it. And because the planner is neither the holder nor the borrower, both sides can trust the call.

Once a slip is confirmed, exchanges help it surface sooner next time:

Lend with recall. The idle capacity is lent with a recall date, and the slipped team keeps ownership and its place. The central reserve backs the promise when a recall collides with other demand.
Return with interest. A team that volunteers its idle capacity before a review catches it regains more priority later. Pay only for voluntary releases, pay in priority rather than quantity, and cap it by the forecast record.
Time swap. The borrower repays from its own future allocation when the slipped launch lands. It’s only as reliable as the borrower’s ability to repay.

Weaker on their own:

Slip insurance. Each team pays a small premium into a shared pool, and a covered slip gets its priority restored later. It pools risk the way insurance does: slips are hard to predict one at a time but steady across many teams. The catch is one insurers know well. Protected teams may commit to riskier dates, and teams that expect to slip join most eagerly. Insurers answer with deductibles and premiums that rise with a history of claims; here, the forecast record could play that role.
Internal market. Selling the reserve rewards truth, but invites over-asking in order to sell.
Reputation. Crediting early, accurate reports in the forecast record is cheap and builds trust, and it strengthens every option above.

Putting it together

No single handshake solves this, and that’s the point. Each force needs its own answer, and each state of the reserve needs its own tool.

Before delivery, the goal is early truth. A refund schedule makes releasing early pay, and a default flip makes silence cost. Together they turn a slip from a confession into a routine check at each commit point.

After delivery, the goal is visibility. The utilization review, read against the forecast record, finds capacity that is idle or hidden. Lending or return with interest then offers a better path than being found: volunteer and keep your place, or volunteer and gain priority.

Across both states, the forecast record does two jobs. It blocks gaming, because an inflated ask stays on the record even if the capacity is lent. And over time it builds the trust that lets a team believe a promise of return.

The right mix depends on that trust. Where trust between teams is low, start with rules: they are simple, visible and hard to argue with. As trust grows, exchanges can carry more of the load, because teams believe the capacity will come back. The aim is a plan where reporting a slip early is simply the easiest thing to do.

Whatever the mix, it has to be clear. A team deciding whether to report a slip should be able to predict exactly what happens next. If the rules are too clever to follow, teams fall back on holding.

That is why we think incentives belong at the center of the discipline, and it brings us back to the planner’s vantage point. Forecasts, supply and methods all rest on what people choose to report. Each team sees only its own numbers, so no team can see where honest reporting breaks down. The planner sees every forecast, every reserve and every team waiting, all at once.

That work spans the whole organization: sideways between teams, and up and down from the engineers who run services to the executives who decide what doesn’t fit. But it isn’t ordinary cross-functional work. A TPM brings teams together toward a shared goal. An SRE works across teams to keep systems reliable. The planner works across teams whose goals compete for the same scarce capacity. It holds none of that capacity and owns none of the outcomes, so its standing comes from being neutral and from keeping the record. It has to speak every party’s language: machines for engineering, money for finance, dates for product, lead times for supply. And it has to design the rules that make honest reporting the sensible choice for each of them. That is what keeps the promises between teams worth keeping.

The proposal

Change: rewrite note 8 in Section 8. The main cost of reusing a slipped reserve is incentives, not supply or tooling. The new note would summarize the forces, the two states and the handshakes above.

Next: I’ll open a separate proposal to treat incentives as a core concern of the discipline. That touches the principles, so it gets the longer open period.

Evidence: unmeasured for how often reserves are reused or hidden. Argued for the two states and the handshakes. Published for the underlying patterns, as cited.

Questions for anyone who has lived this: Which arrangements have you seen work? What made teams report a slip early? How have you found capacity that was hidden rather than idle? And where have incentives quietly broken a plan that looked sound on paper?

Open for comment until 23 November: 14 days, doubled while there’s one editor. I’ll call it then and record the decision here.