
Freeze windows: protect settlement, protect go-live
Trading has a cut‑off for a reason. After market close, you don’t “sneak in one more trade.” You protect control, price integrity, and downstream processing.
Delivery needs the same discipline.
In commodity trading, the most expensive outages rarely arrive as “big bang” failures. They arrive as a small late change: a posting tweak, a data mapping adjustment, a message format update. Then settlements jam, finance loses confidence, and the go‑live plan turns into a war room.
Freeze windows stop that. Not by banning progress, but by forcing clarity: what changes stop, when they stop, and who can approve exceptions.
The freeze window playbook
1) Define the freeze purpose in one sentence
Pick one. Write it at the top of the policy.
- “Protect month‑end close and settlement throughput.”
- “Protect cutover rehearsal and production stability.”
- “Protect holiday operations when coverage drops.”
If the purpose stays vague, the exceptions explode.
2) Set the calendar and the cut lines
A freeze is not a date. It’s a set of cut lines across systems.
List the windows that matter in trading:
- month‑end close (finance + settlements)
- exchange/market deadlines
- holiday coverage periods
- go‑live cutover + hypercare
Define cut lines by system:
- CTRM/ETRM production changes: ___ to ___
- ERP posting and interfaces: ___ to ___
- Market data feed changes: ___ to ___
- Identity/roles changes: ___ to ___
Keep it simple. If the window is complex, split it into two tiers (see below).
3) Classify changes into three tiers
This is where most teams fail. They say “freeze,” then keep changing anyway.
Tier 1 , Allowed (low risk, reversible)
- documentation updates
- runbook updates
- monitoring dashboards
- training materials
- defect triage and prioritisation (no prod change)
Tier 2 , Allowed with evidence (controlled change)
- config change behind feature flag
- report layout changes (no calculation logic)
- non-prod fixes that unblock testing
- patching with a tested rollback step
Tier 3 , Frozen (no exceptions without exec sign-off)
- schema changes
- interface contract changes (message format, API, file layouts)
- posting logic to ERP or settlement rules
- pricing/curve handling, valuation logic, risk engine changes
- access model changes that affect segregation of duties
Make Tier 3 short and blunt. Tier 3 equals “no surprises.”
4) Define the exception process like a trading limit breach
Exceptions happen. The problem is silent exceptions.
Exception gate (minimum):
- a named requester
- a named approver (business + IT)
- a rollback trigger (measurable)
- evidence pack link (tests, steps, comms)
- a start time and end time (no open‑ended work)
Approver set (typical for trading platforms):
- Business owner (Trading Ops / Settlements / Finance, depending on impact)
- IT service owner
- Change manager / CAB delegate
- Vendor delivery lead (if vendor touches the change)
If any name is missing, it’s not an exception. It’s scope creep in disguise.
5) Enforce it in the tooling, not in meetings
A freeze that relies on people remembering is not a freeze.
Hard controls:
- tag change tickets with “Freeze Tier 1/2/3”
- lock deployment permissions during Tier 3 window
- block merges to release branches without exception ID
- require evidence link before status changes to “Ready for Release”
This protects teams from pressure. It also protects vendors from “just push it.”
The “Market Close” metaphor (use it in steerco)
At market close, you don’t debate the rule. You follow the cut‑off.
During a freeze, treat Tier 3 as market close.
Phrase to use:
“Market close rules apply. If it touches Tier 3, it needs an exception pack and sign‑off.”
Outcomes, risks, mitigations
Short outcome
Less chaos in close week. Cleaner settlement throughput.
Risk: stakeholders feel blocked.
Mitigation: Tier 2 path with evidence and fast approvals.
Long outcome
Higher trust in go-live readiness because stability stays stable.
Risk: teams delay hard decisions until after freeze.
Mitigation: decision deadlines before freeze; publish the “no later than” dates.
Unexpected outcome
Vendor behaviour improves when access and releases tighten.
Risk: vendors argue it slows delivery.
Mitigation: show the alternative cost: rework, rollback, and reputational damage.
Five-step rollout this week
- Publish the calendar and cut lines (one page).
- Agree Tier 1/2/3 and lock Tier 3 list.
- Put the exception gate into your change tool (template + required fields).
- Enforce access controls and branch rules during the freeze.
- Run a 15‑minute weekly review: exceptions raised, exceptions approved, lessons logged.
Next step prompt
If you need a clear plan with owners and dates, build a Reset Plan with actions, owners, dates, and vendor accountabilities,then plug the freeze policy into it.
Three questions to tune this to your programme
- Which freeze hurts you most: month‑end close, holiday coverage, or cutover week?
- Which Tier 3 item causes the most damage when it changes late: ERP postings, interfaces, data model, or access controls?
- Who holds the final exception pen today: CFO, COO, Head of Ops, or CIO?

