Skip to content

AI uses for records

Procedural prerequisite dataset: 57 conditions added after the steps

By SourceX Editorial · Updated

9 sections

Short answer

A procedural prerequisite dataset pairs each version of a procedure with the conditions added to it later, the commit or ticket that added them and the scope they apply to. In Foreman's documentation, 83 edits added a prerequisite while leaving every step untouched; on reading, 57 named a condition that had been missing, mostly permissions (17), objects that had to exist first (17) and components to install or enable (15).

Document stack

Procedural prerequisite

This page covers procedural prerequisite dataset: 57 conditions added after the steps.Illustrative

Key takeaways

  • Correct steps are not a complete answer: 57 changes in Foreman's documentation added a missing condition without touching a single step.
  • Of 625 procedures present at the end of 2023, 20 (3.2%) later gained such a condition by October 2026, and 85 (13.6%) gained a new prerequisite by any route.
  • Permissions, objects that must already exist and components to install account for 49 of the 57 gaps.
  • Only 7 of the 35 commits that filled a gap cited a bug or issue, so ticket links alone would miss most of them.
  • A buyer's test should give a model the earlier version and score whether its answer states the condition and ties it to the record that proves it.

What is a procedural prerequisite dataset, and why do correct procedures need one?

A procedural prerequisite dataset records, for every version of a procedure, the conditions that must already be true before step one: the permission the user holds, the object that must exist, the component that must be installed and the scope in which each applies. Its distinctive field is the condition that was added later, joined to the record that added it.

Red Hat's modular documentation guide defines the section the same way: prerequisites are "a bulleted list of conditions that must be satisfied before the user starts the procedure." A knowledge article can be right at every step and still fail its reader, because the failure happens before the first step runs. An assistant that retrieves and repeats such an article inherits the gap, and an evaluation that grades only the steps will mark that answer correct.

The decision for an enterprise knowledge AI team is whether a candidate dataset can teach and test that difference. Doing so takes version history, not a snapshot: the earlier text, the later text, the date between them and evidence that the added condition was genuinely missing rather than reworded, moved or introduced by a product change.

How often did Foreman's procedures gain a missing condition later?

Rarely for any one procedure, but steadily across a documentation set. Foreman's documentation repository, pinned at commit c2ace2c (27 October 2026), keeps each procedure as a module with a labelled .Prerequisites block and a commit history back to May 2019, so every change can be compared with the version before it.

Of the 625 procedure modules present on 22 December 2023, the last commit before 2024, 36 (5.8%) later received a commit that added a new prerequisite and changed nothing else in the module, and 20 (3.2%) received one that a reader judged a genuinely missing condition. Counting additions made alongside other edits, 85 (13.6%) gained a new condition by October 2026.

Across the full history the census examined 10,841 changes to procedure modules. Each filter below is stated as the script applies it in the methods note, and each removes a named kind of change:

How often did Foreman's procedures gain a missing condition later?
Filter applied by the scriptModule changes leftWhat the filter removes
Changed file has a .Procedure block in its earlier version10,841Edits to concept, reference and assembly files
Adds at least one prerequisite item1,222Edits that leave the prerequisite list as it was
Adds an item that is not restated or moved251Rewordings, link edits, per-build copies and items moved within the same commit (971)
Leaves every other line of the module unchanged83Additions made while steps or text were also rewritten (168)
Reader confirms the condition had been missing57Restated conditions (12), product or platform changes (11) and notes that are not conditions (3)
A five-step filter from 10,841 procedure edits in Foreman's documentation to 1,222 that add a prerequisite, 251 that add a new condition, 83 that leave the steps untouched and 57 confirmed missing conditions, with a bar splitting the 57 into permissions 17, objects that must exist 17, components 15, access 5 and other 3.
Counts from Foreman's documentation at commit c2ace2c (October 2026), computed by the census script; the last step uses one reader's labels.

Which conditions were missing, and how long had the steps stood?

Permissions, objects that must exist first and components to install or enable account for 49 of the 57 gaps. The rest are access conditions such as an HTTP proxy, a host name or open ports (5), plus one case each of a backup, a disk-space check and a workaround for a known defect.

The steps had often been in use for months or years. Measured from the last commit that changed a module's steps, the median gap was filled after 84.0 days, and 11 of the 57 came more than a year later. In November 2025 (commit 9775055) the procedure for assigning a host to a location gained two conditions, the assign_locations permission and membership of that location, after its steps had stood unchanged for 778.7 days; a related procedure that assigns a Capsule to organizations and locations gained the matching condition after 1,454.8 days.

Gaps also travel in families. One February 2026 commit (275369e) added the same condition, that the importing organization is configured to synchronize content through exports, to four import procedures at once. A dataset that treats each article on its own will count four errors where the maintainers fixed one missing fact shared by siblings.

The counts by type, each with a dated example from the commit history:

Which conditions were missing, and how long had the steps stood?
Gap typeEventsExample from the commit historySteps unchanged before the fix
Permission or role17Adding an Amazon EC2 connection gained the AmazonEC2FullAccess permission (0c5b09f, 5 September 2023; the message cites a Bugzilla report)605.9 days
Object or setting that must already exist17Four import procedures gained an importing organization configured for export-based synchronization (275369e, 11 February 2026)148.1 days for three, 21.2 days for one
Component installed, enabled or configured15Two Insights installation procedures gained Podman configured for the netavark network backend (bebc258, 15 July 2026)118.9 and 279.9 days
Access: proxy, host name, ports, registry login5Renaming the server gained the check that hostname and hostname -f both return the fully qualified domain name (58c4aae, 26 May 2021; cites Bugzilla)75.6 days
Backup, disk space or defect workaround3An in-place Leapp upgrade gained a repository override for a module the Leapp tool did not enable properly (1705488, 14 October 2022)7.1 days

Did bug reports add the missing conditions, or documentation reviews?

Mostly reviews. Only 7 of the 35 commits that filled a gap cite a Bugzilla report, a Jira issue or a numbered issue in their message; the other 28 arrive as documentation work titled after permissions, prerequisites or a feature's rework. A dataset that harvested only ticket-linked fixes would find a fifth of these gaps.

The timing points the same way. Gap fixes rose from 3 in 2022 to 10 in 2023, 12 in each of 2024 and 2025, and 17 in 2026 to late October. On 21 May 2026 the repository added a written guide for reviewing prerequisites, and 10 of the 57 gaps were filled on or after that date. Neither figure proves the guide caused the fixes; both show that how a team audits its documentation changes how many gaps its revision history will reveal.

For a buyer, a supplier's revision history therefore measures its review process as much as its procedures. Ask when and how the knowledge base was audited before reading a low gap rate as a sign of quality, and expect a recently audited archive to show a burst of filled gaps that an unaudited one simply lacks.

The prerequisite authority map: which record proves each condition

Every type of condition has a record that settles whether it holds, and a dataset should cite that record rather than the article that mentions it. The map pairs each gap type with its authority, the field a dataset needs and the failure an AI answer shows when the field is absent:

The prerequisite authority map: which record proves each condition
Condition typeAuthoritative recordField the dataset needsFailure when the field is absent
Permission or roleThe product's role and permission definitions and the user's assigned rolesExact permission names and the role that grants themThe answer works for an administrator and fails for everyone else
Object or setting that must existThe configuration state, or the earlier procedure that creates itA link to the creating procedure and the setting nameThe user reaches a step that refers to something not yet created
Component installed, enabled or configuredPackage, plugin and repository lists for the stated build and platformComponent, version and the builds it applies toA command fails because the package or plugin is absent
Access: network, proxy, host name, portsPort and firewall requirements, proxy and name-resolution settingsEndpoint, port and direction of trafficThe procedure stalls or times out with no step visibly at fault
Backup before a risky changeThe backup procedure and any statement that the change cannot be undoneWhat to back up and which step makes it necessaryAn irreversible step runs with no way back
CapacityStorage and sizing requirementsResource, threshold and where it is measuredThe step fails partway after consuming space or memory
Known defect workaroundThe bug report and the version that fixes itDefect identifier and the versions it affectsThe workaround outlives the fix, or is missing before it

An evidence-grounding test for answers that skip a condition

The test asks whether a model, given the earlier version of a procedure, states the condition that a later version added, and whether it can point to the record that proves it. Build it from the dataset's version pairs and keep the later text hidden:

At a glance

An evidence-grounding test for answers that skip a condition

  • Freeze the evaluation date before the fix: the model sees the documentation set as it stood the day before the prerequisite was added, plus any tickets open on that day.
  • Ask for the task, not the gap: pose the user's goal, such as assigning a host to a location, rather than a question that names the missing permission.
  • Score three outcomes separately: the condition stated before the first step, the condition tied to an authoritative record from the map, and no invented prerequisite that the later version does not support.
  • Respect scope: a condition added for one product build, platform or deployment variant must not be demanded for the others, so every label carries its applicability.
  • Treat sibling procedures as one item: when one commit adds the same condition to several procedures, count a single gap and test whether the model transfers it across the siblings.
  • Keep restated and product-change conditions out of the positive set: a rewording, or a condition that arrived with a new feature, is not a gap the earlier text should have caught.
  • Report results by gap type: a model that finds missing permissions but misses missing components has a specific weakness that a single average hides.

Strongest objection: the condition was obvious, or written somewhere else

Both documentation guides warn against listing everything. Red Hat's guide says "Do not list obvious prerequisites," and Foreman's own review guide repeats the rule, giving "You are logged in." as an example of a bad prerequisite. Some omissions are therefore editorial choices, and a model trained to add every conceivable condition would produce answers no expert wants to read.

The data supports a narrower reading. The 57 gaps are conditions the maintainers later chose to add, so by their own standard they were not obvious. Red Hat's guide also allows a condition that applies to a whole user story to sit in the assembly file instead of each procedure; checking every file that included the module at the parent commit found the condition already stated in 1 of the 57 cases.

The labels are one reader's judgments, made from each module's diff and saved with commit hashes so anyone can re-read them. The softest boundary lies between a genuine gap and a product change: 11 additions arrived with a renamed setting, a new containerized variant, a plugin that became optional or support for more operating systems, and were kept out. Foreman is also one well-maintained open-source project; an enterprise knowledge base reviewed less often will show different rates, which is exactly why a buyer should measure the supplier's own history.

What to request when licensing procedural knowledge records

Ask for the revision history of each procedure with dates, the change records or tickets linked to each revision, the role and permission definitions the procedures refer to, and the build, platform or customer scope that each condition applies to. A snapshot of current articles cannot support the test, because the gaps exist only in the difference between versions.

Ask how the knowledge base is reviewed, too. A supplier that audits prerequisites on a schedule will show more filled gaps than one that edits only when a ticket arrives, and the dataset documentation should say which pattern applies. Internal help desks, IT runbooks and customer-facing help centers usually keep these histories in different systems, so the request should name each one.

SourceX sources datasets of this kind for buyers; its request page says datasets are "sourced on request, not held in stock" and asks for the scope, format, timing and licensing terms a team needs. A requirement that names the gap types, the version window and the acceptance test from this article gives a supplier search something specific to match; it implies no particular volume, availability or acceptance outcome.

Methods note: repository, counting rules, labels and limits

Data: theforeman/foreman-documentation, a bare clone of the default branch at commit c2ace2cb364c3f902cb889b0afaa80fcb3c4c098 (committed 27 October 2026), with history from 15 May 2019. The census compared 4,635 non-merge commits with their single parents, skipped 8 merge commits while keeping side-branch commits, followed renames with git's default 50% similarity and ran no repository code.

Rules as applied: a procedure module is any changed .adoc file whose earlier version has a line that is exactly .Procedure. Prerequisite items are the list items in blocks opened by .Prerequisites or .Prerequisite, running to the next block title or heading. Bulleted and ordered-list items both count; nested items join their parent, and an item ends at a blank line, a line that is only +, a delimiter line, the next item or an include line. Include lines are replaced by the included file's items and conditional directives are skipped. A new item is item text absent from the earlier version. Words are runs of a-z and 0-9 in lowercased text, read after URLs and xref: and link: macros are deleted, together with everything from "For more information" on. Attribute links such as {InstallingServerDocURL}anchor[text] are not deleted: the attribute name, the anchor and the link text stay and count as words. New items left with no words, such as a bare link, are set aside. A new item counts as restated when its word set has a Jaccard similarity of at least 1/2 with any item of the earlier version, or at least 4/5 of the smaller set (of at least 4 words) appears in the other; it counts as relocated when it matches in the same way any line of at least 3 words deleted anywhere in the same commit, with list markers trimmed. A prerequisite-only event adds at least one remaining item and leaves every other non-blank line of the module unchanged.

Tracker references are Bugzilla links, "bug" followed by a number of 6 or more digits, BZ numbers, SAT- or SATDOC- Jira keys, Foreman Redmine issue links, and fixes, closes, resolves or refs followed by a number; the pull-request number GitHub appends to a subject is excluded. Steps age runs from the last commit that changed the module's steps (the lines after .Procedure up to the next verification, results, next steps, troubleshooting or additional resources title) or created the module, in exact seconds. An even-count median averages the middle pair, and every figure is rounded once, half up. The baseline is the last first-parent commit before 1 January 2024.

Labels (GAP, CHANGE, RESTATED, NOTE) and gap types were assigned by one reader to all 83 prerequisite-only events. The assembly check tests each gap's items against every line of each file that included the module at the parent commit. Not measured: the 168 additions made alongside other edits, prerequisites added to assemblies, whether readers ever hit the gaps, and the linked Bugzilla and Jira records, which were not opened. Scripts and outputs hold commit hashes, paths, counts and documentation text only; commit messages were read in memory and not saved.

Frequently asked questions

What is a procedural prerequisite?

It is a condition that must already be true before the first step of a procedure: a permission held, an object created, a component installed or network access in place. It differs from a step because the procedure does not perform it. The reader must have done it, or must check it, beforehand.

How can revision history separate a missing prerequisite from a rewording?

Compare each version with the one before and keep only added conditions that do not resemble any earlier item or any line removed in the same commit. Then require that nothing else in the procedure changed, and have a reader confirm the condition was missing rather than introduced by a product change. In Foreman's documentation those steps narrowed 1,222 additions to 57 gaps.

Why not find missing prerequisites through bug reports?

Most are not filed as bugs. In Foreman's documentation only 7 of the 35 commits that filled a gap cited a Bugzilla report, a Jira issue or a numbered issue; the rest came from documentation reviews. A dataset built from ticket links alone would miss about four in five.

Does a missing prerequisite mean the knowledge article was wrong?

Not at the level of its steps: in every case counted here the maintainers left the steps exactly as they were. The article was incomplete for readers who lacked the condition. That is why an evaluation needs a separate score for stating the condition, not only for reproducing the steps.

How should an AI answer handle a prerequisite it cannot verify?

State the condition before the first step, say how to check it and name the record that would confirm it, such as the role definition or the port requirements. It should not invent conditions the documentation does not support, and it should say when a condition applies only to one build or platform.

Sources

  • Foreman's documentation repository at the pinned commit: procedure modules with .Prerequisites and .Procedure blocks and their commit history, the data behind every count in this article. Source
  • Foreman's prerequisites review guide defines prerequisites as conditions satisfied before a procedure starts and says not to list obvious ones, with "You are logged in." as a bad example. Source
  • Red Hat's modular documentation guide defines prerequisites as a bulleted list of conditions satisfied before the user starts the procedure, says not to list obvious ones, and allows story-wide prerequisites in the assembly. Source
  • Commit 9775055 (12 November 2025) added the assign_locations or assign_organizations permission and location or organization membership as prerequisites to three procedures. Source
  • Commit 0c5b09f (5 September 2023) added the AmazonEC2FullAccess permission as a prerequisite to adding an Amazon EC2 connection. Source
  • Commit 58c4aae (26 May 2021) added the prerequisite that hostname and hostname -f return the FQDN before the renaming script is used. Source
  • Commit 275369e (11 February 2026) added the same prerequisite, an importing organization configured to synchronize content through exports, to four import procedures. Source
  • Commit 1705488 (14 October 2022) added a workaround to the Leapp upgrade prerequisites for a module the Leapp utility did not enable properly. Source
  • Commit bebc258 (15 July 2026) added the prerequisite that Podman is configured to use the netavark network backend to Insights installation procedures. Source
  • SourceX's request page says datasets are sourced on request, not held in stock, and asks buyers for scope, format, timing and licensing needs. 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