CertaDNS
Skip to lesson

The Permutation Space · lesson 2 of 3

Cutting it to what matters

After this lesson you can

Reduce a permutation space of thousands to a watchlist you would actually act on.

Assumes you have read Generating the space.

The generated space is tens of thousands of names. A watchlist is what remains after you ask, of each one, whether a registration would change what you do that day.

The test

If this domain were registered tomorrow, would I act? If the answer is no, watching it produces an alert you will dismiss, and dismissing alerts is a habit that generalises.

Where the cuts are

CutKeepsDiscards
By plausibility to a readerNames a hurried person would not questionVariants that look obviously wrong in an address bar
By TLDThe TLDs your customers associate with you, plus the cheap ones abused at volumeThe long tail of TLDs nobody transacts in
By brand elementThe brand name itself and the product names customers typeInternal project names nobody outside would recognise
By techniqueCombosquats and single-character errorsBitsquats, unless you are large enough for random bit flips to matter

Bitsquatting, honestly

A bitsquat exploits memory errors flipping a single bit in a cached hostname. It is real, it has been demonstrated, and the traffic it captures is proportional to your query volume.

  • At internet-scale query volumes it is worth registering the variants.
  • Below that, the expected traffic is approximately zero.
  • It belongs in a generated list and near the bottom of a watchlist, for almost everyone.

A worked cut

Generated:                        ~22,000 names
Drop TLDs with no customer link:  ~4,000
Drop implausible-to-a-reader:     ~900
Drop bitsquats:                   ~700
Keep combosquats with high-risk
  words (login, secure, billing,
  support, pay, account, verify):  +120

Watchlist:                        ~800 names

Of which, escalate immediately on:
   - any that resolves
   - any that obtains a certificate
   - any that publishes an MX

Eight hundred names being watched with three escalation triggers produces a handful of alerts a month. Twenty-two thousand names alerting on registration produces a stream nobody reads by week three.

Write down what you excluded, and why

The exclusions are the defensible part of the design, and they will be questioned after an incident involving something you excluded. A recorded rationale turns that conversation into a decision to revisit rather than an oversight to explain.

Knowledge check

Which question decides whether a permutation belongs on the watchlist?

Last reviewed