Skip to content

Systems and records

Organization role change dataset: the title stayed, the authority moved

By SourceX Editorial · Updated

9 sections

Short answer

An organization role change dataset should record what each role could decide on each date, not just its title. A restructure can keep a title while moving its authority: after Python's 2019 governance change, 25 of 64 new proposals with a delegate field used the retired BDFL-Delegate name for 20 months. Map each role to duties and an authority source, tie it to the governing document version, and test labels across the cutover.

System export

Organization role change

This page covers organization role change dataset: the title stayed, the authority moved.Illustrative

Key takeaways

  • A role field is evidence of authority only when it is joined to the duties and decision rights in force on the event date.
  • In 745 Python PEP headers, the retired BDFL-Delegate field appears on 25 of 64 council-era proposals that carry a delegate field, and vanished only when PEP 1 renamed it.
  • Restructures change role data in four ways: same title with a new authority source, new title with the same duty, blanks whose default decider changes, and split or merged duties.
  • Identity exports built on the SCIM enterprise schema hold single current values, and job-title workflows stay silent when a reorganization redefines a title nobody changed.
  • Separate meaning change from task change: Python's share of proposals carrying a delegate field rose from 13.7 to 24.2 percent across the same cutover.

What should an organization role change dataset record?

It should record what each role was allowed to decide on each date, and treat the title as a label that can drift. An organization role change dataset holds records written before and after a restructure: approvals, escalations, assignments and sign-offs whose role fields a model reads as evidence of authority. When a restructure moves a duty but keeps the title, or renames the title but keeps the duty, a model trained on the raw field learns a rule that held for only part of the window.

The repair is an effective-dated crosswalk. For every role value in the data, the crosswalk states the duties and decision rights it carried, where that authority came from, which governing document defined it and the dates it applied. Records join to the crosswalk by the date of the event, never by today's org chart.

Company role histories are rarely public, so this article measures one restructure whose records and rulebook are both versioned in the open: the Python language's move from a single final decision-maker to an elected council. It then turns that record into a role crosswalk card, a checklist for any reorganization inside a record window, and a test that separates a change in meaning from a change in the work.

Python's governance change, measured in 745 proposal headers

In Python's own proposal records, a retired title kept appearing for 20 months after the authority behind it changed. Python Enhancement Proposals, or PEPs, carry structured headers, including a field that names the person who makes the final decision on a proposal when the top authority delegates it. PEP 13 records that Guido van Rossum served as Benevolent Dictator for Life, the BDFL, until July 2018, and that a December 2018 vote selected a council model. PEP 8100 describes the first steering council election, with voting from January 21 to February 4, 2019.

The rulebook moved quickly. On February 21, 2019, PEP 1, the document that governs the proposal process, was revised so that final authority for approving a PEP passed from the BDFL to the Steering Council. The same revision kept the delegate title on purpose, explaining that it still used BDFL-Delegate as a historical reference while the council's authority came from election by the active core developers. The field was renamed PEP-Delegate on October 26, 2020.

The records lagged the rulebook. We parsed the headers of all 745 PEP files at the repository's current commit and grouped them by creation date. Of the 265 PEPs created from February 2019 onward, 25 carry the old BDFL-Delegate field and 39 the new PEP-Delegate field, so 39 percent of council-era proposals with a delegate field use the retired name. The last of them was created on October 24, 2020, two days before PEP 1 renamed the field. Seven of the 25 name the former BDFL himself as delegate, now deciding on authority that flowed from the council, and five leave the field empty.

PEP 1 now tells readers to treat legacy BDFL-Delegate references as equivalent to PEP-Delegate. That holds for the duty, deciding one proposal, and fails for the authority chain: before July 2018 a delegate acted for one person, and from February 2019 for an elected body. A dataset keyed on the field name gets the period wrong; one that treats the two names as identical in every respect gets the authority wrong. The table sets out the counts by period.

Python's governance change, measured in 745 proposal headers
MeasureCreated before July 2018July 2018 to January 2019February 2019 onward
PEPs created46812265
Carry the BDFL-Delegate field62025 (5 left empty)
Carry the PEP-Delegate field2039
Share with a delegate field13.7%0%24.2%
Accepted or Final with no delegate named197 of 2405 of 596 of 141
Final authority when no delegate is namedThe BDFLNot coded: governance in transitionThe Steering Council
Column chart of new Python PEPs per year from 2016 to 2026 that carry the BDFL-Delegate or PEP-Delegate header field, with markers for the BDFL stepping down in July 2018, the first council election in February 2019 and the field rename on October 26, 2020: the old field still dominates in 2019 and 2020.
New PEPs per year by delegate field name, counted by SourceX from python/peps at commit 362ca9d on 2026-10-27.

Four ways a restructure rewrites a role field without touching it

A restructure changes what role data means in four distinct ways, and each calls for a different repair. The PEP record shows three of them; the fourth shows up when a company merges teams or moves approval limits. Treating all four as one problem, usually by renaming old values to new ones, fixes the second and damages the other three.

  • Same title, new source of authority. A BDFL-Delegate on a 2019 proposal decided on behalf of the council, not a BDFL. Repair: keep the recorded title and add the authority source and governing document version for the event date.
  • New title, same duty. The rename to PEP-Delegate changed no duty. Repair: map both values to one duty code, keep the rename date as metadata, and stop a model from reading the rename as a policy change.
  • A blank whose meaning changes. A proposal accepted with no delegate named was decided by the BDFL before the change and by the council after it. Repair: replace the implicit default with an explicit decider derived from the rule in force.
  • Split or merged duties. Illustrative business case: a regional manager title survives a reorganization, but approvals above a threshold move to a central finance desk. Repair: map the old role to each successor duty with its threshold and date, and flag records from the transition months.

Why identity and HR exports rarely show a reorganization

Because most exports describe people's current attributes, while a restructure changes what an attribute means. RFC 7643, the core schema of SCIM, the System for Cross-domain Identity Management, includes an enterprise user extension that defines department, division, organization, cost center and manager as single values: a name, or for manager a reference to another user's identifier. None of those attribute definitions records when a value took effect or what the department was responsible for.

Lifecycle tooling shares the blind spot. Microsoft's documentation describes Entra's Employee job profile change template as configuring tasks that run when an employee has a change in job title, such as notifying the manager and removing access package assignments. That workflow fires when a person's title changes. A reorganization that redefines a title while nobody's title changes fires nothing, so the identity record stays silent about the very change that altered the role's meaning.

Data warehousing solved the storage half long ago. Kimball's Type 2 slowly changing dimension adds a new row for each change and recommends at least three extra columns: a row effective date, a row expiration date and a current row indicator. The W3C Organization Ontology supplies the semantic half: it models a post as a position that exists independently of whoever fills it, gives each membership an optional interval during which it was valid, and links an organization to the change event that altered it. A role crosswalk borrows from both.

The role crosswalk card: fields that keep authority time-aware

Attach a role crosswalk card to any dataset whose window crosses a restructure; it is the part of the data card a buyer reads before using a role field as a feature, a label or an authority check. Each field below shows the value it would hold for the PEP record, so the card can be tested against a real case before it is filled in for a company's records.

Seven fields, the PEP example for each and the check each must pass:

The role crosswalk card: fields that keep authority time-aware
FieldWhat it holdsPEP exampleCheck before use
Role value as recordedThe raw title or field name, unchangedBDFL-DelegateNever overwritten by a later rename
Duty codeWhat the role does, independent of its nameFinal decision on one proposalEvery role value maps to at least one duty
Authority sourceWho granted the power on the event dateThe BDFL before July 2018; the Steering Council from February 2019Agrees with the governing document in force
Effective from and toDates the mapping applies, start included, end excludedFebruary 2019 to October 26, 2020 for council-era BDFL-DelegateNo gaps or overlaps for one role value
Governing document versionThe rule text that defined the rolePEP 1 as revised on February 21, 2019Resolves to a stored copy or commit
Implicit default deciderWho decides when the field is blankThe Steering Council from February 2019Blank values replaced by an explicit code
Transition flagWhether the event fell inside the change windowCreated July 2018 to January 2019Flagged records scored as their own slice

A change-impact checklist for a reorg inside the record window

Run this checklist once for every restructure, team merger or change to a delegation-of-authority policy that falls between the first and last record. Its output is a decision for each role value: pool the records, map them forward or split the series.

The checklist, in order:

Checklist · 9 items

  1. Date the change four ways: when it was announced, when the rule took effect, when records first show the new behavior and when the old behavior last appears. In the PEP case, new proposals kept the old field for 20 months after the rule changed.
  2. Collect the governing documents for both sides, such as delegation-of-authority matrices, approval limit tables and team charters, with their version history.
  3. List every role value in the records, including blanks and system defaults, with its first and last date of use.
  4. Map each value to duty codes per period and label each mapping: renamed with the same duty, same name with new authority, split, merged, retired or new.
  5. Rewrite implicit defaults as explicit values derived from the rule in force on the event date.
  6. Measure the task mix on each side, meaning the share of records involving each duty and decider, not only the record counts.
  7. Hold out or label the transition window and report results for it separately.
  8. Train or calibrate on one side, score on the other and break the errors down by duty code to see which mappings fail.
  9. Write the decision for each role value into the data card with the evidence behind it, so a later buyer can audit the choice.

0 of 9 ticked. Tick each item as you go.

Is the shift a new meaning, or noise from newer records?

Both can happen at once, and the test is whether the change tracks the governing rule or the calendar. Newer records can add noise, shift the mix of tasks or come from a different population, and any of these can pass for a change in role meaning.

The PEP record shows a change in the work running alongside the change in meaning. The share of proposals carrying a delegate field rose from 13.7 percent before July 2018 to 24.2 percent from February 2019, and among Accepted or Final proposals from 43 of 240 to 45 of 141. A model trained on the earlier period would underpredict delegation even with the field names mapped correctly. That is a distribution change rather than a semantic one, and it calls for reweighting or a split, not a crosswalk.

To separate the two, first hold the mapping fixed and compare distributions by duty code on each side, then hold the period fixed and compare raw role values against mapped duty codes. If the differences vanish once values are mapped to duties, the problem was meaning. If they persist by duty code, the work itself changed and historical comparisons need a stated break. If they appear only in the transition window, treat them as noise from a system in flux and keep those records out of training.

Methods note: repository, commit and counting rules

Counts come from the python/peps repository on GitHub, cloned on 2026-10-27 at commit 362ca9d of October 26, 2026. We parsed the header block of every file named pep-NNNN.rst in its peps folder, 745 files, and read the Created, Status, BDFL-Delegate and PEP-Delegate fields. Periods follow the Created date: before July 2018, July 2018 through January 2019, and February 2019 onward. Accepted or Final means the first word of Status is one of those two. Rule changes come from PEP 1's history in a blob-less clone of the same repository, commits 83687f6 of February 21, 2019 and a169cac of October 26, 2020.

Limits: creation date is not decision date, so some proposals created before July 2018 were decided later. Headers can be edited after creation; PEPs 387 and 497, created in 2009 and 2015, already carry the newer PEP-Delegate field. Empty BDFL-Delegate fields were counted as present. No decision outcomes, discussion threads or resolution links were analyzed, and Python's governance is an open-source project, not a company, so the case shows the mechanism rather than its frequency in business records. The four change types, the crosswalk card and the checklist are editorial frameworks, and the regional manager case is illustrative. The counts can be rerun with a short script against the same commit.

Specifying role history in a dataset request

Ask for role meaning as a deliverable, not an afterthought. A request covering a restructure should name the record window, the known reorganization dates, the decisions the model will learn or be graded on, and an acceptance test such as: every approval event resolves to a duty code and an authority source in force on its date, with transition-window records flagged.

SourceX works this way by design: it looks for suitable business records after a buyer explains what it needs, rather than selling from inventory, and its request form collects scope, volume, language, format, timing and licensing needs. Governing documents such as approval matrices can be named in that scope, and the supplying company agrees the scope and terms and signs before anything reaches SourceX. Whether a particular company kept dated role documents is a question its own archive answers, so no dataset is implied here.

Frequently asked questions

Can a job title be used as a label for who approved something?

Only through a dated crosswalk. A title records what a role was called when the record was written, not what it could decide. Map each title to duty codes and an authority source for the event date, using the governing document in force then; Python's BDFL-Delegate field kept its name for 20 months after the authority behind it changed.

How is a role crosswalk different from an org chart history?

An org chart history shows who reported to whom on each date. A role crosswalk maps each recorded role value to the duties and decision rights it carried in each period, including renamed, split and merged roles. Training data needs the crosswalk, because models learn from the role values written into records, not from reporting lines.

What should happen to records created during the months of a reorganization?

Flag them and score them separately. During a transition the rules may be unsettled, systems half migrated and people still following old habits, so the records mix both regimes. Keep them out of training unless the transition itself is the task, and evaluate them as their own slice.

Do identity directory exports keep a history of past roles?

Usually not on their own. Directory feeds built on the SCIM enterprise schema carry single current values for department, division, organization, cost center and manager. Past roles have to come from the HR system of record or change logs, and what each role meant from governing documents such as delegation-of-authority policies.

Can role fields re-identify people in a de-identified dataset?

Yes, when a role is rare, such as the only person allowed to approve above a threshold. Generalize rare roles to duty codes, keep the authority source coarse, and review the combination of role, unit and date before release. The actor role mapping guide linked below covers how coarse a role must be.

Sources

  • PEP 13 records that Python's creator served as BDFL until July 2018 and that the council model in PEP 8016 was selected by a December 2018 vote. Source
  • PEP 8100 describes the first Python steering council election, with voting from January 21 to February 4, 2019. Source
  • The February 21, 2019 revision of PEP 1 made the Steering Council the final authority for PEP approval and kept the BDFL-Delegate title as a historical reference. Source
  • The October 26, 2020 commit changed PEP 1's language from BDFL-Delegate to PEP-Delegate. Source
  • Current PEP 1 calls BDFL-Delegate a deprecated alias for PEP-Delegate and says legacy references should be treated as equivalent. Source
  • The PEP files at commit 362ca9d are the data behind the header counts by period (computed by SourceX). Source
  • RFC 7643 defines the SCIM enterprise user attributes department, division, organization, cost center and manager as single values. Source
  • Microsoft Entra's Employee job profile change template runs mover tasks when an employee has a change in job title. Source
  • Kimball's Type 2 technique adds a row per change with effective date, expiration date and current row indicator columns. Source
  • The W3C Organization Ontology models posts independently of their holders, membership validity intervals and organizational change events. Source

Tell us what data you need

Share scope, volume, format, timing and licensing needs. Datasets are sourced on request.

See if your company qualifies