Attacker Cost Attributes

## What an annotation is An attack tree decomposes an attacker goal into the steps that achieve it, with leaves that are concrete actions. An **annotation** is a value written onto a node describing what that action demands of the attacker. The mistake people make on their first tree is to write one value — usually dollars — as if attacker effort had a single currency. It does not, and the reason to annotate at all is to support a comparison between branches that only survives if the annotation keeps the shape of the difficulty. ## The working attribute set A practical set, small enough to be filled in consistently: | Attribute | What it records | |---|---| | Monetary cost | Cash the attacker must spend: hardware, a purchased service, a bribe | | Skill level | Rarity of the expertise required, not hours worked | | Elapsed time | Wall-clock time from starting the step to holding its result | | Required access | The position the attacker must already occupy: on the local network, an authenticated low-privilege tenant, an employee with a caseload, N colluding insiders | | Equipment | Specialised kit that is not simply bought with money — restricted, bulky, or export-controlled | | Detectability | How likely the defender notices, and whether before or only after the step completes | Teams sometimes add legal or physical risk to the attacker, and repeatability (works once versus works at scale). Whatever the set, fix it for the whole programme so two trees can be read side by side. ## Why the attributes must not be collapsed **Equipment and skill move independently.** Consider a hotel chain's tree, branch "clone a guest room credential in the corridor". The equipment may be an inexpensive handheld device; the skill to use it against that particular lock protocol is rare; required access is two unobserved minutes outside a door; detectability at the moment of use is close to nil. Write "cost: low" and you have told the reader something false about why this branch has not been exploited at scale — the scarce ingredient is skill. That matters because skill is the attribute with the shortest half-life: the day someone packages the technique into a point-and-click tool, the skill annotation collapses while every other value on the node is unchanged. Attributes you have kept separate can be revised separately. **Required access has no linear price.** A crypto exchange moves funds from cold storage only with three of five signer approvals. The branch "obtain three signer approvals by collusion or coercion" is not usefully annotated "$2,000,000". The second signer costs more than the first because each recruitment raises the chance of exposure, and at least one signer may refuse at any price. "Required access: 3 of 5 distinct signers, concurrently; elapsed time: months; detectability: rises with each approach" tells a reviewer what to do about it — split the quorum across jurisdictions, or add an out-of-band delay — which a dollar figure never would. **Time is not money here.** Elapsed time constrains the attacker against your rotation schedules, expiry windows and audit cycles, and it is what a slow branch trades away to buy stealth. A step that costs nothing and takes nine months is a very different threat from one that costs a little and takes an afternoon. ## Filling values in honestly - **Unknown is not zero.** Zero means free, and a reader will treat it that way. Write "unknown" and treat it as a research task; a tree with three unknowns on one branch is itself a finding. - **Annotate leaves first.** Concrete actions are the only things anyone can estimate; interior nodes get their values from their children. - **Record the basis.** A value someone can trace to a red-team's actual hours is worth arguing about; one with no provenance quietly becomes fact. - **Keep the scale consistent per attribute.** Skill in three or four named bands, elapsed time in orders of magnitude (hours / days / months), required access as a named position rather than a score. ## Common failure modes Averaging the attributes into one index, which destroys exactly the information the vector was carrying. Treating equipment and money as the same column, so restricted kit looks purchasable. Annotating only the root, where no one can check anything. Assuming money substitutes for every other attribute — it buys equipment reliably, skill slowly, insider position sometimes, and elapsed time almost never.