Checks and assurance
A check (a predicate in the protocol) is a named question with a yes, no or unknown answer. A CA can only ask about checks in the registry, and only the ones the device owner granted.
The registry
This table is built from registry/v0.json, the same file the agent and the
CA read.
| Check | What a yes means | Parameter | Best on macOS | Best on Linux |
|---|---|---|---|---|
disk_encrypted | Startup disk is encrypted. | agent-reported | agent-reported | |
os_patched_within | No operating system updates are more than {n} days overdue. | 7d 14d 30d 60d 90d | agent-reported | agent-reported |
sip_enabled | System Integrity Protection is on. | hardware-backed | – | |
secure_boot | Secure boot was enforced at startup. | hardware-backed | hardware-backed | |
firewall_enabled | Firewall blocks unexpected incoming connections. | agent-reported | agent-reported | |
screen_lock_within | Screen requires a password within {n} minutes of inactivity. | 1m 5m 10m 15m | agent-reported | agent-reported |
auto_updates_on | Security updates are installed automatically. | agent-reported | agent-reported | |
kexts_disallowed | Third-party kernel extensions are not allowed or loaded. | hardware-backed | – | |
ssh_key_hardware_bound | Device key is hardware-protected and cannot be copied. | hardware-backed | hardware-backed | |
kernel_lockdown | Kernel lockdown blocks changes to the running kernel. | – | agent-reported | |
edr_present | Recognised security software is running. | agent-reported | agent-reported | |
os_supported | Operating system still receives security updates. | hardware-backed | agent-reported | |
restarted_within | The computer has restarted within the last {n} days. | 1d 7d 14d 30d | agent-reported | hardware-backed |
no_restart_pending | No installed update is waiting for a restart to take effect. | agent-reported | agent-reported | |
no_known_bad_binaries | No program measured since the last restart matches a known malicious file hash. | – | hardware-backed | |
edr_assessment_at_least | Your endpoint security tool rates this device at least {n}. | medium high | agent-relayed | agent-relayed |
Parameters come from fixed buckets, so a CA cannot probe for an exact value.
For example, os_patched_within accepts 7d, 14d, 30d, 60d or 90d.
Using checks in policy
Each environment lists the checks it requires:
[[environment]]
name = "staging"
principals = ["alice", "bob"]
require = ["disk_encrypted", "firewall_enabled"]
[[environment]]
name = "production"
principals = ["alice"]
require = ["disk_encrypted", "os_patched_within:14d", "firewall_enabled",
"screen_lock_within:5m"]
Assurance levels
Every answer carries an assurance level. The CA works the level out for itself and never trusts the label a device writes.
| Level | Meaning |
|---|---|
agent-reported |
The Device Passport agent on this enrolled device says so. Fresh, bound to this CA, and signed by the device key. Root on the device could lie. |
agent-relayed |
The agent passes on a verdict that a third party signed, such as an EDR tool's assessment. |
hardware-backed |
The answer rests on evidence signed by hardware (a TPM, or Apple's attestation) and checked by an evidence issuer. This is the job of the attestation service. |
The table's last two columns show the best level each check can reach on each platform. A dash means the check does not apply there.
Agent-reported checks catch honest mistakes and unenrolled machines. They do not catch a device owner with root who lies. The threat model covers this in full, and you should read it before you rely on a check for anything important.