Control register · six families rev 2026-11-16 · doc-set 03

Control register

Six families, described as procedures

NPX Exchange describes a set of security practices in its own materials: multi-factor authentication, secondary confirmation for sensitive actions, role-based tiered access, multi-party approval for high-risk internal actions, encryption in transit and at rest, and continuous risk monitoring with escalation. What follows is an engineering reading of each one.

Basis company description Independent test none cited Format does / does not / ask
Stated plainly. The control families below are the company's described practices, not independently verified here. We cite no audit report, no penetration test and no certification, because none was provided with the company's materials. A description of a control is evidence that someone has designed it — not evidence that it operates.
01

The register

01

Multi-factor authentication

+ What it does

Requires a second, independent credential beyond a password before a session is established. The engineering value is not the second factor itself but the separation: a password stolen from a breach or a phished page is insufficient on its own.

− What it does not cover

  • An attacker who also holds the second factor, for example through a relay that proxies a live session.
  • The account-recovery path, which is the usual way an enforced factor is bypassed.
  • What the authenticated user is permitted to do once inside.

? What to ask

  • Which factor types are accepted, and is any of them SMS?
  • Is the reset path as strict as the login path?
  • Can MFA be disabled, by whom, and is that event logged?
mfa
required · factors described: 2
reset path
described as reviewed
coverage
session establishment; not the recovery flow
02

Secondary confirmation for sensitive actions

+ What it does

Interrupts a specific high-consequence action — changing a withdrawal destination, altering a contact method, resetting credentials — with a fresh confirmation. It converts a single decision into two, with a gap in between where a mistake can be noticed.

− What it does not cover

  • Actions the operator does not classify as sensitive.
  • A user who confirms out of habit after repeated prompts.
  • Confirmation fatigue created by prompting too often.

? What to ask

  • Is there a published list of actions that trigger it?
  • Is the confirmation independent of the session being confirmed?
  • Is there a delay before a new destination becomes active?
trigger
sensitive action · fresh confirmation
context
displayed at the point of confirmation
coverage
classified actions only; the list is not published
03

Role-based tiered access

+ What it does

Maps a person's role to a fixed scope of permissions, so that authority is granted by function rather than by request. Tiering adds a second dimension: some actions require not just the right role but a higher tier within it.

− What it does not cover

  • Role accumulation — one person holding several roles whose combined scope exceeds any single role.
  • Permissions granted outside the role model as temporary exceptions.
  • Whether roles are reviewed as people change jobs.

? What to ask

  • How many tiers exist, and what action defines the highest one?
  • Who can change a role assignment, and is that change itself approved?
  • Is there a periodic access review, and what happens to findings?
model
role → scope, tiered by sensitivity
principle
least privilege, as described
coverage
assignments; review cadence not published
04

Multi-party approval for high-risk internal actions

+ What it does

Requires more than one person to agree before an internal action of significant consequence proceeds. Its value rests on independence: the approvers must not be the requester, and must not be able to approve a request they raised.

− What it does not cover

  • Collusion between approvers.
  • Approval given without reading the request — a signature that certifies nothing.
  • Speed: a quorum is slower than a single approver, and the difference is a deliberate cost.

? What to ask

  • How many approvers, and must they be in different functions?
  • Can one person hold two approval seats simultaneously?
  • Is the request content fixed at submission, or can it change before approval?
quorum
more than one approver, as described
independence
approver ≠ requester
coverage
internal high-risk actions; threshold not published
05

Encryption in transit and at rest

+ What it does

Protects data on the wire between client and service, and on storage media, so that an intercept or a stolen volume yields ciphertext rather than records. It is the least informative of the six on its own: a baseline, not a differentiator.

− What it does not cover

  • Compromise of an endpoint where data is decrypted and displayed.
  • Insider access to plaintext through legitimate permissions.
  • Metadata — who transacted, when, and for how much — which often survives encryption.
  • Key custody. Encrypting data with a key that any operator can retrieve changes the failure mode, not the risk.

? What to ask

  • Where are keys held, and who can use them without a second person?
  • Is encryption applied to backups and to derived stores, or only to primary databases?
  • Is transport encryption enforced, or merely supported?
scope
data in transit · data at rest
key handling
not detailed in the published documents
coverage
transport and storage; endpoint and metadata excluded
06

Continuous risk monitoring with escalation

+ What it does

Watches for behavior that departs from an expected pattern and routes what it finds to a person who must act. The second half carries the value: detection without an owner is telemetry.

− What it does not cover

  • Anything the rules do not describe. Novel behavior is, by definition, outside the pattern.
  • Prevention. Monitoring observes an action; it does not stop it.
  • Alert quality. A queue dominated by false positives trains its readers to ignore it.

? What to ask

  • What is the acknowledgement window, and who is on call to meet it?
  • Which escalation decisions are recorded, and where?
  • How is alert precision measured, and what is done with it?
signal
anomalous behavior, continuously evaluated
path
alert → triage → owner → escalation → record
coverage
detection and response; no timing figures published

Monitoring and escalation is treated in its own note →

02

What the six families add up to

Read individually, five of the six are ordinary — the default furniture of any financial system built in the last decade. Read together, they form a shape: every action that moves value passes through at least two independent checks, and every check emits a signal that someone is obligated to read.

That shape is what a reviewer should test. Not whether each component exists, but whether the composition holds under the cases where one of them fails — a stolen factor, an exhausted approver, a rule that never fires because the behavior it was written for has changed.

  • Redundancy. Does any single compromise clear two gates at once? A phished session that clears both authentication and re-confirmation is one failure, not two.
  • Independence. Are approvers structurally separate, or does one team hold both the request and the approval role?
  • Evidence. Can a dispute be resolved from records the operator cannot quietly edit?
  • Recovery. When a gate misfires, what is the path back — and does it bypass the gate it failed?

The gaps named here are not accusations. They are the questions the public description leaves standing, and they are answerable — an operator who answers them in writing has given a reader something more useful than an adjective.