Skip to content

Barrier Coverage Matrix

Coverage-mapping tool — instantiates Adaptive Barrier-Circumvention Response

A cross-tabulation of control layers against variant classes and contexts that marks demonstrated coverage apart from unknown, stale, correlated, or merely-inferred coverage — making uncovered cells and shared blind spots visible before escape finds them.

"We have five controls" says nothing about which variants each control actually stops, or whether the five are really one control wearing five badges. Barrier Coverage Matrix is the artifact that converts a pile of controls into a legible map: control layers on one axis, variant classes and contexts on the other, and in each cell an honest statement of engagement and its evidence. Its defining discipline — the thing that separates it from a reassuring inventory — is that it records demonstrated coverage apart from unknown, stale, inferred, and correlated coverage, so the blanks and the false diversity become visible cells rather than silent optimism. It is the loop's static reference map: not a model of dynamics, not a live sensor, but the shared picture everyone points at.

Example

A financial-crime team maps its controls against fraud typologies. Rows are control layers — device fingerprinting, velocity rules, sanctions/PEP screening, manual review, an ML risk score. Columns are variant classes — account takeover, synthetic identity, mule networks, authorized-push-payment fraud — split by context such as new versus established account and channel. Each cell answers three things: does this layer engage this class, under what conditions, and on what evidence?

Marking the evidence is where the matrix earns its keep. Demonstrated (tested last quarter) is kept distinct from inferred (a vendor's claim), stale (validated a year ago, before the typology shifted), unknown (never tested), and correlated (two layers that both rest on the same device signal). Filled in this way, the matrix surfaces two actionable gaps: synthetic identity in the new-account context has only inferred coverage everywhere, and three "independent" layers all lean on the same identity-verification vendor — a shared blind spot dressed as defense in depth. Neither gap is visible in a headcount of controls.

How it works

  • Make the axes the specification. Defining the protection claim and the relevant variant classes and contexts is the first step — the columns state what the barrier set is meant to cover.
  • Record engagement per cell, with conditions. Which layer engages which class, and under what operating conditions, not a blanket yes.
  • Mark evidence status separately. Unknown, stale, correlated, and inferred coverage each get their own mark, never collapsed into "covered."
  • Re-review on change. The matrix is refreshed whenever composition, evidence, or control design shifts — a living document, not a launch artifact.

Tuning parameters

  • Variant-class granularity — how finely classes are split. Finer splitting reveals conditional gaps but explodes the matrix's size.
  • Evidence-status taxonomy — how many coverage states are distinguished. A richer taxonomy resists false assurance but is heavier to maintain.
  • Context dimensionality — how many contexts become separate cells. More context catches conditional coverage gaps; more cells go stale between reviews.
  • Freshness threshold — when demonstrated decays to stale. Tighter keeps the map honest; looser keeps it cheap.
  • Correlation marking — whether shared dependencies are flagged in-cell. Flagging surfaces common-mode risk but requires an independence analysis as input.

When it helps, and when it misleads

Its strength is making absence and false diversity visible — the uncovered cell and the three-layers-one-vendor cell — exactly the gaps a count of controls conceals. As the shared reference the rest of the loop annotates, it is where coverage arguments get settled with evidence instead of assertion.

Its failure mode is false assurance: a wall of green cells invites the belief that a claim of coverage is protection, when a cell is only a statement about the map, not the territory.[1] Sloppy taxonomy lets "inferred" quietly read as "real," and run backwards the matrix gets filled to look complete rather than to find gaps. It is also a snapshot — it does not detect change, so an un-reviewed matrix rots into fiction. The discipline is to keep unknown, stale, and inferred visibly distinct, re-review on every change, and treat a correlated cell as a gap until proven otherwise.

How it implements the components

  • protective_barrier_and_target_specification — the matrix's axes state what each barrier is meant to prevent, which target features it engages, and which classes and contexts the protection claim covers.
  • variant_specific_coverage_profile — the body of the matrix is the per-variant coverage estimate, carrying evidence status, cross-context differences, and correlated-failure flags.

It flags correlated cells but does not test whether they truly fail together (that's Common-Mode Escape Review); it does not detect live composition change (that's the Variant-Composition Surveillance Dashboard) or decide what to change (that's Champion–Challenger Barrier Revalidation and the rotation protocol). This tool records the coverage claim; siblings test, detect, and act on it.

Notes

A correlation flag in the matrix is a suspicion, not a proven common mode: the matrix notices that layers share a dependency, but confirming they would actually fail together against a real variant class is the job of Common-Mode Escape Review. Treating a flag as a verdict over-reacts; treating a green cell as tested protection under-reacts — the matrix's whole value is keeping those two states apart.

References

[1] Korzybski's dictum that "the map is not the territory" is the right caution for a coverage matrix: a filled cell is a claim about protection, evidenced or not, and an adaptive population escapes through the territory, not the map — which is why the evidence-status marks matter more than the coverage marks. ↩