Reliability Objective

Building toward 99.9% uptime

Not a promise we've already kept. A standard we're working toward with 12 full-time engineers (41 counting contractors). The companies that hold this line run 200x to 500x more engineers. Every point of reliability here is earned through deliberate trade-offs between shipping and stability.

Where we actually are

Last 90 days

Target

99.9%

What this means

At 99.9%, we budget for 43 minutes of downtime per month. That's the margin between “reliable” and “broken.” Every incident eats into that budget, and when it runs low, we stop shipping features and fix foundations.

How we calculate uptime

Only severities that meaningfully degrade the platform count against uptime. Everything else is tracked on the timeline but spends no error budget.

SEV1100%Critical outage, counts as full platform downtime
SEV2n/NWeighted by affected services out of total
SEV3–40%Shown on the timeline, no impact on uptime %
PROVIDER0%3rd-party provider degradation, informational only
Untagged0%No impact until a severity is declared

Not counted isn't the same as not noticed. The dashed line on the trend chart counts every degraded minute at full weight — when it drops below the green line, users felt something even though the SLO never moved.

Why 99.9 and not higher

Each additional “nine” costs roughly 10x more to achieve — dedicated SRE rotations, multi-region failover, the works. At our size, 99.9% is the honest ceiling: ambitious enough to take seriously, grounded enough to actually hold.

SLO vs SLA

This is an internal objective, not a contractual guarantee. An SLO tells us when to invest in stability. An SLA tells you the consequences if we don't. We publish our SLO because we believe transparency builds more trust than fine print.

Allowed downtime by level

UptimeMonthlyQuarterlyYearly
95%
1d 12h4d 12h18d 6h
99%
7h 12m21h 36m3d 15h
99.9%target
43.2m2h 10m8h 46m
99.95%
21.6m1h 5m4h 23m
99.99%
4.3m13m52.6m
99.999%
25.9s1.3m5.3m

The team behind it

41 in the department
12full-time on the SLO
29contractors

29% full-time

VP, Software Engineering1leads the org
Product Engineering
8full-time24contractors
Platform Engineering
3full-time2contractors
Platform Operations
0full-time2contractors
Quality Assurance
0full-time1contractors

All forty-one carry this number. Our contractors aren't a buffer around the work — they do it alongside our full-time engineers, often absorbing the same on-call load and the same crunch, without the equity or long-term security that's supposed to make that trade fair. Holding this SLO on a team this size means leaning on people harder than any company should, and that gap is ours to close, not theirs to keep covering. Naming it here is the floor, not the fix.

Who targets what

Counted against our 12 full-time engineers — the same kind of dedicated, long-term investment these organizations made. Our 29 contractors carry this line right alongside them; the gap is that we haven't yet backed them with the same commitment, not in anything they bring to it.

95–98%No formal target

Most seed and Series A startups

No uptime target, no on-call rotation, no incident process. Downtime happens when someone notices. 95% means 18 days of outage per year.

~3 engineerssmaller than us
99%Two nines

Early Twitter, early Shopify, fast-scaling startups

Where companies land when they're growing faster than their infrastructure can keep up. Monitoring exists, someone carries a pager, but 3.5 days of downtime per year is the reality.

~100 engineers8.3×our team
99.9%Three ninesOur target

GitHub, Slack, Zoom — and us

Our target. Names like these hold it with dedicated platform and SRE teams in the hundreds of full-time engineers; we hold it with twelve. Our contractors are exceptional and carry this line with us — we just haven't backed them with the equity and long-term security those names gave their full-time teams. Same bar, a fraction of the headcount we've committed to.

~300 engineers25×our team
99.99%Four nines

Stripe, Google, Amazon

Dedicated reliability orgs, multi-region infrastructure. About 4 minutes of downtime per month.

~2,500 engineers210×our team
99.999%Five nines

Visa, Mastercard, NYSE

Global payment rails processing 65,000+ transactions per second, run by reliability orgs in the thousands. About 26 seconds of allowed downtime per month.

~10,000 engineers830×our team
99.9999%+Six nines and beyond

Air traffic control, medical devices, nuclear systems

Safety-critical systems with unlimited budgets. An entirely different engineering discipline from web services.