Exchange outages matter more than they look
Exchange downtime gets treated as a customer service problem. The site was unreachable for ninety minutes during a sharp move, some people complained, it came back, everyone moved on. Framed that way it is an annoyance.
Framed correctly it is a disclosure. An outage during a large move is the venue telling you how it behaves precisely when its behaviour matters most, and that information is otherwise very difficult to obtain.
Failure correlates with the moment you need it
Trading venues do not fail randomly. They fail when volume spikes, and volume spikes when prices move violently. This means downtime is structurally concentrated in exactly the windows where access has the most value — the ones where you might need to reduce a position, meet a margin call, or move funds out.
An availability figure averaged over a year hides this completely. Ninety-nine point nine percent uptime sounds excellent and is entirely compatible with being unreachable during every significant move of the year. The average is measured against calendar time; what you care about is availability weighted by how much the access was worth.
Some responses are worse than the outage
It is worth distinguishing between kinds of failure. A venue that goes fully offline is being straightforwardly unavailable. A venue that stays online but suspends withdrawals while trading continues has made a choice — it decided which users to disadvantage. A venue that keeps taking orders but stops honouring stops or liquidations at quoted prices has done something worse still.
These distinctions rarely survive into the summary. All three get reported as “technical issues”, and the incident post-mortem, if one appears, tends to describe infrastructure rather than decisions. Reading what a venue actually did, rather than what it called it, is one of the few genuinely informative exercises available to an outside observer.
What it tells a long-term holder
If you hold on an exchange, downtime is a direct preview of your counterparty risk. The engineering failure and the solvency failure are not the same thing, but they draw on the same underlying quality of operation, and venues that handled load badly have a poor historical record of handling stress well in other respects.
This is the practical argument for self-custody that does not depend on ideology. It is not that exchanges are untrustworthy in principle. It is that keeping a long-term position somewhere whose availability is worst exactly when you might need it is a structural mismatch between the holding period and the venue.
The exchange is a place to transact. It is a poor place to keep something for years, and every outage during a violent move is a reminder delivered at no cost to you — provided you were not trying to act at the time.
What to actually do with the information
Keep a note. When a venue you use goes down, write the date, the duration, what was suspended, and how the market moved during the window. Over a couple of years this becomes a far better picture of operational quality than any marketing page, and it costs a minute per incident.
Spread transactional balances across more than one venue, so a single outage does not remove all your optionality at once. And keep the long-term position somewhere that has no uptime at all, because a private key does not have an availability figure — it works or it does not, and whether it works has nothing to do with how many other people are trying to use it that afternoon.
None of this requires distrust of any particular company. It requires noticing that the times you would most want access are the times access is least likely to exist, and arranging things so that gap costs you nothing.