A freeze window is a client saying "nothing changes during this period unless we personally
accept the risk." Keep those windows recorded where intake can find them, enforce them when
changes are proposed, and run the exception path when something genuinely can't wait.
RECORDING A FREEZE:
1. Capture the freeze as structured facts: client, start/end dates (explicit dates, not
"tax season" — resolve the phrase to dates with the client), what is frozen (all changes
vs. specific systems), what is exempt (typically security-critical patches and
break-fix — get this stated, not assumed), the client-side authority who can grant
exceptions, and who requested the freeze.
2. Store it where the desk's calendar checks read from — the client's KB/wiki record — and
confirm it's findable in the client's documentation by the client's name. A freeze nobody
can find protects nobody.
3. Freezes expire: every record carries its end date, and open-ended freezes are recorded
with a review date instead. Flag freezes past their end date for confirmation-or-removal
rather than deleting them unilaterally.
ENFORCEMENT (at intake / scheduling):
4. When a proposed change window is checked, match affected clients against active freeze
records. Overlap → the change is blocked from scheduling: post a plain-text note citing
the specific freeze (client, dates, scope, source record) and set the change back to the
requester.
5. Distinguish frozen from exempt: if the change matches the freeze's stated exemptions
(e.g. critical security patch), say so explicitly and route it onward as normal — but
cite the exemption clause, don't infer one.
EXCEPTION PATH:
6. An exception requires all of: a written business case for why it cannot wait until the
freeze ends, the risk assessment (change-risk-assessment) attached, and explicit
approval from the client-side authority named in the freeze record. Route it for approval through the system where available; otherwise document the client's written approval verbatim
in a note. Client silence, or approval from someone other than the named authority, is
not an exception.
7. Record granted exceptions on both the change ticket and the freeze record's history — a
freeze that leaks exceptions every week is really a calendar problem, and the pattern
should be visible.
Guardrails: the default answer during a freeze is no. The agent never grants exceptions —
it assembles the case and routes it to the named human authority. Never treat an emergency
change as an automatic freeze exception: break-glass work during a freeze still notifies the
client's named authority immediately, because the freeze is their risk posture, not the
MSP's. Resolve vague freeze phrases to explicit dates at recording time. If freeze records
may be incomplete or stale (no review in >6 months), say so when clearing a window rather
than asserting "no freeze exists." Plain-text notes; freeze citations name the source
record so the block is verifiable.
