Ask who may decide access
Mandatory access control introduces a system-defined policy that is not merely the resource owner’s freely adjustable sharing choice. In a document example, an owner’s grant is insufficient if an organizational policy forbids disclosure to the proposed reader. Mandatory does not mean that no administrator can ever change configuration. Separate authority to define policy, the component enforcing it and the choices available to ordinary users. The key distinction is between controls an owner can relax at discretion and controls that the owner’s grant alone cannot bypass. Actual policy systems differ; a single comparison of numeric labels is not a description of every implementation that belongs to this broad category.
Combine discretionary and central conditions
Suppose A shares document D with B. The discretionary sharing table allows B to read, but a teaching central policy labels D as level two and permits B to read only through level one. Under our explicitly chosen rule that an object’s label must not exceed the reader’s permitted level, the final answer is deny. The two conditions are not added as alternative ways to gain access; this model requires both. Write each condition in a separate table column. Also decide who may relabel D. If A may freely lower its central label, the intended restriction changes. Accepting a caller-supplied number as the authoritative central label would defeat the very contract that the model was introduced to demonstrate.
Distinguish the numeric model from real mechanisms
Our numeric labels illustrate how an additional policy constrains an owner’s grant. They are not an explanation of SELinux as merely comparing two integers. Real mechanisms can consider subject types, object types, operations and explicit policy rules. Linux Security Modules provide places for security checks, but which mechanisms and policies are active depends on the environment. Seeing a label in a display or placing a policy file on disk does not establish that the intended operation was denied. Verification must connect the active mechanism, applied policy, target state and actual attempted action. The model teaches the structure of layered decisions; it is not a claim that a policy was installed or enforced by a particular operating system.
A decision table is not complete enforcement
An API may contain the right policy table while another route reaches the same storage without that check. Defining policy and placing unavoidable enforcement at the right boundary are separate responsibilities. A privileged maintenance helper needs analysis of whose requests it accepts, not an assumption that the ordinary viewing screen covers it. Distinguish a mechanism that only records a proposed denial from one that actually blocks the operation. Test legitimate actions and deliberate rejections in an independent environment before applying an operational policy change. A high denial count is not itself a safety metric. The meaningful result is that permitted and forbidden operations match the intended specification across the relevant paths.
Separate policy administration from ordinary access
Actors allowed to change policy or labels have authority over the protection itself. A permitted reader need not also be a policy administrator. Record why a policy changed, which objects were affected, when it applied and which cases were tested, without indiscriminately recording document bodies or credentials. A denial may come from ordinary file permissions, an additional policy or an unrelated resource constraint. A generic permission-denied message alone does not prove that the mechanism under study enforced the intended rule. Finally, MAC is also used to abbreviate message authentication code. This access-control concept and a keyed data-authentication mechanism answer different questions despite sharing letters in English.
Exercise: read both permission columns
Under the model requiring both conditions, a true sharing grant and false central-policy result produce denial. A false sharing grant and true central-policy result also produce denial; satisfying the central condition does not invent an owner’s sharing grant. Finally, ask whether a denial record alone proves that no document body was returned. It does not: inspect the actual response and state change as well. Prepare one allowed case and one case rejected by each distinct condition to trace the composition. Explaining those cases moves the discussion away from whether a mechanism has a strong-sounding name and toward which policy actually blocks which actor’s operation in the tested environment.