06-reference/research

ld allowance prevalence enterprise eligibility

2026-08-28·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
macpricingl-and-d-stipendbuyer-researchcalibration

Prevalence is still unmeasurable, but the selection-bias worry was right: the first enterprise policy located names software and subscriptions in its exclusion clause

The question

"What share of engineering employers structure L&D as a named individual per-employee allowance versus a centrally-managed training budget — and does the permissive software/subscription eligibility language found at handbook-transparent small companies still hold at 10,000+ headcount?"

Carried forward from [[2026-08-23-ld-stipend-buying-wallet-mac-pricing]], which verified the wallet exists and permits digital products on a sample of two small handbook-transparent employers, and named this as the single finding that would move its recommendation from moderate to high confidence.

What we already know (from the vault)

What the web says

Convergences and contradictions

Synthesis for RDCO

The honest headline is that half this question got answered and half did not, and the half that got answered went against us. Prevalence remains unmeasurable from public sources. The ceiling on confidence there is SHRM's unquantified "Few organizations do this currently," from a 2022 report with 356 HR managers, offered as a recommendation rather than a measurement. Nothing above that ceiling exists in the non-vendor world: BLS measures benefit access by type, CIPD and ATD measure spend and hours, and the vendors who would happily supply a number sell the software that administers the thing they are measuring. Getting a real number would require an original hand-count of published policies, and this pass shows why that is slower than it sounds. The University of Toronto case is the clean illustration: PERA is a genuine named per-faculty allowance, the policy page is public, and the dollar amount is deliberately held on an intranet that does not resolve publicly. Policies are public more often than amounts are.

The eligibility half did resolve, and it resolved against the parent brief's generalization. Adobe at roughly 31,000 employees runs a named individual allowance that is broadly available to the workforce, and its exclusion clause reads "software, equipment or subscriptions to publications." Falsifying a universal claim takes one counterexample, and this is one. Establishing a rate would take a sample we do not have, so the correct statement is narrow: the claim "enterprise L&D allowances permit software and digital products" is now known to be false in general, and we still do not know how often it is false. The shape of the counterexample is worse than its existence. The wallet does not merely narrow at scale, it shrinks by roughly an order of magnitude. MAC at $350 is 3.5% of GitLab's fund and 17.5% of Free Law's. Against Adobe's $1,000 it is 35% of an engineer's entire annual allowance, which is a materially different ask, and at UIUC's $500-per-two-years it does not fit at all.

There is one live thread in the Adobe text that matters more than anything else here, and it is a classification question rather than a size question. Adobe excludes software and subscriptions but explicitly permits "online learning" and "business-related books." MAC is not a subscription and it is not equipment. Whether it reads as online learning or as software is decided by how we describe it, not by what it is. That makes the reimbursement kit already filed as P1 on [[2026-05-14-mac-prelaunch-readiness-checklist]] more valuable than the parent brief assessed, and it makes its copy specification concrete rather than generic. The invoice line, the certificate of completion, and the manager email template should describe MAC as a course with a completion artifact. Any string in that kit that describes MAC as a skill pack, a tool, a plugin, or a subscription is handing an approver the exact phrase in the exclusion clause. That is a falsifiable and cheap instruction, and it is the one thing in this brief that changes what we build.

What this does not change is the price, and the reasoning is the same as the parent's. Wallet cannot be conditioned at a self-serve checkout, the unreimbursed personal buyer remains the binding constraint, and the $199 single-seat recommendation survives. If anything the enterprise finding strengthens the case for the lower single-seat number, since $199 against a $1,000 allowance is a 20% draw that an engineer can plausibly justify without a second conversation, where $350 is a third of the year's budget. The parent asked whether this finding would move its recommendation from moderate to high confidence. It does not. It moves the copy recommendation from moderate to somewhat better specified, and it moves the prevalence claim from unmeasured to unmeasured-and-now-known-to-be-unmeasurable-from-public-sources, which is a smaller but more durable result. We should stop spending research cycles trying to size this segment.

Why this is in the vault

This closes the first of the two open follow-ups on [[2026-08-23-ld-stipend-buying-wallet-mac-pricing]] and directly constrains the reimbursement-kit copy that is filed as a P1 item on [[2026-05-14-mac-prelaunch-readiness-checklist]]: MAC must be described as online learning with a completion artifact, never as software, a tool, or a subscription, because the first enterprise policy we located names those three words in its exclusion clause. It also retires the prevalence question as a research target, so future /curiosity passes do not re-queue it.

Open follow-ups

Related

Sources

Vault

Primary employer policy text (retrieved 2026-08-28)

Prevalence sources, all of which fail to answer the question

Explicitly not cited