Account for the Period When Two Models Must Coexist
Tác Giả: OmniStack
Ngày đăng: 11/20/2026

On this page
- The period of coexistence is a management decision
- Use Parallel Change without pretending it removes the cost
- Find the work created by two live models
- Build the proposed dual-operation work breakdown
- Attach unit economics to the coexistence interval
- Decide what to fund, shorten or stop
- Set the handoff to the retirement gate
- FAQ
- What is a dual-operation work breakdown?
- What should a coexistence budget include during data migration?
- Does Parallel Change prove zero downtime?
- Is no observed traffic enough to retire the old model?
- Sources and limits
The period of coexistence is a management decision
A hypothetical retailer changes its fulfilment model from one order status field to separate payment, picking, dispatch and delivery states. The new structure is live for the order service, but warehouse screens, customer emails, finance exports and marketplace feeds still read the old field. The decision is not only whether the migration can be implemented. Leaders must decide whether to fund dual operation, narrow the scope, change the consumer sequence or defer the change.
That distinction matters because old and new meanings create an operating period. It has owners, exception paths, retained work, support procedures and evidence requirements. Treating it as a short technical overlap hides work that operations and finance may still carry after deployment.
Proposed template: ambiguity and decision table. Every proposed template below is a planning artefact, not an OmniStack deployment.
What breaks or becomes ambiguous | Why it happens | Decision required |
|---|---|---|
Warehouse staff see a status that does not match the new fulfilment state | The screen still interprets the old field while the order service writes the new representation | Fund an adapter or move the warehouse consumer before the new model affects fulfilment behaviour |
Finance exports produce different totals or categories | The new status semantics change report filters and historical grouping | Approve semantic mapping and finance acceptance criteria |
Marketplace updates are delayed or rejected | An external consumer cannot interpret the new value set | Retain a compatibility output with an accountable integration owner |
Support cannot explain an order state | Runbooks describe one model while alerts and screens expose two | Assign triage authority and update the operating procedure |
The buyer question is therefore a funding question: what must remain true while both representations are live? A useful answer names the work, the owner, the evidence needed to stop it and the consequence of leaving it open. A migration plan without those fields is an engineering schedule, not an operating decision.
For a multi-site operator, the unit of impact may be a warehouse, store, carrier lane or region. For a B2B SaaS platform, it may be a tenant, integration, account or processed event. The unit determines which exceptions matter and who can accept them. It also determines whether a coexistence interval protects a critical workflow or merely preserves an obsolete convenience.
Use Parallel Change without pretending it removes the cost
Parallel Change is a sequencing pattern in which a system expands to support old and new forms, migrates callers or data, and contracts by removing the old form.Martin Fowler's Parallel Change pattern is useful because producers and consumers rarely deploy together. The pattern separates risk across releases, but it does not make conversion free, reversible by default or free of operational work.
In the hypothetical retailer, expansion could add the new fulfilment fields while preserving the old status. The order service may write both forms. Readers can move one at a time, with compatibility logic translating the new representation for consumers that have not moved. Comparison jobs can identify divergence before the old path is removed.
The pattern has a hard limit: structural compatibility does not prove semantic compatibility. Two fields can accept valid values while expressing different business rules. A new delivery state may distinguish a carrier handoff that the old status collapsed into ‘shipped’. A report that maps both values to ‘shipped’ may pass a schema check and still fail finance's interpretation of the workflow.
Rollback also needs a specific design. Keeping the old field does not prove that the system can return safely to the old behaviour. New records may contain information the old model cannot represent. External messages may already have been sent. A backfill may have changed historical interpretation. The rollback question must identify which writes, messages, reports and operational actions can be reversed, and which can only be compensated.
PostgreSQL's ALTER TABLE documentation illustrates the boundary between structural change and data validity. PostgreSQL ALTER TABLE mechanisms, such as adding constraints without immediate validation, can reduce disruption in some cases, but they do not prove that historical rows satisfy the intended rule. Database behaviour must be checked against the relevant PostgreSQL major version if code-level examples are added. Local validation also cannot cover external reports, messages or third-party reads.
The correct implementation record is not ‘Parallel Change applied’. It is a sequence of acceptance decisions:
- Define the old and new meanings in business terms, not only column names.
- Identify every producer, reader, batch job, report and external integration.
- Introduce compatibility only where a named consumer still needs it.
- Compare representations and route differences to an accountable exception owner.
- Move consumers based on evidence, not on a release date.
- Approve contraction through a separate retirement gate.
A dedicated engineering team can maintain adapters, test fixtures and an evidence trail when the internal roadmap cannot absorb the coexistence work. The condition is clear: internal finance, warehouse and product owners must still control semantic acceptance. Engineering can implement the bridge. It cannot invent the business meaning.

Find the work created by two live models
The hidden workload in coexistence is the work required to keep two representations trustworthy at the same time. A useful discovery exercise should trace each workflow from write to operational action, rather than counting only schema changes or migration scripts.
- Duplicate write logic: the authoritative service may need to populate old and new forms, define failure handling and decide whether a partial write blocks the transaction.
- Compatibility reads: old readers may need a translation layer, with explicit behaviour for values that have no equivalent.
- Adapter logic: external feeds, warehouse screens and legacy exports may require different mappings. One generic adapter often hides materially different semantics.
- Comparison jobs: old and new outputs need comparison at a business-relevant grain, such as order, shipment, tenant or settlement batch.
- Exception handling: differences need a queue, severity rule, owner and disposition. A dashboard without an action path is observation, not control.
- Backfill and retry: historical populations need sampling, retry rules and treatment for records that cannot be converted cleanly.
- Reporting parity: finance and operations must test meaning, totals, filters and timing. Matching row counts is weak evidence if categories changed.
- Monitoring and alert routing: alerts must distinguish expected dual writes from divergence, missing updates and adapter failures.
- Support scripts and training: staff need a way to identify which representation is authoritative for each workflow.
- Test fixtures: supported combinations, unmappable values, delayed events and retry outcomes must remain in the test set until contraction.
These items should be separated into temporary, retained and discovery work. Temporary work should disappear after contraction. Retained work may remain because an external contract, archive or delayed consumer still requires it. Discovery work represents unresolved questions that can block retirement. Mixing the categories creates a false coexistence budget and leaves leaders unable to see what they are continuing to fund.
Data governance and data management practices are relevant here because the decision record needs ownership, definitions, lineage and retention reasons. The link is useful only if it leads to an actual governance action, such as assigning authority for a fulfilment definition or approving a finance mapping.
Proposed template: coexistence discovery register. For each consumer, record the old meaning, new meaning, data source, business owner, technical owner, failure mode, comparison method, exception destination and planned migration decision.
Proposed template: exception disposition record. Record the affected business unit, record or batch, observed difference, severity, decision, approver and evidence. ‘No action’ is not a valid status without a reason and an owner.
Proposed template: coexistence budget data migration worksheet. Separate temporary controls, retained obligations and unresolved discovery. Include the workflow unit, calculation source, business value protected and review point.

Build the proposed dual-operation work breakdown
A dual-operation work breakdown turns migration overlap into accountable work. Each row should explain why the item exists, which owner can approve its scope, what evidence permits closure and what happens if it remains open at the next checkpoint.
Proposed template: dual-operation work breakdown.
Work item | Why it exists during coexistence | Owner decision | Exit evidence |
|---|---|---|---|
Duplicate write path | Old and new consumers still need their model | Approve scope and failure handling | All required consumers accept the new path |
Reconciliation | Two representations may diverge | Define tolerated differences and exception owner | Open exceptions are closed, accepted or escalated |
Backfill follow-up | Historical data may need population after schema expansion | Approve retry, sampling and exception treatment | Acceptance criteria are met, not merely script exit |
Retained old records and jobs | Rollback dependencies or delayed consumers may still need them | Set retention reason and review point | Retirement gate approves deletion or continued retention |
Temporary support process | Operations will see mixed-state failures | Assign triage authority and escalation path | Runbooks and alerts no longer reference the old path |
A real work breakdown should add the affected workflow and system for each row. ‘Reconciliation’ is too broad if one job compares orders while another compares settlement batches. The owner needs to know which population is covered and which is not.
Use one row for a decision, not one row for a team. A row that says ‘engineering migration’ cannot expose whether finance has accepted changed report semantics or whether warehouse operations still relies on an old exception path. The work breakdown should survive a change in implementation team because the business obligation remains.
The strongest exit evidence is behavioural. A consumer has moved when it reads and acts on the new meaning, its failure path is tested, its reports have been accepted and its support procedure no longer depends on the old representation. A deployment timestamp proves none of those things.
Where the approach fails is equally specific. Coexistence is the wrong choice when the two models cannot express the same business event safely, when old and new actions would both trigger an irreversible external effect, or when no owner can accept semantic differences. It is also a poor choice when the change is so narrow that a controlled maintenance window can alter all consumers together without exposing the business to a mixed state. Parallel Change should not become a reflex that preserves every legacy path indefinitely.
For legacy platforms with many readers, a system modernisation and migration approach can help structure the technical sequence. The decision still belongs to the organisation that owns the workflow. Modernisation is not permission to leave an unowned compatibility layer running.
Proposed template: decision checkpoint. At each checkpoint, ask whether the coexistence interval is protecting a required consumer, whether the temporary work is reducing a named risk, whether retained work has an approved reason and whether the next consumer sequence is still the safest route.
Proposed template: failure rehearsal record. Test a failed duplicate write, a stale compatibility read, an unmappable value, a backfill retry and an external consumer rejection. Record the operational action, not only the system error.
Attach unit economics to the coexistence interval
FinOps unit economics provides a disciplined way to connect coexistence work to business value: document the calculation sources, link the work to the workflow it protects and keep temporary, retained and discovery costs separate. It does not justify an invented multiplier or a universal budget percentage.
Choose a unit that reflects the change. An order platform may use an order, shipment, refund or marketplace listing. A B2B SaaS platform may use a tenant, integration, account or processed event. A multi-site operation may use a warehouse, store, region or carrier lane. The unit is a measurement boundary, not a claim that every unit has the same value.
- Temporary work: dual writes, comparison jobs, interim dashboards, extra release coordination and short-lived support procedures.
- Retained work: archived old records, compatibility exports for a contractual integration and manual exception review that remains necessary after the main migration.
- Discovery work: identifying unknown consumers, unresolved finance definitions, undocumented warehouse exceptions and uncertain rollback dependencies.
- Evidence source: the repository, query, dashboard, ticket, report owner or operational record that supports the calculation.
- Business value protected: the workflow that would be disrupted if the control disappeared before the consumer moved.
A coexistence budget should show the work against the unit and the decision point. A proposed worksheet might contain: workflow unit, affected population, temporary control, retained obligation, discovery question, source of calculation, accountable owner, review point and exit evidence.
Do not turn this into a false saving claim. If a comparison job consumes infrastructure, engineering attention and operations review, those are real inputs even when they prevent a larger incident. If a compatibility export remains for a contractual consumer, it is retained work, not a temporary item that can be marked complete because the new database schema is deployed.
Cloud usage can make temporary controls visible, but infrastructure telemetry alone is insufficient. A cost dashboard may show the compute used by reconciliation while missing the finance analyst's review or the support team's manual disposition. The cloud cost control guidance for SaaS teams is relevant when it is used to connect technical consumption to an operating unit and an owner, not when it reduces the exercise to infrastructure spend.
For APAC leaders, the useful question is not ‘what percentage should migration cost?’ It is ‘which business activity requires this control, what evidence says the activity is safe on the new model, and who can approve removal?’ That question produces a budget that can be challenged without pretending that all work has a single financial conversion.
Proposed template: unit economics calculation note. State the unit, source systems, included work, excluded work, period covered, assumptions and approver. Record uncertainty as an open discovery item rather than burying it in a blended figure.
When data pipelines or reporting consumers are part of the change, data engineering and analytics capability may be relevant to building comparison and lineage controls. The internal data owner must still define what constitutes an acceptable result.
Decide what to fund, shorten or stop
The work breakdown gives leaders four legitimate decisions: continue dual operation, sequence lagging consumers, remove low-value compatibility or pause the change until a blocker is resolved. It prevents ‘keep both paths for safety’ from becoming an indefinite operating policy.
Imagine a hypothetical approval meeting for a retailer. Engineering reports that the new fulfilment model is authoritative for the order service, but a marketplace feed still requires the old status. Finance has accepted the new internal report semantics, while warehouse operations has three unresolved exception cases. The right decision is not a blanket approval or rejection.
Proposed template: funding decision table.
Decision | Use it when | Required evidence | Consequence |
|---|---|---|---|
Continue dual operation | A required consumer still depends on the old meaning and the bridge is controlled | Named owner, monitored divergence and a review point | Fund temporary and retained work explicitly |
Sequence lagging consumers | The technical model is ready but a consumer lacks an adapter, test or business acceptance | Consumer-specific blocker and migration plan | Move the lagging workflow without expanding compatibility everywhere |
Remove low-value compatibility | A path has no approved business reason and evidence shows it is not required | Owner approval, consumer inventory and support confirmation | Reduce retained work and update runbooks |
Pause the change | Semantic differences, irreversible effects or unresolved accountability make the bridge unsafe | Documented blocker and decision to narrow or redesign | Stop adding consumers to a model that cannot yet be governed |
In the hypothetical meeting, engineering should say which failures the bridge prevents and which it cannot prevent. Finance should state which report or control has accepted the changed meaning. Operations should identify the staff action required when the two models disagree. The executive owner should decide whether the protected workflow justifies continued coexistence.
This is where delivery ownership matters. Assigning people to tickets is different from asking a dedicated engineering team to own a defined technical outcome, maintain continuity and expose blockers through the organisation's roadmap. OmniStack may be considered for the second model when a company wants one continuous team to carry the technical work across migration, testing and operations. Hire in house when the unresolved questions depend mainly on internal finance policy, warehouse authority, undocumented business exceptions or permanent domain ownership.
The distinction is not a label. A team that owns the outcome should maintain the work breakdown, keep evidence current, raise an exit blocker before it becomes an incident and remain accountable for the technical handoff. A team assigned only to implementation can still be useful, but it should not be described as owning the business result when the organisation has to coordinate every dependency.
Set the handoff to the retirement gate
Coexistence budgeting and retirement approval are separate decisions. The old path should be removed only after required consumers have moved, exceptions have been dispositioned, retained data has an approved reason, rollback dependencies have been resolved and an accountable owner has approved contraction.
A symptom of weak retirement control is a dashboard showing no observed reads from the old field, followed by a delayed finance batch or external integration failure after deletion. No traffic during an observation window is weak evidence when batch jobs, delayed reports and third-party consumers do not run inside that window.
The retirement gate should review the actual evidence set:
- Consumer inventory is complete enough to support the deletion decision.
- Required consumers interpret the new model and have accepted its failure behaviour.
- Open reconciliation exceptions are closed, accepted or escalated by named owners.
- Backfill acceptance criteria are met, including treatment of records that could not be converted.
- Reports and exports have passed semantic parity review where parity is required.
- Runbooks, alerts, support scripts and training no longer direct staff to the old path.
- Any retained storage or compatibility job has a documented reason, owner and review point.
- Rollback and compensation decisions are recorded for effects that cannot be reversed.
The gate can approve deletion, approve continued retention or reject contraction. Continued retention is not a failure if the reason is real and funded. It becomes a failure when the old path remains because nobody owns the decision.
Proposed template: retirement decision record. Include the old representation, deletion scope, consumers checked, evidence links, unresolved exceptions, retained obligations, approver, implementation window and post-removal observation plan.
For a multi-site operator, the next action is to name the workflow unit and create the first dual-operation work breakdown before approving further migration scope. For a CTO or CIO, the decision is whether the organisation has the owners and evidence discipline to run two models safely. If it does not, narrow the change or stop it before compatibility logic becomes permanent infrastructure.
Put the retirement decision after the evidence, not before it. The next checkpoint should approve one of three actions: move the next named consumer, fund a specific blocker, or reject contraction and record why.
FAQ
What is a dual-operation work breakdown?
A dual-operation work breakdown is a proposed record of the temporary, retained and discovery work required while old and new data models remain live. Each item should name its purpose, decision owner and exit evidence.
What should a coexistence budget include during data migration?
A coexistence budget should separate temporary controls, retained obligations and unresolved discovery. It should link each item to a workflow unit such as an order, shipment, tenant, integration, warehouse or settlement batch, and document the source of each calculation.
Does Parallel Change prove zero downtime?
No. Parallel Change provides a sequencing mechanism for expanding, migrating and contracting a system. It does not prove zero downtime, prove reversibility or establish that old and new meanings are semantically compatible.
Is no observed traffic enough to retire the old model?
No. Delayed reports, batch jobs and external integrations may not run during the observation window. Retirement requires a consumer inventory, business acceptance, exception disposition, runbook review and an accountable approval.
Sources and limits
These sources support mechanisms only. They are not business acceptance records and do not replace local semantic review, owner approval or retirement evidence.
- Martin Fowler, Parallel Change: supports the expand, migrate and contract sequencing pattern.
- FinOps Foundation, Unit Economics: supports linking cost to business value and documenting calculation sources.
- PostgreSQL documentation, ALTER TABLE: supports the distinction between structural change mechanisms and validation of data. Check behaviour against the relevant major version before adding code-level guidance.

