Is it Time to Retire Uptime Percentages?

On 2026-09-03, I came across a piece by Jason Gorman titled “The Wall Confronting Reliable Coding Agent Autonomy.” In it, he makes a poignant observation:

The journey from 90% to 99% reliability is just as difficult as the initial climb to 90%.

This resonated with me deeply, especially since I've noticed a trend of increasing stability instability in my daily workflow.

The Status Page Struggle

Whenever things feel "off," I find myself navigating to various service status pages. I'm usually greeted by a sea of colors and statistics, often claiming something like 100% uptime.

When I see these numbers, I have to consciously apply Gorman's logic to translate them. It feels less like reading a report and more like interpreting seismic data.

The Magnitude Problem

In seismology, the gap between a $6.2$ and a $7.8$ earthquake might look small numerically, but the actual energy released is vastly different. Uptime metrics suffer from the same non-linear deception:

• 99.9% looks almost identical to 99.99%.
• In reality, the latter represents 10$\times$ less downtime.

The "Nines" Hierarchy

Infrastructure engineers have developed a shorthand to cope with this:

• "Two nines"
• "Three nines"
• "Four nines"

While those who live in the server room intuitively understand these distinctions, the general public—who now frequently use status pages—does not.

A Flawed Interface

Status pages serve as the primary window into a service's reliability. However, they are currently designed for compliance and standardization rather than human communication.

Most users aren't looking for a standardized industry metric; they just want to know:

"Dude, how much have you been down lately? It feels like a lot..."

Comparing the Approaches

The Logic of Downtime

To understand why percentages fail us, consider the math. The amount of allowed downtime $D$ for a given percentage $P$ over a time period $T$ is:

$D = T \times (1 - P)$

If we want to calculate this programmatically, it looks like this:

def calculate_downtime(percentage, days=30):
    total_minutes = days * 24 * 60
    downtime_minutes = total_minutes * (1 - (percentage / 100))
    return downtime_minutes

# Example: 98.31% uptime over 30 days
print(f"{calculate_downtime(98.31)} minutes of downtime")

A Better Path Forward

We should stop relying solely on the "nines." Instead, we should prioritize clarity.

Checklist for a Better Status Page

• Move away from percentage-only displays.
• Lead with actual time lost (hours/minutes).
• Keep the percentage as a secondary, parenthetical detail.
• Use plain language instead of "infra-speak."

The Bottom Line: Instead of saying GitHub Actions: 98.31% uptime, why not say: GitHub Actions: 12 hours affected in the last 30 days (98.31% uptime)? One requires a math degree in reliability; the other just requires a clock.