Skip to content

Systems and records

Creative scope change dataset: is this revision included or new scope?

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Classify each creative revision against the signed scope as it stood on the request date. It is included when it alters a listed deliverable within the remaining rounds, formats, channels and rights; it is a scope change when it adds a deliverable, round, format, channel, territory, term or asset needing new clearance. A usable dataset task links the baseline, the verbatim request, the rule applied, the decider and an independent outcome.

License terms

Creative scope change

This page covers creative scope change dataset: is this revision included or new scope.Illustrative

Key takeaways

  • Classify a revision by comparing the request with the scope baseline in force that day, not by the effort it takes.
  • Nine scope dimensions cover the common cases: deliverables, rounds, brief objective, formats, channels, territory and term, source assets, schedule and approval authority.
  • A proof decision records quality, not commercial scope, so the task keeps the proof decision and the scope classification as separate records.
  • A new channel, territory or term is a scope and rights event even when the creative file does not change.
  • Each task needs an outcome record from a separate source, such as a change-order closure or invoice line, not a status field.
  • Paired counterexamples with identical requests and different baselines keep a model from learning scope from wording.

How do you tell an included revision from a change in scope?

A creative revision is included when it changes something the signed scope already promised, inside the rounds, formats, channels and rights the scope names. It is a change in scope when it asks for something the scope never promised: another deliverable, an extra round, a new format or channel, a new territory or term of use, or source material that needs its own clearance.

The test is a comparison, not a judgment of effort. A request that takes ten minutes can be new scope, such as a vertical crop for a channel the estimate never listed. A request that takes two days can be included, such as reworking a headline concept the client rejected in round one. Teams that classify by effort or by how annoyed the studio feels produce labels nobody can audit later.

For a dataset, the classification has to be reproducible by someone who was not in the room. That means each task carries the scope baseline as it stood on the day of the request, the request in its own words, the rule that was applied, the person who decided, and the record of what was actually delivered and paid for. Without those links, a revision label is just an opinion attached to a file.

Which terms in the scope decide the classification?

Many disputes reduce to nine scope dimensions, and each has a clear included case and a clear scope change case. Design industry contract templates draw the same boundary in general terms. The AIGA Standard Form of Agreement for Design Services, for example, has the client pay additional charges for changes outside the scope of the services, lets the designer adjust the delivery schedule to accommodate them, and allows a new proposal when revisions grow substantially, with no work on the revised services until it is signed.

Use the table below as the classification rule set. Fill the right-hand column from the actual estimate, statement of work or proposal version in force on the request date, not from memory.

Which terms in the scope decide the classification?
Scope dimensionIncluded when the requestScope change when the requestBaseline record to cite
DeliverablesAlters a deliverable already listedAdds a deliverable, variant or language version not listedDeliverables list in the signed proposal or SOW
Revision roundsFalls within the remaining rounds for that deliverableNeeds a round beyond the stated countRound count per deliverable and the round log
Brief objectiveServes the objective in the approved brief versionChanges audience, message or campaign goalBrief version approved before production
Formats and sizesAdjusts a listed formatAdds aspect ratios, durations or cutdowns not listedFormat and spec sheet attached to the scope
Channels and mediaPlaces work in a listed channelAdds a channel such as paid social, broadcast or out-of-homeMedia plan or usage section
Territory and termStays inside the licensed territory and durationExtends territory, duration or exclusivity of useUsage and license clause
Source assetsUses assets already cleared for this useIntroduces stock, talent, music or fonts needing new clearanceAsset rights log for the project
ScheduleFits the agreed review windowsRequires rush work or reopens an approved milestoneSchedule and approval milestones
Approval authorityComes from the named client approverComes from a new stakeholder after sign-offApprover list in the brief or SOW

Why a proof status cannot settle the question

A proof decision records whether the work is acceptable, not whether the requested work was paid for. Proofing tools are built for the first question. Adobe Workfront, for example, offers decisions such as Approved, Approved with changes and Changes required. Its documentation says Approved with changes means the proof needs edits but the reviewer does not need to see the revision before it moves to the next stage, while Changes required means the reviewer needs another revision first.

None of those values says whether the requested change was in scope. A Changes required decision on round three of a two-round deliverable is a scope event; the same decision on round one is not. The overall status can also hide who asked for what. Workfront documents that a proof's status is driven by the worst-case participant, so one restrictive decision overrides several approvals, and the request that triggered the next round may come from a reviewer who was never named as an approver.

So the task pack keeps two separate decisions: the quality decision from the proofing record and the commercial classification from the scope record. A dataset that collapses them teaches a model that every rejection is a revision, which is exactly the wrong lesson for scope management.

When a rights condition turns an edit into new scope

Rights conditions can turn a request into a scope change even when no pixel moves. Asking to run an approved print ad as a social video, in a second country, or past the licensed term changes what the agency may deliver, because the cleared rights for stock images, talent, music and fonts usually follow the original use.

Ownership matters too. Under U.S. copyright law, a specially ordered or commissioned work is a work made for hire only if it falls into one of nine statutory categories and the parties sign a written agreement expressly saying so; the Copyright Office's Circular 30 sets out those conditions. Otherwise ownership depends on the contract and any assignment. A pack that does not record who owns each asset cannot say whether the agency may reuse it, license it, or even show it.

For image assets, embedded metadata helps but does not decide. The IPTC Photo Metadata Standard defines fields such as Copyright Notice, Rights Usage Terms and Licensor. They show what a file claims; the contract and the license documents remain the evidence.

  • New channel or medium for an approved asset: scope change and rights screen, even with no creative edits.
  • Approved with changes followed by a new stakeholder's comments: classify the new comments against the approver list before counting a round.
  • Swapping a cleared stock image for an uncleared one inside an included revision: the revision is included, the asset is not deliverable until cleared.
  • Client-supplied material: record the client's warranty or license for it, since the agency's clearance does not cover it.
  • Reopening an approved milestone: treat as a scope change unless the scope reserves a final correction pass.

What the rights screened scope and decision evidence task pack includes

The pack is a set of linked records for one revision request, built so a reviewer can rebuild the decision from evidence. Each record has required fields, a statement of what it proves, and a reject rule. If any required record is missing, the task is excluded from the dataset rather than filled with a guess.

At a glance

What the rights screened scope and decision evidence task pack includes

What the rights screened scope and decision evidence task pack includes
RecordRequired fieldsWhat it provesReject the task if
Scope baselineProposal or SOW id, version, signed date, deliverables, rounds, formats, channels, territory, termWhat was promised on the request dateOnly an unsigned draft exists
Brief versionBrief id, version, approval date, objective, audience, approverWhich objective the work had to serveBrief approval date is after the request
Revision requestRequest id, timestamp, requester role, channel, verbatim text, deliverable id, round numberWhat was asked, by whom and whenRequest text is paraphrased or undated
Proof decisionsProof id, version, reviewer role, decision value, decision timeThe quality judgment on each versionDecisions cannot be tied to a version
Scope classificationRule row applied, classification, reason, decider role, decision timeWhy the request was included or new scopeNo rule row or decider is recorded
Change orderChange order id, reference to the original proposal, added work, schedule effect, signature statusThat new scope was agreed before workWork was delivered before the signature with no written authorization
Rights screenAsset ids, owner, license source, permitted media, territory, term, expiryWhat may be delivered and reusedAny delivered asset lacks an owner or license record
Delivery recordFile ids, versions, hash or checksum, delivery date, recipient roleWhat actually shippedDelivered version differs from the approved version
Outcome checkInvoice line or change-order closure reference, client acceptance, dispute flagThat the classification held commerciallyOutcome is only a status field or a closing message

A worked case: the cutdown that arrived as a revision

This case is illustrative; no real client, campaign or agency is described. All counts and durations are invented for the example.

Starting state: the signed scope lists one 30-second video and two static banners, two revision rounds per deliverable, web and paid social in one country, and a 12-month usage term. Round one on the video is closed with a Changes required decision; round two is open.

  • Request: during round two the client's brand manager writes: "Please tighten the opening and give us a 6-second version for pre-roll." The request is logged verbatim with its timestamp and deliverable id.
  • Evidence linked: the scope baseline lists no 6-second format and no pre-roll channel; the brief version names the brand manager as approver; the music track's license covers the 30-second edit in the listed channels.
  • Rule applied: tightening the opening alters a listed deliverable within the remaining rounds, so it is included. The 6-second version adds a format and a channel, so it is a scope change, and the music license must be checked for the new edit and channel.
  • Decision: the producer splits the request into two tasks, records the classification and reason for each, and issues a change order that references the original proposal for the cutdown.
  • Outcome check: the change order is signed before work starts; the delivery record shows the tightened 30-second version and the new cutdown as separate files; the outcome record cites the change-order closure, not a status field.
  • Counterexample: the same request on a scope that lists a 6-second cutdown for pre-roll is entirely included. The text of the request is identical; only the baseline differs. Keep this pair in the dataset so a model cannot learn to classify from wording alone.

Proving the classification held after delivery

The decision is proved by the classification record citing a specific baseline row, and the outcome is proved by a commercial record that closes it. A project status of complete or a friendly final email proves neither. The AIGA agreement template allows invoices to include changes the client authorized by email or through a project management platform, which is why the authorization itself, not just the invoice, has to be linked.

At a glance

Proving the classification held after delivery

Validate every task with the same checks before it enters the dataset:

  • Chronology: the scope baseline and brief version are dated before the request; the classification is dated after the request and before delivery.
  • Linkage: every record carries the project id and deliverable id; the request id appears in the classification and, where present, the change order.
  • Version match: the delivered file version equals the last approved proof version.
  • Rights coverage: every delivered asset has an owner and a license that covers the channel, territory and term of the final use.
  • Outcome independence: the outcome record comes from a different system or person than the classification, such as finance or the client approver.
  • Label provenance: each task is marked as real and approved for release, or illustrative, never mixed.

Where each record lives, and which key joins them

In many studios the nine records sit in four or five systems that were never designed to agree, so the join key matters more than any single field. The scope baseline usually lives in a signed PDF or an e-signature platform; briefs, requests and proofs in a work management or proofing tool; the rights log in a digital asset manager or a spreadsheet; change orders and invoice lines in accounting. Email threads hold the authorizations nobody copied anywhere else.

Pick one project identifier and one deliverable identifier and write both onto every record before export. Where a system cannot hold them, keep a crosswalk table that maps its native ids to yours and record who built it. Timestamps should be converted to one time zone at export, because a request and a classification that look out of order are often just two systems writing local time.

Expect gaps. Early projects often have a proposal and a final invoice but no round log, and verbal approvals leave no record at all. Do not backfill those gaps from memory. Mark the task incomplete, keep it out of the release, and use the pattern of gaps to decide which systems need a field added for future projects. A smaller set of complete cases is worth more to a buyer, and to your own operations team, than a larger set where half the decisions are reconstructed guesses.

What to strip or hold back before a pack leaves the studio

Remove what the studio has no right to share, even when the decision logic is valuable. Client briefs and unreleased campaigns are often confidential under the client agreement. Talent releases, model releases and personal contact details are personal data. Third-party stock, music and fonts are licensed for a project, not for redistribution in a dataset.

A practical pattern is to keep the decision structure and replace protected content with stable placeholders: the deliverable type, format, channel and rights terms stay; the creative file becomes a hash and a description; names become roles. De-identification reduces privacy risk but does not give the studio a right to license client material, so the rights screen comes first and the redaction second.

Stop and narrow the pack when the client contract restricts any reuse of project records, when ownership of key assets is unclear, or when the only evidence of authorization is a phone call. Those tasks can stay in internal training material without entering a licensed dataset.

If your agency's project history could become a licensed dataset

Agencies and in-house studios with years of scoped projects hold exactly the decision records described here. Turning chosen business records into datasets that AI developers license is what SourceX does, and it carries the rights review, the contract, the preparation and the delivery. Buyers generally look for consistent history across years and a large number of records in one system, though rarer work can qualify with less, and every case is assessed on its own.

The studio approves the scope and terms and signs before a single file leaves its hands, and checking eligibility begins with a brief company assessment, with no upload required.

What does a brief change after production began force the producer to decide?

It forces a call on every deliverable the change touches, not one verdict on the request. Work already exists, other deliverables may inherit the changed element and the plan must be redrawn, so the right answer is often partial: continue some items, pause others, stop one, and proceed on the rest once conditions are written. A creative workflow scope change dataset that stores one label per request loses that structure.

Commercial production guidance makes the paperwork a precondition, with one exception. The Association of Independent Commercial Producers' post-production guidelines (2022 file, read through search excerpts) treat a change in project specifications, scope or schedule included, as an overage billable as a change order, and ask that overages be agreed before the work they cover begins. In paraphrase, they also treat a client representative's instruction to start incurring costs before signature as binding, to be confirmed in writing promptly, and they let a purchase order or contract override them. Hence one label below proceeds before signature.

Label each touched deliverable with the first rule that fits, using the nine-row rule set above for the baseline test:

  • R1. The deliverable does not use the changed element: CONTINUE_ON_BASELINE.
  • R2. The requester is not an approver named in the approved brief or statement of work: HOLD_FOR_AUTHORITY, with one written question to the named approver.
  • R3. The change removes the deliverable: STOP_AND_SETTLE. Mark it canceled, keep its logged time and versions, and settle under the cancellation terms.
  • R4. The change stays inside the signed baseline: CONTINUE_ON_BASELINE.
  • R5. The change brings in a use or asset the clearances on file do not cover: RECLEAR_BEFORE_DELIVERY. Work may continue; delivery waits.
  • R6. A written authorization states the added cost, or a cap, and the new dates: PROCEED_WITH_CONDITIONS, with those conditions stored on the case.
  • R7. Anything else: HOLD_DEPENDENTS until a change is signed, while unaffected work carries on.

What can a studio's production tracker prove about the change, and what must come from elsewhere?

A production tracker can prove what depended on the changed element, what work existed and what was paused or re-planned, but not the brief versions, the authority behind the change or the client's agreement to pay. We tested that on Kitsu, CGWire's open-source platform whose README says it lets studios "collaborate on your creative productions and manage your deliveries"; its project styles include commercial and motion design.

At commit 5099f60 of the Kitsu backend (28 October 2026), the schema defines 80 tables and 751 columns in the 54 files of its models folder. No table or column is named for a brief, an approver, a signature, a change order, an invoice or a price. Casting links record which assets are cast in which shots, time logs record a duration per person, task and day, and shots and assets carry standby and canceled statuses. Schedule versions can be locked and name the version they came from, but the interface warns that applying one "will override the current task values", so the pre-change plan survives only if someone saves it first.

Deletion is the export trap. Deleting a shot that has tasks only marks it canceled unless deletion is forced; a forced delete removes its tasks with their comments, previews and logged time, plus its version history, which Kitsu records when a frame range or name changes, and its casting links. A dropped deliverable should stay canceled.

Method: a table is each class mapped with db.Model plus each db.Table definition, columns include those inherited from the shared base class, and names were split at underscores and matched as whole words. The count shows what the software can store, not any studio's data:

What can a studio's production tracker prove about the change, and what must come from elsewhere?
Piece of a brief-change caseWhat the Kitsu schema holdsNames searched, whole wordsWhat a licensor adds from other records
Brief wording and its versionsNo brief table or column; wording could sit in one of 13 description columns (project, entity, task and others); the 4 version tables (entity snapshots and schedule versions) copy no descriptionbrief, briefs; columns named description; tables with the word version or versionsBoth brief versions with approval dates, from documents or email
Who asked, with what authority6 user roles, client and vendor among them; 16 column names use the word client or clients, all true-or-false visibility or permission switches; 0 approver, approval, authority or signature namesclient(s), approver, approval, approved, authority, authorized, signature, signedThe approver list from the brief or statement of work, and the requester's role on the change date
What depends on the changed element5 tables hold two or more references to production entities, among them entity_link (casting, with an occurrence count) and entity_concept_link; status_automation chains one task's status to another'sTables with two or more foreign keys to the entity table; the status_automation tableA dependency snapshot frozen on the change date, since casting links keep no history
Work already donetime_spent (date, duration, person, task); task retake_count, real_start_date and done_date; revision numbers on preview, working and output filestime_spent; duration, retake_count, revision, estimation, done_date, real_start_dateRates and the agreement's cancellation terms to price that work
Held, canceled or re-plannedEntity statuses standby, running, complete and canceled, plus canceled and casting-standby flags; schedule versions with locked, canceled and source-version fields and per-task start, due and estimateValues containing standby, cancel, hold, pause, suspend or wait; canceled, standby, locked; schedule and milestone tablesThe date each hold began: statuses keep the current value, and a shot update event carries the shot and project ids, not the changed values
What the client agreed to pay, and when0 change order, invoice, client contract, price, fee, quote or scope names; the 8 hits are staff-cost tables and fields (budget with revision and currency, budget_entry and person daily salaries, salary_scale) and an employee contract_typechange_order, invoice(s), contract(s), price(s), pricing, fee(s), quote(s), scope, budget, salaryThe change order or interim authorization, invoice lines and any dispute, from accounting and email

Seven deliverable calls from one illustrative campaign, scored by the rule order

Score each call against the rule order, then grade it from records someone else wrote later. The campaign is invented: a 30-second 3D product film of 14 shots, two 6-second cutdowns with some shots of their own and four key visuals, with the client's brand director named as approver in brief v2. A packshot asset is cast in 9 of the 14 shots and in every key visual. A script applies the rules to each row; every label is one of the six above.

Read the outcome after the project closes: OUTCOME_HELD when the settling record agrees with the call, OUTCOME_REVERSED when work was redone, credited or disputed, and OUTCOME_OPEN when the project had not closed by the cut-off date. Rows A1 and A2 are a twin pair from one change, so a model that holds everything, or nothing, fails one of them.

Seven deliverable calls from one illustrative campaign, scored by the rule order
Change event (illustrative)Deliverables it touchesLabel and deciding ruleHard negative to beatSettling record, written later
A1. Brief v3 from the brand director changes the product colorway after lighting beganThe 9 shots and 4 key visuals that cast the packshotHOLD_DEPENDENTS (R7): no written cost or dates yetKeep lighting those shots in the old colorwayChange order signature date against the first time logged on those shots after the change
A2. The same colorway changeThe 5 shots that do not cast the packshotCONTINUE_ON_BASELINE (R1)Pause all 14 shots because the brief changedTime logs and approvals on those shots continue with no gap, and no change order line names them
B. A regional marketing lead asks for a new end line in a review commentThe end-card shotHOLD_FOR_AUTHORITY (R2)Rework the end card from the comment aloneThe approver list in brief v2 and the brand director's written answer
C. The brand director confirms a new end line by email, with a cost cap and new dates, while the change order is draftedEnd cards in the film and both cutdownsPROCEED_WITH_CONDITIONS (R6)Hold until signature and miss the fixed launch, or proceed with no cap recordedThe email dated before the work, the signed change order and an invoice line within the cap
D. Brief v4 swaps library music for a commercial track, and the film is re-timedThe film and both cutdownsRECLEAR_BEFORE_DELIVERY (R5)Deliver once the re-timed cut is approved, treating the frame changes as an included revisionThe track license on file before the delivery date, and shot version history showing the new frame ranges
E. Brief v3.1 supplies the disclaimer text that brief v2 said the client would provideEnd cardsCONTINUE_ON_BASELINE (R4)Raise a change order because a new brief version arrivedBrief v2's placeholder line and an invoice with no added line, accepted without dispute
F. The brand director drops the second cutdown after work on its own shots beganThe second cutdownSTOP_AND_SETTLE (R3)Force-delete its shots, or finish the week's work firstCanceled status kept, time logged before the cancellation date, and the invoice or credit line that settles it

What should a producer and counsel confirm before a brief-change case is licensed?

Confirm each case's evidence before it leaves the studio, and drop any case that fails an item. A buyer specifying such a case set can use the rule order as the labeling instruction and this list as the acceptance test:

  • The approver list in force on the change date, with the document it came from.
  • Both dated brief versions, with changed lines marked by deliverable.
  • A casting or dependency snapshot frozen on the change date, not rebuilt at export.
  • Time logged before and after the change date on each touched deliverable, with person ids replaced by roles.
  • The schedule version locked before the change and the one applied after it.
  • The written authorization, its cost or cap and its date, compared with the first time logged after the change.
  • A clearance result for any new asset or use, from whoever owns clearance at the studio.
  • Counsel's reading of the client agreement's change, cancellation and confidentiality terms; this outline is general information, not legal advice.

Sources for this section

  • Kitsu's backend (cgwire/zou) at commit 5099f60, read as a git clone, defines 80 tables and 751 columns in the 54 Python files of its models folder; no table or column name uses the word brief, approver, approval, authority, signature, change order, invoice, price, fee, quote or scope, and the cost tables plan staff cost (budget, budget_entry, salary_scale). Source
  • Kitsu's entity model gives shots and assets the statuses standby, running, complete and canceled, keeps canceled and casting-standby flags, and models casting as many-to-many links between entities with an occurrence count. Source
  • Kitsu's shots service records a shot version when the shot's frame range or name changes, and its shot update event carries the shot id (the event helper adds the project id), not the changed values. Source
  • Kitsu marks a shot that has tasks as canceled instead of deleting it unless deletion is forced; a forced delete removes its tasks, version history and casting links. Source
  • Kitsu's forced task deletion removes the task's comments, preview files and time spent records. Source
  • Kitsu's README describes it as a web application to collaborate on creative productions and manage deliveries, written by CGWire for animation, VFX and video game studios. Source
  • Kitsu's English interface warns that applying a schedule version to the production will override the current task values. Source
  • AICP's post-production guidelines treat a change in project specifications, including scope or schedule, as an overage billable as a change order, and ask that overages be negotiated and executed before the work they cover begins; in paraphrase, a client representative's instruction to start incurring costs before signature is binding and should be confirmed in writing, and purchase order or contract terms can override the guidelines. Source

Frequently asked questions

Is a client asking for a different color an included revision?

Usually yes, if the deliverable is listed, a revision round remains and the change serves the approved brief. It becomes a scope change when it arrives after the last included round, reopens an approved milestone or comes from someone outside the named approvers. Record which of those conditions applied.

How many revision rounds should a creative scope include?

There is no standard number; the count is whatever the signed proposal says. What matters for the dataset is that the round count is written per deliverable and that every round is logged, so a later reviewer can see which request used which round.

Does a change order have to be signed before the work starts?

That depends on the contract. Design industry templates such as the AIGA agreement describe a change order the client reviews and signs, and allow a new proposal with no work until it is signed when revisions grow substantially. Record what authorization existed before work began.

Can an approved creative asset be reused for a new campaign?

Only if the owner and every license in it allow the new use. Stock, music, fonts and talent releases are usually limited by medium, territory and term, and client-owned work may need the client's consent. A new use is a scope and rights event even without edits.

What makes a scope decision useful as AI training data?

A useful case links the baseline, the verbatim request, the rule applied, the decider and an independent outcome record. Paired counterexamples, where the same request is classified differently under a different baseline, stop a model from learning to judge scope from wording alone.

Sources

  • The AIGA Standard Form of Agreement has the client pay additional charges for changes outside the scope of services, lets the designer adjust schedules for them, provides for a new proposal when revisions grow substantially, and lets invoices include changes authorized by email or a project management platform. Source
  • AIGA's 2022 agreement update describes a change order as a document acknowledging a client request outside the original scope, stating added time and cost, sent to the client for review and an authorized signature. Source
  • AIGA's pricing guidance says a fixed-fee proposal should be specific about what is and is not in scope, because requests can only be identified as out of scope when the scope was well defined. Source
  • Adobe Workfront documents that Approved with changes means a proof needs changes but the reviewer does not need to see the revision before the next stage, while Changes required means another revision must be seen first. Source
  • Adobe Workfront documents that a proof's overall status is driven by the worst-case participant's decision. Source
  • The U.S. Copyright Office's Circular 30 says a commissioned work is a work made for hire only if it falls in one of nine categories and a written agreement signed by the parties expressly says so. Source
  • The IPTC Photo Metadata Standard defines rights fields including Copyright Notice, Rights Usage Terms and Licensor. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies