SENTINEL SAFE MODEOperating layer
Preserve control. Do not go dark.
DEGRADED BY DESIGN. NOT BROKEN.
Under a severe incident the instinct to shut everything down destroys exactly what an investigation needs. Safe Mode contracts the estate into a protected operating core instead: identity, read access where safe, evidence capture and incident response remain, while dangerous and high-risk operations are held.

Illustrative render. The posture is real; this picture is not a live system.
- 01ObserveContinuous telemetry from every service and edge you connect.
- 02VerifyTelemetry checked at the gateway before it is trusted.
- 03DetectAnomaly, policy violation, and absent signal alike.
- 04ProtectPolicy enforced at the identity and tenant boundary.
- 05PreserveEvidence written down before it can be lost.
- 06RecoverContinuity coordinated, state restored, result verified.
Not a stage. The loop keeps running inside it.
Recover returns to Observe
Safe Mode is not a stage. The loop keeps running inside it, with high-risk operations held.
The posture, as an operator sees it
What stays, what is held, and who decides.
Command shows Safe Mode as one state at one scope: the reason it was engaged, what is still running inside it, what is held at the edge, and who may lift it.
Safe Mode engaged
Operational state: Safe Mode
- Tenant
- Site
- Deployment
- Incident
- Scope
- One site, the narrowest scope that contains the incident
- Reason
- Credential material in flux
- Engaged by
- An operator with sufficient authority
- Lifted by
- Not yet. An authorised operator decides, and gives a reason.
Active in the core9 groups
- Identity and sign-inKept
- Read accessKept
- Evidence capture and Ledger writesKept
- Evidence verification and legal holdKept
- Incident responseKept
- Preservation requestsKept
- Watchtower sweeps and Range runsKept
- Narrowing accessKept
- Remove or demote a memberRe-confirm
Held until lifted6 groups
- New and lifted containmentHeld
- Provider credential and allowlist changesHeld
- Protected-download grantsHeld
- Deployment create, rotate and patchHeld
- New sites, site release and verification issuanceHeld
- Invitations and role promotionHeld
A held operation is refused with its scope and the reason it is held under.
What remains active
It shrinks to what matters. The rest waits.
The core keeps what an investigation and a recovery need, lit and in reach. What could compound the incident is parked, dark and safe, until a person decides otherwise.
Kept. What an investigation and a recovery need stays active: six layers on one axis, each lit.
- KeptKept. What an investigation and a recovery need stays active: six layers on one axis, each lit.
- HeldHeld. What could compound the incident is parked, dark and safe, on the plate round the foot, until an operator lifts Safe Mode.
- DecisionDecision. One platinum light marks the only way back: a person lifts Safe Mode and the reason is recorded.

The same core taken apart along its axis. Six lit slabs rise on one axis: incident response, evidence capture, verification and legal hold, Watchtower sweeps and Range runs, read access, and identity and sign-in. Five low dark capsules sit parked on the plate round the foot: provider credential changes, new and lifted containment, protected-download grants, deployment lifecycle writes, and invitations and role promotion. The three states are kept, held and decision.
What it is not
Not an outage. Not a shutdown.
Safe Mode is easy to mistake for failure, and it is the opposite: a posture Sentinel chooses, with identity, evidence and response still running. Five things it is not, and the fact that replaces each.
Safe Mode is not An outage
Identity stays available and reads are not refused.
Evidence capture, incident response and the actions that narrow access keep running. What is held is the write that could compound the incident.
Safe Mode is not A shutdown
Nothing is switched off.
A held operation is refused with the scope and the reason it is held under, and it returns when Safe Mode is lifted. Evidence capture and Ledger writes are never among what it pauses.
Safe Mode is not Automatic
An operator decides.
No signal, heuristic or browser can engage it. A Shield signal can never reach the Safe Mode action on its own, and engaging it needs a reason and a role.
Safe Mode is not Global
It holds at the narrowest scope that contains the incident.
A tenant, a site, a deployment or an incident. A tenant-wide hold needs an owner, and the scope is stated on every refusal.
Safe Mode is not A recovery check
Lifting is an operator's recorded decision.
Lifting needs authority and a reason. No automated recovery verification gates it today: that gate is a named seam, not built. Recovery verification sits on closing the incident, as an operator's attestation.
Scope
Held at the narrowest scope that contains the incident.
A single compromised credential does not stop an organization working. Safe Mode is a scoped control, never a single switch: a tenant holds its sites, a site holds its deployments, and an incident holds its own surface. Every mutating action is classified in advance, and an action nobody classified is held.
Illustrative demonstration
Engage Safe Mode at
TenantUnaffected
SiteIn Safe Mode
DeploymentIn Safe Mode
IncidentUnaffected
Its own write surface, cascading to the site or deployment it names.
Site scope
One site and everything under it: its sentinels and its deployments.
- Tenant
- Every mutation across the tenant is subject to the matrix: every site and every deployment under it.
- Site
- One site and everything under it: its sentinels and its deployments.
- Deployment
- One deployment, and nothing around it.
- Incident
- The incident's own write surface. When the incident names a site or a deployment, the hold cascades to that scope too.
Engaged, worked, lifted
Recovery proceeds from a controlled state.
The posture is a place to work from. The incident is worked inside the core, recovery is verified by an operator, and the posture is lifted by a person with a reason.
Illustrative demonstration
One incident, held and recovered
Step 6 of 6
Safe Mode lifted
An authorised operator lifts the control and records why. Held operations are available again.
A severe incident is declared. Sentinel is operating normally and nothing is held yet.
An operator with the authority engages Safe Mode at a scope and states a reason. A tenant-wide scope needs an owner.
A request to change a provider credential arrives inside the scope. It is refused, with the scope and the reason it is held under. Nothing else stops.
Incident notes, evidence preservation, a preservation request to Continuity and a Range run all go ahead.
An operator completes the recovery-verification task, which closing the incident requires. It is an attestation, not an automated check.
An authorised operator lifts the control and records why. Held operations are available again.
Simulated states, written to show how Safe Mode behaves. Nothing here reads or changes a live system. No automated check gates a lift today: recovery verification is enforced when an incident is closed.
Signal path
What enters it, and what leaves.
Signals in
- An operator's decision, with sufficient authority
- The scope: a tenant, a site, a deployment or an incident
- The reason the operator gives, recorded with the engagement
Safe Mode
THE DELIBERATE RESTRICTIVE POSTURE THE LOOP CAN ENTER
Outputs
- High-risk operations held within the scope, each refused with its scope and reason
- Kept: identity, reads, evidence capture, incident response
- An audit entry for engagement and for lift, each with its reason
- Operational state shown as Safe Mode
Ten questions
Safe Mode, answered.
The questions every Sentinel system must answer, in the same order on every system page.
Safe Mode is a deliberate restrictive posture, not an outage. Under a severe incident Sentinel narrows what may happen while keeping what an investigation and a recovery need.
Writes and high-risk operations within a scope: a tenant, a site, a deployment or an incident's own surface.
An operator's decision to engage it, with a scope and a reason, and the incident it is linked to.
For each request in scope: does this operation remain available, or is it held?
- Kept: incident notes and progress, evidence verification and legal hold, Range runs, recovery confirmation.
- Held: provider credential changes, new network containment, protected-download grants, deployment lifecycle writes.
Hold dangerous and high-risk operations at the narrowest scope that contains the incident, while identity, reads, evidence capture and incident response keep running. Lifting needs an authorised operator and a recorded reason.
Design directionThere is no per-operation exception today: an operator who needs a held operation lifts Safe Mode first, which is fast and audited. Gating a lift on automated recovery verification is a named seam, not built.
It is not engaged by a signal: Safe Mode is an operator decision. It never pauses evidence capture or Ledger writes, and it is not a global switch.
The active Safe Mode scopes with their reasons, who engaged and lifted them, and the operational state shown as Safe Mode.
Guard enforces the holds, Evidence keeps recording throughout, Continuity recovers from the controlled state Safe Mode holds, and Range proves the enforcement.
Safe Mode is the answer to failure elsewhere: the estate contracts into a protected core rather than going dark. Lifting it requires authority and a recorded reason.
Every mutating handler declares whether it is kept or held, and a contract test fails if one does not. Range runs the Safe Mode state and enforcement scenarios.
Relations
Where it connects.
- Guard Sends toWhich operations to hold within the scope
- Evidence Sends toCapture and Ledger writes keep running
- Continuity Exchanges withRecovery proceeds from a controlled state
- Command Receives fromEngaged and lifted by operator authority
- Range Receives fromState and enforcement scenarios
- Sends to
- Receives from
- Exchanges with
- Outside Sentinel
Failure and health
When it fails, it says so.
In operation
Full operation
- Every operation available within its authority
When it fails or loses visibility
Safe Mode engaged
- High-risk operations held
- Identity, reads, evidence and response kept
- Lifted only with authority and a reason
Safe Mode states
- AvailableKept running inside Safe Mode.
- HeldRefused, with its scope and reason, until Safe Mode is lifted.
- Re-confirmAllowed after a fresh password confirmation.
- Safe Mode engaged (lit: on failure)Degraded deliberately at a named scope.
- Full operation (lit: in operation)
How its own health is verified
- Every mutating handler classified
- A contract test catches one that is not
- Range: Safe Mode enforcement scenarios
The Sentinel system family
