War-game a capacity reservation before reserved pools, overcommit rules, and forecast claims harden into how every team books headroom.
Capacity reservations fail when the draft invents headroom finance never funded, when utilization dual-counts the same cores as reserved and as burst, when platform still promises silent overcommit in private Slack, and when SRE cannot show which reservations actually cleared under peak load. A neat reservation spreadsheet is not evidence.
Lock one reservation design
State the proposed reserved pools and overcommit ceilings, which services and regions it covers, who owns forecast audit and exception relief, what success looks like after two peak windows of utilization and abort rate, and the kill criteria if shadow borrows or fairness disputes past a named threshold. Attach the current reservation inventory, sample peak metrics, cost attribution, and the measured path from request to granted pool. If platform, SRE, and finance disagree on whether the reservation buys predictability or just deferred borrow fights, reconcile before seating.
Name the decision you will make if the war game finds nothing new, and the hold criteria if any money-path service still lacks a named owner or a verified abort path.
Who should attack it
- Product. Fear that "reserved capacity" still means a quiet borrow with no named owner.
- SRE. Utilization math that looks clean while peak abort cohorts cannot be listed.
- Platform. Talk tracks that invent headroom the last peak never showed.
- Finance. Cost language that still hosts retired reservation rates.
- Skeptic. The claim that looks strongest and is least sourced.
Optional legal seat if customer capacity commitments bind the form. Same pack for every seat.
Run the passes
Feed Pingpong the reservation draft and exhibits. First pass steelmans the pool design. Later passes attack from the seats above. Final pass turns surviving objections into clearer ceilings, a published abort path, or a hold. Delete invented "we already hold enough headroom" claims and dual-counted utilization rates.
Force a day-after narrative: what happens if a product team screenshots mismatched reservation terms, if platform still teaches unwritten overcommit, or if finance cannot list the open exception cohort. If those stories are stronger than your mitigation plan, fix the package before you publish it. Separate a capacity reservation from a capacity plan or a usage limit change so each decision stays testable.
Ask every seat to mark which abort steps are optional in practice. Optional peak drills become skipped within a month if nobody names the owner and the time box. Related: stress-test a multi-region cutover, pretend you are the SRE manager, pretend you are the CTO, and the war-game decisions hub. Process: how to run a Pingpong.