Skip to content

Rights and contracts

Dataset department exclusion scope: reason, date and register

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Dataset department exclusion scope is settled by three written decisions made before any filter runs. The reason for the exclusion decides whether the boundary follows the department's work, its people or its subject matter. A stated date fixes which version of the department counts, because units get renamed, merged and retired. The department's own register then tests the export. In OpenStack's governance files, about half of 2018's team repositories had changed status by 2026.

Record timeline

Dataset department exclusion

This page covers dataset department exclusion scope: reason, date and register.Illustrative

Key takeaways

  • A department is a label on an org chart, not a field on most records, so the exclusion has to be rewritten as a test over record attributes before anyone runs it.
  • The reason picks the test: privilege and HR reasons follow subject matter and participants, contract and regulatory reasons follow the work the unit did, and commercial reasons usually follow ownership.
  • Name the date basis. Of 831 repositories OpenStack listed under a team in October 2018, 422 were under the same team in October 2026.
  • Test completeness against an independent list the department keeps, in both directions: none of its register items in the export, and every excluded record explained by the approved reason.
  • Redaction is no substitute: masking names changes identifiability, not the privilege, contract or regulatory reason the department was excluded for.

What does a department exclusion actually have to decide?

It has to decide which records leave the company, not which team is out of favor. A request such as "leave out Legal" or "exclude the People team" names a unit on an org chart, but most business systems file records by queue, folder, mailbox, account or project. The request becomes an exclusion only when someone rewrites it as a test that a query can run and a reviewer can check.

That rewrite is a rights decision, and it needs a record. The record names who decides (the person authorized to license the company's data, with the department head and counsel for anything uncertain), which document governs (the signed scope that will name the dataset), what evidence settles each boundary question, what stays out, and how the exclusion reaches the delivery files. This article builds that record: a reason-to-test table, a check on how unit maps drift, a permission matrix by record class, a two-way completeness test and a review log.

A useful answer is one a stranger could rerun: the stated reason, the date basis, the extraction rule with its version, the department's own register as a reference list, and counts that reconcile. This is general information, not legal advice; your own counsel resolves authority and scope questions on your facts.

Which boundary test does the reason for the exclusion call for?

The reason decides the test. A department excluded because its files hold legal advice needs a boundary that follows content and participants wherever they sit. A department excluded because one customer contract binds its work needs a boundary that follows the work done under that contract, including records filed in other teams' systems.

Regulation already draws unit lines this way. Under the HIPAA rules for a hybrid entity, one legal entity designates health care components, and 45 CFR 164.105 says such a component "also may include a component only to the extent that it performs covered functions." The same section requires the entity to stop the component from sharing protected health information with the rest of the organization where the rules would prohibit it "if the health care component and the other component were separate and distinct legal entities." The line follows function and information flow, not the org chart. That is background on how one regulation draws a unit boundary, not a view on whether HIPAA applies to your records.

The table pairs common reasons with the boundary each one implies. Counsel confirms the reason and the test before any rule is written:

Which boundary test does the reason for the exclusion call for?
Reason given for the exclusionBoundary followsRecords that usually slip past a queue filterEvidence that settles it
Legal advice or privileged materialContent and participants: in-house counsel, legal staff, outside firmsAdvice forwarded into project threads; tickets opened by lawyers in shared queuesCounsel's written screen rule and the matter list
HR and employee mattersSubject matter about named employees, wherever it is filedIT or facilities tickets raised for an HR case; payroll questions in finance queuesThe HR case register with linked record IDs
A customer or government contract that binds one unitWork performed and data received under that contractShared-service tickets for that customer handled by other teamsThe contract's data-use clause and its account or program code
A regulated function inside the companyThe function as designated, including staff who perform it part timeRecords of the function created in general-purpose systemsThe written designation of the unit and its systems
Commercial sensitivity: pricing, deals, strategyOwnership and creation by the unitShared drives copied to other teams; board material in executive mailboxesThe owner's list of containers and date ranges
The signer does not control the unitAuthority: only records the signer may licenseEverything the unit holds, until authority is shownGroup structure and delegation records, reviewed by counsel

Department as of when? Eight years of a public unit map

Write the date basis into the decision, because unit maps move. To measure how much, we read the OpenStack project's governance repository, whose reference/projects.yaml file is the list of official project teams and the code repositories each team delivers, and compared the October 2018 snapshot with the October 2026 one.

Of 831 repositories listed under a team in October 2018, 422 (50.8 percent) sat under the same team in October 2026. Another 39 had moved: 18 to a different project team, 19 to a special interest group and 2 to another governance body. Another 240 were listed as retired, and 130 appeared in none of the governance files we read, 128 of them from the Infrastructure team, which a February 2020 governance resolution split so that OpenDev could run as its own team. Of the 63 teams of 2018, 23 no longer existed under their 2018 names.

For the 40 teams present in both years, defining each team by its 2026 listing would miss 76 repositories those teams held in 2018 and pull in 18 that a different team held in 2018; for 22 of the 40 teams the two definitions disagree. A company excluding its analytics department across seven years of tickets faces the same choice on a smaller scale.

Three date bases are defensible, and each suits a different reason. As of creation keeps records with the unit that made them, which fits ownership and commercial reasons. At any point in the record's life catches everything the unit ever handled, which fits privilege and HR reasons. As of the extraction date is the easiest to run and the most likely to miss records filed under a predecessor's name. U.S. federal records rules lean the same way: 36 CFR 1231.14 lets records of a terminated function go to "another executive agency that inherits the function," and 36 CFR 1231.16 says restrictions imposed under a statute or Executive order "must continue to be imposed after the transfer." An exclusion tied to a function should travel with the function through a reorganization.

Where the 831 team repositories of October 2018 sat in October 2026:

Department as of when? Eight years of a public unit map
Status in October 2026RepositoriesShare of 831What an exclusion drawn from the 2026 map would do
Same team as in 201842250.8%Find them, under the current team name
Under a different project team182.2%Exclude them under the wrong unit, or miss them
Special interest group or other governance body212.5%Miss them unless every kind of unit is searched
Listed as retired24028.9%Miss them unless retired units are mapped
In none of the governance files read13015.6%Miss them entirely
A flow from 831 repositories listed under OpenStack project teams in October 2018 to their status in October 2026: 422 under the same team, 240 listed as retired, 130 in no governance file read, 21 in a special interest group or other body and 18 under a different team.
Where OpenStack's October 2018 team repositories sat in October 2026. Source: OpenStack governance repository, snapshots a04fc6b and 4c99ba0.

The department exclusion permission matrix, one row per record class

The matrix turns the reason and the date basis into a call for each class of record the department touches. Fill in one copy per excluded department. The default column applies until the evidence column is complete, and nothing in the hold state ships.

The department exclusion permission matrix, one row per record class
Record classDefault until reviewedEvidence that settles itWho signs offHow it reaches delivery
Records the department created and kept in its own queues, folders or mailboxesExcludeContainer list with system IDs and date rangesDepartment headContainer IDs written into the extraction rule
Records it owned for a time, then passed to another teamExclude for privilege and HR reasons; review for othersAssignment history showing the department as ownerDepartment head and counselAn ever-assigned condition in the rule
Records filed under a predecessor unit's name or codeSame call as the current unitUnit history table: old codes, new codes, effective datesRecords owner for each systemMapped codes listed in the rule
Records owned elsewhere that the department approved, commented on or attached files toReviewA sample of such records read against the reasonCounselField-level removal or a whole-record hold
Records about the department's subject matter held by other teamsExclude when the reason is subject basedScreen terms and a reviewed sampleCounsel and department headScreen version recorded in the log
Shared master data the department maintains, such as an employee or vendor masterHoldField list showing which attributes the department controlsData ownerExcluded fields or a reduced master
Records with a blank or unmapped unit fieldHoldCounts by year and system, with a reviewed sampleRecords ownerA held set kept out until decided
Reports and aggregates built from the department's recordsExcludeReport lineage back to source queuesDepartment headReport IDs on the exclusion list

How do you test that the export respects the exclusion?

Test it against a list the filter did not write. The department's own register (a matter list, an HR case log, a contract register or a program roster) is an independent reference: every entry that links to a record should land in the excluded set, and none should appear in the export. Searching the export for the rule's own identifiers, the canary method in our exclusion-list guide, proves the rule ran. The register test asks whether it was the right rule.

At a glance

How do you test that the export respects the exclusion

GAO's guide to assessing data reliability defines completeness as "the extent to which relevant data records and fields are present and sufficiently populated." For an exclusion, completeness runs in both directions: the excluded set must hold everything the reason covers, and the delivered set must still hold everything the reason does not cover.

The checks, in the order to run them:

  • Reconcile totals: source records for the period equal included plus excluded plus held, by year and by system.
  • Run the register forward: count register-linked records found in the included set. The pass condition is zero.
  • Run the excluded set backward: read a fixed-size sample of excluded records and count those the approved reason does not explain. Record the count rather than forcing it to zero; it is the price of the boundary.
  • Count the unknowns: records with a blank or unmapped unit field, by year, held until someone decides them.
  • Map every unit code that existed during the period to include, exclude or hold, retired and renamed codes included.
  • Rerun every check after each rule change, and keep each run's counts beside the rule version that produced them.

A review log for one exclusion, run twice (illustrative)

Illustrative: a fictional services company licenses 412,600 service desk tickets created from 2019 to 2025 and excludes its People team. The People team's case register lists 1,140 cases linked to a ticket. The first rule excluded tickets whose current queue was People.

Run 1 excluded 14,980 tickets and left 69 register-linked tickets, 6.1 percent of 1,140, in the export: tickets reassigned to IT or facilities after an HR case opened, and tickets filed under Personnel, the queue name the team used before a rename. It also passed 2,310 tickets with a blank queue straight into the included set. Run 2 excluded any ticket ever assigned to People or Personnel or matching the approved HR subject screen, and held the blank-queue tickets: 18,412 excluded, 2,310 held, 391,878 included and no register-linked ticket in the export.

Run 2 had a cost. In a 400-ticket sample of the excluded set, 29 tickets (7.3 percent) did not match the reason, mostly routine requests from People staff such as password resets, against 6 in Run 1's sample. The company kept the wider rule and logged the trade-off; a narrower subject screen is the next fix. The log that settled it:

A review log for one exclusion, run twice (illustrative)
WhenQuestionDecisionDecided byEffect on the extraction rule
Week 1Why is the People team excluded?HR matters about named employees; the boundary follows subject and the team's queuesSponsor, with counselReason code recorded; the rule must cover subject and queues
Week 1The department as of when?Any point in a ticket's life, including the former Personnel queueRecords ownerEver-assigned condition added; Personnel code mapped
Week 2Run 1 left 69 register cases in the exportRule rejectedCounselRule version 2 written
Week 22,310 tickets have no queueHold until sampledRecords ownerBlank-queue tickets routed to the held set
Week 3Run 2: no register cases in the export; 29 of 400 sampled exclusions off-reasonAccept, with the trade-off notedSponsorRule version 2 frozen for the signed scope

Would masking the department's names do the same job?

No. Masking or redacting names changes how identifiable people are; it does not change why the department was excluded. The UK ICO's guidance states that "pseudonymised data is personal data in the hands of someone who holds the additional information," so swapping HR staff names for codes still leaves personal data, and it leaves privileged advice privileged and contract-restricted work restricted.

Technical access is the mirror-image mistake. An administrator who can export every queue has access, not authority. Permission to license comes from the company's rights in the records and the promises it made about them, which is why the reason memo, not the export tool, starts the record.

The opposite shortcut, excluding everything the department ever touched, fails in its own way. Approvals, sign-offs and handoffs are often where a department appears in other teams' work; remove them and a procurement or claims workflow loses the step that explains its outcome. Count what the wider rule removes from workflows you intend to license, as with the 7.3 percent above, and decide with that number in hand.

Decision rule, limits and the evidence still missing

Proceed with a department exclusion when five things are on file: a signed reason memo, a stated date basis, an extraction rule with a version number, a run with no register items in the export and every unit code mapped, and a decision on the held set. Hold the scope when any one is missing, and never ship the held set to make the totals tidy.

The public analysis has limits. OpenStack is an open-source community, not a company; its governance files list code repositories, not business records; and the counts cover only the files named in the methods note, so a repository missing from them may still exist elsewhere. It shows that a unit map can change substantially within a few years, not how fast yours does.

The next evidence is your own: a unit history table for every system in scope with old codes, new codes and effective dates, the department's register exported with record links, and the first reconciliation run. Those three items settle more boundary questions than any policy wording.

Bringing a department exclusion to a SourceX assessment

Describe the exclusion when you list your systems, before any agreement exists. SourceX's rights check happens while the inventory is being built, and finding out whether a company qualifies needs no data uploads, so the reason, the date basis and the rough size of the excluded set can be discussed without sharing a single record.

If a deal follows, the Dataset Deal Confirmation is where the boundary becomes binding. SourceX's how-it-works page says the Confirmation "fixes the exact dataset" and that "a material change later needs a signed amendment." Among the points a supplier confirms is that it has the rights to license the data, so a department whose records you cannot vouch for belongs outside the scope or on hold. An assessment is not an offer and does not promise a payout.

Methods note: how the unit-drift counts were made

Source: the OpenStack governance repository, read through its GitHub copy (github.com/openstack/governance) of the canonical repository on opendev.org, at commit 4c99ba0 of October 16, 2026. A year's snapshot is the last commit on the first-parent history before October 28 at 00:00 UTC of that year: a04fc6b (October 25, 2018) for 2018, 7755c8f (October 24, 2019) for 2019, used only for the rename count below, and 4c99ba0 for 2026. Units are the top-level keys of reference/projects.yaml; records are the repository names listed under each team's deliverables. Each 2018 repository was listed under exactly one team.

Each 2018 repository gets one status per snapshot, checked in this order: listed under a team whose normalized key matches its 2018 team; listed under another team; listed in sigs-repos.yaml; listed in technical-committee-repos.yaml, foundation-board-repos.yaml or user-committee-repos.yaml where present; listed anywhere in legacy.yaml; otherwise not listed. Team keys are lowercased with runs of spaces, hyphens and underscores collapsed to one hyphen, and repository names are lowercased; no two keys collided after normalization. A repository missing under its 2018 name is followed to a renamed one when exactly one repository with the same name after the slash is listed in any file read at that snapshot: 142 renames at the 2019 snapshot, from the April 2019 namespace change for the move to OpenDev, and 25 at the 2026 snapshot. No ambiguous case occurred. For the 40 teams present in both snapshots, each team's 2018 repository set, mapped through the renames followed at 2026, is compared with its 2026 set; a repository counts as pulled in when the 2026 team lists it but a different team listed it in 2018, and repositories absent from the 2018 lists are not counted as differences. Shares are computed from integer counts and rounded once, half up.

Not measured: repositories outside these files, the code or contributors themselves, and anything about business records. legacy.yaml says retired projects are tracked only temporarily, so the not-listed count mixes long-retired repositories with the Infrastructure team's move to OpenDev. The ticket example's figures are illustrative and computed from stated inputs. The scripts and their outputs, which hold team and repository names only, are kept with the editorial files for the recount.

Frequently asked questions

Can a company leave one department's records out of a data license?

Yes. A supplier decides what goes into scope, and a department can be left out entirely. The work lies in writing the exclusion as a test over record attributes, with a stated reason and date basis, and checking the export against the department's own register before anything is delivered.

Does excluding a department also remove its employees from other teams' records?

Not automatically. A queue or container filter removes the department's own records, but its people still appear as approvers, requesters and commenters elsewhere. Whether those mentions must go depends on the reason; HR and privilege reasons usually call for a subject or participant screen as well.

What if the department was renamed or merged during the licensed period?

Map every unit code that existed during the period to include, exclude or hold, with effective dates, and state whether the department is defined as of record creation, at any point, or as of the export. In OpenStack's governance files, 23 of the 63 teams of 2018 no longer existed under those names by 2026.

Who should approve a department exclusion?

The person authorized to license the company's data approves the scope, the department head confirms where its records live and what its register holds, and counsel confirms the reason and the test when privilege, employment, contract or regulatory questions arise. Each approval belongs in the review log.

Will excluding a department make the dataset less useful?

It can. Departments often hold the approval or handoff steps of other teams' workflows, so a broad exclusion can strip out the step that explains an outcome. Count what the rule removes from each workflow you plan to license, and narrow the rule where the reason allows.

Sources

  • 45 CFR 164.105 lets a hybrid entity's health care component include another component only to the extent it performs covered functions, and treats the component as if separate and distinct from other units for disclosure. Source
  • 36 CFR 1231.14 lets records of an agency whose functions end be transferred to another executive agency that inherits the function. Source
  • 36 CFR 1231.16 says restrictions imposed under a statute or Executive order continue after records are transferred between agencies. Source
  • GAO's Assessing Data Reliability guide (GAO-20-283G) defines completeness as the extent to which relevant data records and fields are present and sufficiently populated. Source
  • The UK ICO states that pseudonymised data is personal data in the hands of someone who holds the additional information. Source
  • OpenStack's governance repository describes reference/projects.yaml as the list of all official OpenStack project teams and the deliverables they produce. Source
  • OpenStack's projects.yaml lists each project team's deliverables and repositories; read at snapshots a04fc6b (2018) and 4c99ba0 (2026) for the unit-drift counts. Source
  • OpenStack's legacy.yaml lists projects that were previously official and dropped from the official list, tracked temporarily. Source
  • A February 2020 OpenStack resolution split the infrastructure team into an OpenDev team and an OpenStack infrastructure team. Source
  • An April 2019 governance commit updated repository namespaces for the OpenDev migration, which renamed many listed repositories. Source
  • SourceX's how-it-works page says the Dataset Deal Confirmation fixes the exact dataset, a material change later needs a signed amendment, and the supplier confirms it has the rights to license the data. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies