Infrastructure
Platform & Reliability Snapshot
Read where the platform actually stands, and decide what the first move should be — migration, SRE, or observability.
One page, one platform. Score it with the people who run it, not about them. Where two of them score a row differently, that gap is the finding.
Score each row0 Not true1 Sometimes, by effort2 Usually true3 True without anyone trying
| What you are judging | Strong looks like | Weak looks like | Score |
|---|---|---|---|
| Availability | |||
| Service levels existIs there an agreed level of service, and does anyone check it? | Targets are written, measured, and reviewed by a named person. | Availability is whatever it turned out to be last quarter. | |
| Failure is understoodDo you know how this platform fails, and what happens next? | Known failure modes, with a rehearsed response. | Each incident is a surprise, and the response is improvised. | |
| Change safety | |||
| Changes are reversibleCan you undo a bad change quickly, without a meeting? | Rollback is routine and takes minutes. | Rolling back is itself a risky operation. | |
| Deploys need no heroesCan a normal engineer ship on a normal day? | The path to production is the same for everyone. | Certain changes wait for certain people. | |
| Observability | |||
| Alerts are trustedWhen it pages, do people believe it? | Alerts are acted on, and silence means healthy. | Alerts are muted, filtered, or ignored by habit. | |
| "What changed?" takes minutesDuring an incident, how long to find what moved? | Minutes, from one place, without guessing. | Hours, across tools, by asking people. | |
| Cost | |||
| Spend is attributableCan you say what a team, product or client costs to run? | Cost maps to something the business recognises. | There is one large bill and a lot of theories. | |
| Someone owns the billWho is accountable for what this platform costs? | A named owner, reviewing it on a cadence. | Finance notices, engineering explains, nobody decides. | |
| Ownership | |||
| Every service has an ownerFor any service here, can you name the person accountable? | A name, not a team mailbox. | Ownership is inferred from git history. | |
| Out-of-hours is definedWhat happens at 3am, and who agreed to it? | A defined path, staffed and compensated. | Whoever answers the phone. | |
| Total | |||
How to read it
- Observability lowest
- Start there. Nothing else can be measured or trusted until you can see the system.
- Change safety lowest
- Start there. Reversibility is what makes every later change affordable.
- Availability lowest
- Service levels and an SRE practice — targets first, then the work to meet them.
- Cost lowest
- Attribution before optimization. You cannot cut what you cannot attribute.
- Ownership lowest
- This is not a platform problem. Fix ownership before funding any platform work.
- Mostly 2–3, still painful
- The constraint is probably the operating model, not the platform.
Read the lowest area, not the total — a platform rarely fails everywhere at once. If a migration is already underway, score the platform as it is today, not as it will be.