Skip to content

Reading a finding

Every claim in the threat model answers three questions, and each has its own words. A finding can be open, confirmed and stale at once. Reading one word as another is how a report ends up cleaner than what was measured.

bravos guide states prints them:

READING A FINDING
open / mitigated / accepted
Does a control cover it? Open means no control is recorded; mitigated
names the control; accepted is a human's written decision to live with it,
never an agent's.
untested / confirmed / refuted
Has an authorized run tested it? Confirmed means a run reproduced it, with
evidence; refuted means a run showed it does not hold. Untested is neither
cleared nor proven.
verified / stale
Is the claim still true of today's code? Stale means the code under it
changed since it was last checked; bravos-recheck clears it.
Severity
Rated by impact on this product × reach × ease, with the reason written
down. Without .guardlink/intent.md, severities are defaults, not
calibrated.
unknown is not zero
When a tool is unavailable the briefing says "unknown", never 0. Nothing
is called clean unless a gate exited 0.
  • Coverage (open, mitigated, accepted) is the annotations themselves: @exposes, @mitigates, and a person’s @accepts. See the guardlink docs.
  • Testing (untested, confirmed, refuted) comes from guardlink’s hypothesis state, including ledger evidence and source confirmations. bravos-prove writes to it only through guardlink, and guardlink hypothesis next lists what to test next. GuardLink also reconciles a source @confirmed with matching asset/threat exposures; it does not reproduce the claim again. See source confirmation. Bravos checks kit, claim, source and native engine evidence before recording a run; a process exit or template-authored verdict alone cannot update a claim.
  • Currency (verified, stale) is guardlink’s record of the code each claim was checked against.

bravos watch shows all three side by side: open, high and above, confirmed, tested, coverage and stale.

A SAST finding joined to a claim is taint evidence for it. It does not confirm the claim, and the absence of one does not refute it. See Run SAST.