MAXOPS 
THE
DIALS
Tune too hard and every scan becomes noise you learn to ignore. Tune too softly and you quietly miss the savings you installed this for. Here is how to find the middle.
Every one ships with a value that already works. You are adjusting, never configuring from scratch.
WHY THERE ARE
DEFAULTS
Eighty-four of my hundred and three questions have a dial on them. Not one of them starts empty — every dial arrives at a value I already think is right for most accounts.
That matters more than it sounds. A tool that asks you to define "idle" before it will tell you anything is a tool nobody finishes setting up. MaxOps takes a position first — an instance under 5% CPU for a week is idle — and lets you argue with it afterwards.
So should I be changing these at all?
Not on day one. Run a scan, read what comes back, and change a dial only when a finding is wrong about your workload. Tuning ahead of evidence is just guessing with extra steps.
THE THREE
PRESETS
Normal is not a compromise — it is literally the shipped defaults, untouched. Conservative makes every check harder to trigger: it waits longer and demands stronger evidence, so you see fewer findings and trust them more. Aggressive does the opposite — it surfaces more, sooner, and expects you to do the filtering.
Pick by appetite, not by ambition. Conservative in production where a wrong call is expensive; aggressive in a sandbox where you actually want to see everything.
The preset is the first thing to reach for and the only one most accounts ever need.
A PRESET DOESN'T
REPLACE YOUR
NUMBERS
It scales them. And it knows which way each one points — because for some settings "aggressive" means a bigger number, and for others it means a smaller one.
Take one check — idle EC2 instances — and move the lever. Watch the two columns go opposite ways:
| Setting | Conservative | Normal | Aggressive |
|---|---|---|---|
| Days idle | 12 | 7 | 4 |
| CPU under | 3.5% | 5.0% | 7.0% |
| Network under | 700 | 1000 | 1350 |
Going aggressive shortens the wait to 4 days but raises the CPU bar to 7%. Both changes point the same way — catch more instances — even though one number went down and the other went up. That is the whole trick.
Decides how eagerly every check reports. Set once, in Settings.
Rightsizing tier — per recommendationDecides how much headroom one instance keeps. Chosen on the recommendation itself.
Different controls. Note the middle name differs — Normal here, Balanced there.
THE PRESET IS NOT
THE TIER
I've seen conservative and aggressive in two different places. Same thing?
No — and they're worth keeping straight. The preset in Settings decides how eagerly every check speaks up. The tier on a rightsizing recommendation decides how much breathing room one instance keeps after being resized.
You can sit on the Conservative preset and still choose the Aggressive tier on a particular instance you know is over-provisioned. They answer different questions and never override each other.
The giveaway is the middle option: the preset's is Normal, the tier's is Balanced. If the word in front of you is "Balanced", you are looking at a rightsizing recommendation.
WHEN ONE CHECK
IS WRONG ABOUT YOU
The preset moves everything. Sometimes only one question is wrong — your batch box really is idle for three weeks a month, and that's fine. Change that one dial and leave the rest alone.
Any check can be adjusted individually, and an adjusted check stops following the preset for the settings you changed — your value stands until you clear it. Settings shows a running count of how many checks you've customised, so it's always visible how far you've drifted from the shipped behaviour.
Keep that number small and you can still reason about your own configuration. Once it's forty, nobody remembers why any of them are what they are.
How long the quiet must last before it counts.
Quiet enough to call idle.
Because a box can be idle on CPU and still busy on the wire.
Every one must hold. A check's dials are joined by AND, never OR.
This check has more CPU dials than the three shown — peak, 90th and 95th percentile too.
WHAT A THRESHOLD
ACTUALLY ASKS
The single most useful thing to know about a check's dials: they combine with AND. An instance is only idle if it is under 5% CPU and under its peak and percentile limits and under the network floor and has been for seven days. Loosening one dial does not open the floodgates — every other one still has to agree.
Which is why raising CPU from 5% to 7% is safer than it sounds, and why dropping the days from 7 to 1 is more dangerous than it sounds. Time is the dial that protects you from transients.
Each control also has limits. Values are held inside a sensible range and snapped to sensible increments, so a preset can never push a setting somewhere the check cannot use — and neither can a slip of the mouse.
TUNED TOO
HARD
This is the way tuning actually goes wrong. Not an outage — just three hundred findings on Monday morning, and a team that quietly stops opening the dashboard.
Over-tuning is seductive because it feels like diligence. More findings look like more value, right up until the ratio of real to trivial tips over and the whole list becomes something to dismiss. A tool you have learned to ignore saves nothing.
"Replace the current tuning" is the phrase to read twice.
BEFORE YOU
LOSE IT
Settings warns you in two situations, and they are not the same warning.
The order that saves you grief: pick the preset first, look at what it produced, and only then adjust individual dials. Doing it the other way round means watching your work get recalculated away.
CHANGING THE
ACCOUNT
Every other setting on this page is reversible. This one is not. Changing the AWS account number doesn't re-point me at a new account — it wipes everything I learned about the old one.
Confirming that dialog removes, for the current account:
That last one stings most. Months of decisions about which findings were legitimate exceptions — and the reasons written beside them — are attached to the account they were made in. Continue only if you genuinely want to start fresh.
What if I just typed the wrong number?
Then cancel. The dialog exists precisely because a typo and a deliberate migration look identical to me — only you can tell them apart.
THE DIALS WORTH
TOUCHING
Of the 119 settings, a handful account for nearly every adjustment a real account needs.
| Dial | Found on | Turn it when |
|---|---|---|
| Days idle | idle instances | Your workloads are genuinely quiet for long stretches — batch, seasonal, dev boxes. |
| CPU under | idle instances | Your "busy" baseline is unusually low, so 5% never triggers on real waste. |
| Minimum age | unattached volumes | Maintenance routinely leaves volumes detached for longer than a week. |
| Days stopped | unused instances | You keep instances stopped deliberately, for longer than a month. |
| Network under | idle instances | Chatty agents keep quiet boxes above the floor and hide real idleness. |
| Excluded names | various | A known set of resources should never be considered at all. |
Everything else is best left where it is until a finding proves it wrong. That is the whole method: change a dial to answer a specific complaint, not to explore.
PRESET FIRST.
OVERRIDE ONLY
WHAT'S WRONG.
The defaults are an opinion, the preset is an appetite, and a per-check override is an argument you won against a specific finding. Keep them in that order and the dashboard stays worth reading.