Skip to content

Systems and records

Software license entitlement dataset: installed is not entitled

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

To build a software entitlement reconciliation task, start from one installation record, link it to the entitlement meant to cover it (contract, order, metric, quantity, version rights), apply the license rule, and record the decision with its evidence and a later inventory that proves the outcome. An install shows software is present; whether it is entitled depends on documents the installer never sees.

License terms

Software license entitlement

This page covers software license entitlement dataset: installed is not entitled.Illustrative

Key takeaways

  • An installation shows software is present; an entitlement is a usage right recorded in contracts and orders.
  • The unit of a reconciliation task is one decision, not a product-wide license position.
  • Packs, version rights and license metrics are where naive install counts give the wrong answer.
  • Steps that depend on contract language need counsel or specialist sign-off recorded in the packet.
  • The outcome is proven by a later inventory or license record, never by a closed ticket.
  • License keys, hostnames, user identities and vendor terms stay out, and access is not a right to license.

How do you build an entitlement reconciliation task from installation records?

Start from one installation record and ask a single question: which entitlement, if any, covers this install, and what evidence shows it. The task packet then carries the starting state, the linked evidence on both sides, the license rule that applies, the decision a qualified person made, and a later record that shows the decision held.

The difference from a normal software asset management report is the unit. A report summarizes a product's position across the estate. A task packet isolates one decision, such as whether a group of installs is covered, should be reclaimed or needs a purchase, and keeps everything a reviewer or a model would need to reach that decision again.

Built this way, a software license entitlement dataset teaches the reasoning that discovery tools cannot do on their own: reading a contract metric against an inventory, noticing when a pack license covers several seats, and knowing when to stop and ask counsel instead of guessing.

Why an install count is not a license position

An installation proves that software is present on a device; an entitlement is the usage right someone bought. The two are recorded by different systems, owned by different teams and measured in different units, which is why reconciling them is a judgment task rather than a join.

Standards draw the same line. ISO/IEC 19770-3 deals only with software entitlements, which it describes as the part of a software license concerned with usage rights, and it defines a transport format for them, including their metrics. Software identification tags, defined in ISO/IEC 19770-2 and covered by NIST's NISTIR 8060 guidelines, identify what is installed. Identification answers what is on the machine; it says nothing about whether the organization may use it.

Open source asset tools are candid about the gap. GLPI's documentation says license management is not automated and needs human follow-up, and it tells administrators to check the number of installations carefully against the number of uses a license allows. GLPI also keeps a purchase version separate from the version in use, because the two can differ, and allows a license to be the child of another so that packs and grouped licenses can be represented. Each of those fields is a place where a naive count gives the wrong answer.

Metrics add a further layer. The same product can be licensed per device, per named user, per concurrent user, per processor core or by subscription, and the contract decides which. An install count answers the per-device question at best; for every other metric, the reconciliation needs assignment or usage data as well.

A reconciliation task draws on five kinds of record. Listing what each can and cannot prove keeps the task honest about where the decision comes from.

The CIS Controls ask for a detailed inventory of licensed software installed on enterprise assets, recording at least the title, publisher, initial install or use date and business purpose, and they recommend automated inventory tools. NIST's IT asset management practice guide, SP 1800-5, describes the broader goal of tracking assets through their life cycle so management can see what is used, where and how. Neither replaces the contract.

Capture each record with its extract date. Inventories change daily, and a task that pairs this week's install list with last year's license record will show gaps that never existed at any single point in time.

The records a reconciliation task links, and what each can prove
RecordTypical systemWhat it can proveWhat it cannot prove
Installation inventoryEndpoint management, discovery agent, SWID tagsSoftware title, version and device, with install or first-seen dateThat any right to use it exists
License recordSAM tool or asset database such as GLPIDeclared quantity, purchase version, expiry, parent and child linksThat the record matches the signed terms
Contract and orderContract repository, procurement or ERPThe metric, quantity, term and use conditions actually agreedWhere the software is installed today
Assignment and usageIdentity directory, license portal, usage meteringWho or what a seat is assigned to, and whether it is usedThat the assignment complies with the metric
Change and removalService desk ticket, change record, later inventory scanThat an uninstall or reassignment was requested and doneWhy the decision was right

The counsel-reviewed entitlement evidence task packet

The packet is the asset. Each row is a section that must be filled before the task can be used for training or evaluation. The counsel review row is what separates it from a spreadsheet: any step that turns on reading contract language is signed off by a lawyer or a licensing specialist working with counsel, and the packet records that sign-off rather than the reasoning of whoever ran the report.

At a glance

The counsel-reviewed entitlement evidence task packet

Use the same structure for real cases, which need the company's approval, and for illustrative ones, which must be labeled as such in the packet header.

The counsel-reviewed entitlement evidence task packet
Packet sectionRequired contentEvidence referenceSign-off
HeaderTask ID, packet version, real or illustrative flag, product family, as-of dateNonePreparer
Starting stateInstall count by version and device group, declared entitlement count, open questionsInventory extract ID and date; license record IDPreparer
Entitlement evidenceMetric, quantity, term, version rights and use conditions, quoted from the contractContract or order reference, clause numberCounsel or licensing specialist
Applicable ruleThe rule the decision applies, in one sentence, with any assumption statedClause reference; vendor product terms versionCounsel
DecisionCovered, reclaim, true-up purchase needed, refer to counsel, insufficient evidenceAll of the aboveAsset manager
ActionUninstall, reassign, purchase request, contract queryTicket or change record IDAsset manager
Outcome checkPosition after the action, from a later inventory scanSecond inventory extract ID and datePreparer
CounterexampleA nearby case where the same rule gives a different answerIts own evidence referencesCounsel
Redaction logWhat was removed or generalized and whyRedaction checklistSecurity reviewer

Worked task case: a design suite on 40 workstations (illustrative)

This case is invented to show the packet. The product, counts and dates are illustrative, not real records or a real vendor's terms.

The labels a grader cares about are the decision (reclaim four, buy two), the evidence that supports it (per-device metric confirmed by counsel, usage showing four installs idle) and the outcome (covered, proven by the second extract). A model that answers buy six has read the inventory but not the usage data; one that answers covered has trusted the parent license count without reading the order lines.

  • Header: task ID, packet version 1, illustrative, a commercial design suite licensed per named device.
  • Starting state: the inventory extract shows 46 installs across one design team's workstations and a shared render room. The license record declares 40 entitlements. Six installs appear uncovered.
  • Entitlement evidence: the order shows two purchases, a pack of 30 device licenses and a later pack of 10, each recorded in the asset tool as a child of one parent license. Counsel confirms the metric is per device and that the contract has no secondary-use right for a second machine.
  • Investigation: usage data shows four of the six extra installs have not been launched in the review window. The other two sit on render-room machines whose users already have licensed workstations, which does not help under a per-device metric because each device needs its own license.
  • Decision: reclaim four unused installs; a true-up purchase is needed for two render-room devices, unless those machines can be retired.
  • Action: two change records, one for uninstalls and one for a purchase request. The service desk closes both.
  • Outcome check: an inventory scan two weeks later shows 42 installs and, after the purchase arrives, 42 entitlements. The position is recorded as covered, with the second extract ID as evidence.
  • Redaction: device hostnames replaced with group labels, user names removed, license keys never extracted.

Counterexample: the version that looked unlicensed

A good packet set includes cases where the obvious answer is wrong. The most common one in software licensing is version.

Illustrative counterexample: the inventory shows a newer major version of a database client than the version on the order, so a simple match flags every install as unlicensed. The asset tool, like GLPI, holds a purchase version and a version in use. Whether the newer version is covered depends on contract language about upgrade or maintenance rights and on whether those rights were current when the upgrade was installed. That is a contract reading, so the decision label is refer to counsel, and the packet records counsel's answer and the clause it relied on.

A model trained without such cases learns that version mismatch means non-compliance. A model trained with them learns that version mismatch is a question, and that the answer sits in a document the inventory never sees.

Two other counterexamples are worth building early. One is a subscription seat assigned in a directory to a person who has left, which shows up as a license in use with no install anywhere; the right decision is reassignment, not purchase. The other is software installed on a test or disaster recovery machine, where coverage depends on whether the contract grants non-production or failover rights. Both look like mismatches to a counting tool and both need the contract to settle.

Which record proves the decision, beyond a closed ticket?

The decision is proven by the entitlement evidence and counsel's sign-off; the outcome is proven by a later inventory scan or license portal export that shows the new position. A closed ticket proves only that someone marked work as done.

At a glance

Which record proves the decision, beyond a closed ticket

This distinction matters for anyone grading a model. If the outcome label is copied from the ticket status, a task where the uninstall failed silently will be graded as a success. Require the outcome check to cite an extract dated after the action, and treat a missing second extract as an unknown outcome rather than a pass.

The same rule applies to purchase decisions. A purchase request is not an entitlement until the order is placed and the license record is updated, so the outcome check for a true-up cites the updated license record, not the request.

  • Decision evidence: contract clause, order reference, metric and counsel or specialist sign-off.
  • Action evidence: change record or ticket, with requested and completed dates.
  • Outcome evidence: second inventory extract or license portal export dated after the action.
  • Failure signal: a later inventory that still shows the reclaimed installs, or a license record never updated.

Choosing which reconciliation decisions become tasks

Not every gap deserves a packet. Pick decisions that a reviewer could get wrong, and make sure the set covers each decision label rather than mostly easy matches.

A practical mix for a first set is a handful of straightforward covered cases as controls, a larger share of reclaim and true-up decisions, a deliberate group of refer-to-counsel cases, and some insufficient-evidence cases where the right answer is to stop. Label the share of each in the set's documentation so an evaluator can see whether a model is simply predicting the most common label.

Prefer products with clear metrics and complete records for early tasks, then add harder families: server and virtualization licensing, user-based subscriptions assigned through a directory, and bundles where one purchase covers several products. Each family needs its own counterexample, because the rule that fails differs.

Choosing which reconciliation decisions become tasks
Candidate decisionUse as a task?WhyMinimum evidence
Install matches a single-device license one to oneOnly as a controlNo judgment involvedInventory row and license record
Installs exceed declared countYesReclaim versus buy is a real choiceUsage data, contract metric, order
Version in use differs from purchase versionYes, often refer to counselTurns on upgrade or maintenance rightsContract clause and maintenance dates
Pack or child licenses involvedYesNaive counts miss parent and child linksOrder lines and license hierarchy
No contract can be foundYes, as insufficient evidenceTeaches the model to stopSearch log showing what was checked

What must be stripped from the packet before it leaves IT

Reconciliation data sits close to sensitive material, so the rule for this kind of packet is simple: reconstruct the decision without live secrets or details someone could act on.

At a glance

What must be stripped from the packet before it leaves IT

Software contracts can also restrict disclosure of their own terms, and a company that holds an inventory may not hold the right to share a vendor's pricing or contract text. Having the files is access; whether they can be licensed depends on the contracts behind them, and removing secrets does not create that right. Counsel reviews both questions before anything leaves the company.

  • License keys, activation codes, API tokens and portal credentials: never extracted, not merely masked.
  • Hostnames, IP addresses, network paths and directory structures: replaced with group labels.
  • User names, email addresses and employee IDs: removed or replaced with stable pseudonyms by role.
  • Vulnerability, patch-level or security tool details that would map the estate: removed.
  • Vendor pricing, discount levels and contract text beyond the clause the decision needs: removed unless the contract allows disclosure.
  • Audit correspondence with a vendor and any privileged legal advice: excluded.

Licensing reconciliation histories as task data

Reconciliation decisions linked to their evidence and verified outcomes are uncommon, because most organizations keep only the latest position. IT service providers and companies with years of asset records and change history can hold unusual material for AI developers building agents that work with contracts and inventories together.

Through SourceX, such a company can license chosen asset and change records to AI developers while SourceX does the sourcing, rights review, contracting, preparation, delivery and payment. A rough benchmark is two or more years of consistent history and tens of thousands of records in one system; less common work can qualify at lower volume, and every dataset gets its own review.

Reconciliation packets move only after the company has signed off on the scope and terms in its agreement. Qualification starts with a brief company questionnaire, and no inventory extract is requested at that point.

Frequently asked questions

What is a software license entitlement?

It is the usage right an organization acquired: the product, metric, quantity, term and conditions under which it may use software. ISO/IEC 19770-3 treats entitlements as the part of a software license concerned with usage rights and defines a format for exchanging them.

Why does an installation not prove a license?

Discovery tools record that software is present on a device, while entitlements live in contracts and orders. Coverage depends on the license metric, version rights and assignment rules, which an inventory cannot see.

What is software license reconciliation?

It is the comparison of what is installed or used against what the organization is entitled to, followed by a decision for each gap: reclaim, reassign, buy or ask counsel. The result is a license position for each product.

What should a software entitlement reconciliation task include?

A starting state from inventory, the entitlement evidence quoted from the contract, the rule applied, the decision, the action taken and a later record that proves the outcome. A counterexample and a redaction log make it usable for training or evaluation.

Why involve counsel in license reconciliation data?

Questions such as upgrade rights, secondary use, affiliates and virtualization depend on reading contract language. Recording counsel's sign-off on those steps keeps a dataset from teaching a model to guess at legal interpretation.

Can a company license its software asset records for AI?

Possibly, but only after removing secrets and identities and confirming that vendor contracts allow sharing any terms included. Access to the records is not the same as the right to license them, so counsel reviews before anything is shared.

Sources

  • ISO/IEC 19770-3 covers only software entitlements, the part of software licenses concerned with usage rights, and specifies a transport format for them and their metrics. Source
  • NISTIR 8060 provides guidelines for creating interoperable software identification (SWID) tags as a companion to ISO/IEC 19770-2. Source
  • CIS Control 2 calls for a detailed inventory of licensed software installed on enterprise assets and for automated software inventory tools. Source
  • GLPI documentation states license management is not automated, says installations must be checked against a license's number of uses, and separates purchase version from version in use. Source
  • GLPI documentation describes installations as versions on assets and licenses as the record of usage rights, linked only when a license is assigned to a computer. Source
  • NIST SP 1800-5 describes an IT asset management approach for tracking, managing and reporting on assets throughout their life cycle. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies