Two ways to ship something that is not to specification

Both are agreements to accept something the drawing does not permit. The distinction between them is when it was agreed, and modelling either as a change order makes the released design history unreadable.

ERPShipped in segment S07 · 817 words
ONE OBJECT TYPE · SEVEN KINDS · ONE LIFECYCLEProblem reportsomething is wrongECRshould we change it?ECOwhat exactly changesECNtell everyoneReleasedatomicDeviationship off-spec, boundedWaiveraccept as-is, onceStop shiphalt, nowdo not alterthe designSeven kinds, one object type — because seven types means seven copies of the same attribute list drifting apart.

The distinction, stated first

A deviation is agreed in advance. Engineering permits shipping off-specification within stated bounds, for a stated quantity or period — three hundred units, or until the end of March, with the tolerance opened from ±0.05 to ±0.10.

A waiver accepts it afterwards. A batch was built outside specification, it exists, and somebody with the authority to say so accepts it for that specific quantity. Sometimes called a concession, and the word matters less than the timing.

Neither changes the design. That is the whole reason they are separate object kinds rather than a change order with a flag. A deviation says the drawing is still right and we are departing from it deliberately; an ECO says the drawing was wrong and here is the new one.

Why modelling them as changes breaks the history

If a deviation is an ECO with a flag, then reading the change history of a part produces a sequence in which some entries revised the design and others did not, distinguished only by a field somebody has to notice.

Six months later, answering *what is this part supposed to be* means reading every change and filtering. The answer is recoverable, and it takes an hour instead of a query — and an hour-long answer is one people stop asking for.

The released design history should be readable as a design history. A part at revision C has been through the changes that made it revision C; the deviations against it are attached to it and are not part of that sequence. That separation is what keeps both questions cheap.

The three share a lifecycle, an approval mechanism and an audit trail with the change kinds, because the machinery genuinely is the same. What differs is what they mean, and meaning is what the object kind records.

The failure it prevents

A supplier ships a batch with a surface finish outside the drawing tolerance. It is functionally fine, the programme needs the parts, and somebody senior agrees to take them.

In an organisation without a waiver object, that agreement is an email. The parts are used, the product ships, and eighteen months later a customer asks why units in that serial range have a finish the drawing does not permit. The email is in an archive belonging to somebody who has left.

The failure is not that the decision was wrong — it was a reasonable call made by the right person. The failure is that a defensible decision became indefensible for want of a record with a bounded quantity, an approver and a reason attached to the parts it concerned.

How it meets the rest of the product

Both are bounded objects. A deviation names the quantity or period it covers and expires; a waiver names the specific quantity it accepted. An unbounded deviation is a design change nobody wrote down, and the bound is what stops it becoming one.

They attach to the part at the revision they concern, so the question of which units were affected resolves through the same machinery as everything else. A deviation covering units 400 to 700 is answerable against a serial number.

Quality connects both ways. A nonconformance whose disposition is use-as-is produces a waiver rather than a note in a text field, and a recurring waiver on the same characteristic is a signal that the specification is wrong and an ECO is overdue.

Approvals carry recorded meanings, so a waiver approved by somebody who abstained on the engineering judgement and approved on the commercial one is expressible rather than flattened into a yes.

How it meets your ERP

Neither publishes as a structure change, because neither is one. Your ERP's bill of materials is unaffected by a deviation — the design has not moved, and pushing a temporary tolerance change into an item master would create a record that outlives the deviation itself.

What does matter at the seam is quantity. A deviation bounded at three hundred units needs somebody to know when three hundred have been consumed, and that count lives in your ERP. Manufacturing PLM holds the bound and reads the context; the counting happens where the transactions are.

Disposition on affected inventory is likewise recorded here and transacted there. The waiver records what was accepted, by whom, against which quantity; the stock movement happens in your ERP.

Where the boundary is

Manufacturing PLM does not enforce a deviation's bound. It records that three hundred units were permitted and surfaces the count where your ERP exposes it, but nothing stops a three hundred and first unit being built — that control belongs to manufacturing execution.

It also does not decide whether a deviation is acceptable. Whether a surface finish outside tolerance is functionally fine is an engineering judgement resting on things no system holds, and a tool that offered an opinion would be offering confidence rather than analysis.

Facts

DeviationAgreed in advance, bounded by quantity or period
WaiverAccepted afterwards, for a specific quantity
BothLeave the design unchanged — that is why they are separate kinds
Shared with changesLifecycle, approvals, audit trail
Attached toThe part at the revision they concern
Unbounded deviationA design change nobody wrote down
Recurring waiversA signal the specification is wrong and an ECO is overdue
Not enforcedThe bound — counting units belongs to execution

Frequently asked

What is the difference between a deviation and a waiver?

Timing. A deviation permits shipping off-specification in advance, bounded by quantity or period. A waiver accepts a departure after the fact, for a specific quantity already produced. Neither changes the design, which is what separates both of them from a change order.

Why not model them as change orders with a flag?

Because reading a part's change history then produces a sequence where some entries revised the design and some did not, distinguished only by a field somebody must notice. Answering what a part is supposed to be becomes an hour rather than a query.

Do they use the same approval process?

Yes — the same lifecycle, approval mechanism and audit trail as the change kinds, because that machinery genuinely is the same. What differs is what they mean, and the meaning is exactly what the separate object kind exists to record for later readers.

Why must a deviation be bounded?

Because an unbounded deviation is a design change nobody wrote down. The bound — three hundred units, or until the end of March — is what stops a temporary departure quietly becoming the permanent specification without ever passing through change control.

How does this connect to quality?

A nonconformance whose disposition is use-as-is produces a waiver rather than a note in a text field. In the other direction, a recurring waiver against the same characteristic is a strong signal that the specification is wrong and an engineering change is overdue.

Will it stop us exceeding a deviation's quantity?

No. Manufacturing PLM records that three hundred units were permitted and surfaces the consumed count where your ERP exposes it, but nothing prevents a three hundred and first being built. That enforcement belongs to manufacturing execution, where the units are actually counted.

Does a deviation change what our ERP holds?

No, because the design has not moved. Pushing a temporary tolerance change into an item master would create a record outliving the deviation itself. What matters at the seam is the quantity count, which lives in your ERP and is read rather than written.