Web edition of the report dated August 31, 2026, published online on October 24, 2026. The text is published as written on August 31, 2026. One correction only: section 1 announced “six steps” for a list of seven (levels 0 to 6); the text now says “seven steps”. The 16 primary sources listed at the end of the report were reopened on October 24, 2026; the states described (open, closed or merged issues, official wording) were unchanged. The two OpenAI pages cited refuse automated reading (403 response): they were last read on October 22, 2026. The figures are those of the report. The data and method files mentioned in the text (
data/*.csv,METHODOLOGY.md,CLAIMS-REGISTER.csv) are not published online. No independent human review has been signed for this report. French version: /rapports/claim-to-evidence-gap-2026/.
Independent publication, dated August 31, 2026
Version 1.0.0
Universal Commerce Protocol Blog
Editorial responsibility: Universal Commerce Protocol Blog
Genre: evidence register and independent editorial analysis
Scope: twelve evidence items attached to official claims by OpenAI, Stripe, Google, Shopify and UCP.
What this report measures: public evidence checked at a given date.
What this report does not measure: an organization’s internal capability, its actual adoption, its transaction volume or its market share.
Executive summary
Agentic commerce produces announcements faster than observers can test the flows. To avoid two symmetrical errors, believing every announcement or rejecting every capability that has not been reproduced, this report classifies public evidence into distinct levels.
Twelve items were verified on official channels. One falls under level 0, that is, an intention or a future availability. Nine reach level 3 thanks to a public repository, documentation, schema or tutorial. Two reach level 4 because a surface is officially described as available. None reaches level 5 or 6, because this report did not reproduce a merchant flow or execute a real payment.
The most instructive result is temporal. In September 2025, OpenAI and Stripe announced Instant Checkout for purchases from US Etsy sellers and described Shopify as “coming soon”. As of August 31, 2026, the ChatGPT merchants page states that OpenAI is moving away from the standalone Instant Checkout experience and favoring discovery, with the purchase completed on the merchant’s site or app. Both statements are true at their respective dates. Conflating them would produce a false conclusion.
The second result concerns compatibility. UCP states that it is fully compatible with AP2. This statement has public technical documentation and therefore stands at level 3. It is not treated as end-to-end proof, because UCP #599 reports a targeted incompatibility between a UCP pattern and a serialization from the reference AP2 SDK.
Key findings
Twelve items, all linked to an official source. The register retains no third-party article as final evidence for an organization’s claim.
Nine level 3 items. The ecosystem has public repositories, guides, SDKs, schemas and tutorials. This level is useful to developers, but it does not prove that a flow is available for every merchant or every user.
Two level 4 items. The historical launch of Instant Checkout and the current state of shopping in the United States are both officially described. The second replaces the first in describing the present offering.
One level 0 item. “Coming soon” remains a future formulation. It is never converted into availability, even when it appears in a launch press release.
Zero level 5 or 6. No account was created, no purchase was made and no transactional evidence was collected. The absence of level 5 or 6 is a limitation of this report’s protocol, not proof that these capabilities do not exist.
One general claim that is strongly bounded. UCP-AP2 compatibility is documented, but a contrary public report exists for one specific wire format. The report keeps both elements without turning the issue into a global verdict.
Table of contents
- Why an evidence register
- Evidence scale
- Corpus and source control
- Observed distribution
- OpenAI Instant Checkout, historical state
- OpenAI shopping, current state
- ACP, documentation and repository
- AP2, launch and artifacts
- AP2 and the FIDO Alliance
- Google UCP and codelab
- Shopify and UCP
- UCP-AP2 compatibility
- Time as an evidence variable
- What level 3 means
- What level 4 means
- What remains unmeasurable
- Guide for executives
- Guide for procurement
- Guide for journalists
- Updates and right of correction
- Limitations
- Conclusion
- References and glossary
The file data/claims.csv is the machine source. The editorial formulations are developed in this document and bounded in CLAIMS-REGISTER.csv.
1. Why an evidence register
Agentic commerce announcements often combine several types of statements on the same page: a standard is open, documentation is available, an integration works for some users, a partner is coming soon and a more complete flow is planned. These statements do not carry the same weight.
A hurried reader can turn proof of existence into proof of availability. Conversely, an overly strict analyst can consider that a capability does not exist until they have tested it themselves. Both postures lose information.
The proposed register separates seven steps: intention, announcement, eligibility, technical artifact, surface described as observable, reproduction and transactional evidence. It also keeps track of access states. An allowlist, an uncovered region, a required account or a test not carried out are not failures.
This method answers a simple question: what is the best public evidence available for this precise claim, at this precise date? It does not judge the organization as a whole. It does not assign a reputation score. It does not claim to see private systems.
The value of the register lies in its ability to be corrected. When a page changes, the historical claim remains dated and a new item describes the current state. History is not overwritten by the latest version of the marketing.
2. Evidence scale
Level 0, intention. The text describes a future, an ambition or an upcoming availability. Examples: “coming soon”, “over time” or “we plan to”.
Level 1, announcement. An organization announces a specific capability, but no additional condition, documentation or surface is checked.
Level 2, documented eligibility. Countries, audience, merchants, program or access procedure are described. The level does not prove that the capability works for everyone who is eligible.
Level 3, technical artifact. Public documentation, a repository, a schema, an SDK or a tutorial makes it possible to examine the proposal and start an integration.
Level 4, officially described surface. An official page states that an experience is active for a defined scope, or a surface is directly observable without full reproduction.
Level 5, authorized reproduction. The flow is replayed according to a published protocol, without hidden privilege, with documented steps and results.
Level 6, transactional evidence. An end-to-end transaction succeeds, with authorized, redacted and reconciled proof.
The final level is not a quality grade. An open standard can be very useful at level 3. A level 4 surface can remain limited. A level 6 may demonstrate only a single scenario.
3. Corpus and source control
The twelve items come from official OpenAI, Stripe, Google, Shopify and UCP channels and the associated GitHub repositories. Search tools were used to discover candidate pages. Each retained claim was then re-read on its primary source.
The register atomizes formulations. The OpenAI announcement of September 29, 2025, for example, becomes two items: availability of Instant Checkout from US Etsy sellers, then the future arrival of Shopify merchants. The first formulation describes a launched state. The second remains future.
Dates are those displayed by the announcements when available. For evolving documentation without a clear date, the observation date of August 31, 2026 is used. It must not be confused with a publication date.
Validation includes no merchant account, no payment and no private access. It observes the official text, the existence of artifacts and the public state of pages. Any step not carried out remains marked as such.
Third-party sources were used for discovery and possible contradiction, but not to attribute a claim to an organization. An organization is cited only for what it publishes on its own domain or in its official repository.
4. Observed distribution
The corpus contains one level 0 item, nine level 3 items and two level 4 items. Levels 1 and 2 are not represented as final levels, because the corresponding claims have either a stronger artifact or a weaker future formulation. Levels 5 and 6 are absent because the flows were not executed.
As a proportion of the corpus, level 0 represents 8.3%, level 3 represents 75.0% and level 4 represents 16.7%. These percentages are descriptive. The sample is not designed to estimate the whole agentic commerce market.
The concentration at level 3 is consistent with a building period. Organizations publish standards, guides and repositories. They thereby create a verifiable base for technical teams. But documentation does not make it possible to calculate how many merchants have activated a capability, how many transactions have succeeded or which countries have identical coverage.
The presence of two OpenAI states at level 4 also shows that evidence is not monotonic. A surface can be launched and then replaced by another flow model. The historical level remains valid for its period, but it no longer describes the present.
Figure 1. Distribution of the twelve items. The scale is ordinal and does not constitute a quality score.
5. OpenAI Instant Checkout: historical state
On September 29, 2025, OpenAI announced that Instant Checkout was available to US ChatGPT users buying from US Etsy sellers. Stripe published a converging announcement.
The claim is classified at historical level 4. Two official sources describe an available surface and bound its geographic and merchant scope. This report did not execute a purchase. It therefore cannot confirm level 5 or 6.
The same page stated that more than one million Shopify merchants, with several brand examples, would be arriving soon. This statement forms a separate level 0 item. The number describes a potential population attached to a future promise, not a number of merchants activated at the time of the announcement.
The separation is essential. Without it, an analyst could add the live Etsy state to the announced Shopify universe, then present the total as adoption. The sources do not support that conclusion.
Finally, the historical claim must no longer be used on its own to describe the current experience. The ChatGPT merchants page has since changed the flow model, as the next section analyzes.
6. OpenAI shopping: current state as of August 31, 2026
The current ChatGPT merchants page describes an experience centered on product discovery. It states that shopping is active for users in the United States and that Shopify or Etsy catalogs are already integrated with no additional step by the merchant.
The same page explicitly answers the question of direct purchase. OpenAI states that it is moving away from a standalone Instant Checkout experience in ChatGPT and favoring discovery together with merchant-owned checkout. Users discover and evaluate products in ChatGPT, then complete the purchase on the merchant’s site or app.
This state is classified at current level 4. The surface is officially described, with a scope and a conversion architecture. No purchase was reproduced by this report. The level does not prove that every Shopify or Etsy catalog is complete, up to date or visible for every query.
The change does not make the 2025 announcement false. It makes it historical. An authoritative publication must preserve this chronology: launch of Instant Checkout, future Shopify promise, then pivot to a discovery and merchant checkout flow.
This chronology illustrates the register’s reason for being. A claim must never be detached from its observation date.
Figure 2. Three distinct official states, including one future state and one current state replacing the historical flow.
7. ACP: open standard, documentation and repository
OpenAI and Stripe present ACP as a co-developed open standard. The official ACP repository and OpenAI’s commerce documentation constitute public technical artifacts.
The claim reaches level 3. Teams can read the schemas, follow the versions, examine the examples and understand the contracts. The level is stronger than a mere announcement. It allows an independent review of the technical material.
It does not, however, prove uniform availability. An open repository does not indicate how many proprietary integrations are in production, which merchants are activated or what percentage of flows succeed. Current documentation can also describe a broader direction than the surface available today.
The current merchants page in fact specifies that ACP today supports discovery and more accurate recommendations thanks to product data, and that it will expand over time to the full flow. This sentence contains a present and a future. The present is bounded to discovery. The full extension remains an intention.
For an executive, the correct verdict is therefore: ACP has a verifiable public technical base; the current transactional scope must be checked separately for each surface and each program.
8. AP2: launch and public artifacts
Google announced AP2 on September 16, 2025 as an open protocol for agent-initiated payments. The official AP2 repository publishes documentation, code and examples.
The claim is classified at level 3. The announcement alone would have established the existence of the project. The repository makes it possible to examine the objects, the flows and the samples. Integrators can therefore begin a technical evaluation without relying solely on a press release.
This level does not prove that AP2 is deployed identically across all the partners mentioned, nor that a given transaction will be accepted by a payment provider. Launch partners may contribute, test, support or plan an integration under different terms. A logo or a list must not be turned into proof of a live flow.
Report 1 also shows that the public repository contains hardening and verification issues. This transparency is useful. It does not change the level of the launch claim, but it is a reminder that the existence of an artifact is not a certification of all its branches.
To move to level 5, a complete reproduction protocol would have to be published, with versions, roles, identities, environment, test payment methods, expected states and redacted proof.
9. AP2 and the FIDO Alliance: announced governance and version 0.2
On April 28, 2026, Google announced the donation of AP2 to the FIDO Alliance together with the publication of AP2 v0.2. The FIDO Alliance page provides a converging institutional source on the initiative.
The claim reaches level 3. A dated announcement exists, the receiving institution publishes its own communication and the technical repository is public. This report does not claim to audit the legal instruments of the donation, the composition of the working groups or the completion of each operational transition.
The word “donation” can be interpreted too broadly. It does not mean that all governance decisions, all licenses or all responsibilities are instantly transferred in every artifact. The careful formulation is: Google announced the donation and the FIDO Alliance announced that it would develop standards related to trusted agent interactions.
For teams, the implication is a need for monitoring. A specification that changes its governance framework can change its contribution processes, its versions and its priorities. The repository and the official announcements must remain the references, with absolute dates.
10. Google UCP and the AP2-UCP codelab
Google publishes a UCP guide for merchants and developers. The existence of this official documentation provides level 3 evidence: the organization documents an integration, its concepts and its surfaces.
Google also publishes a codelab titled Secure Agent Commerce with AP2 and UCP. The tutorial links the two protocols in a public teaching scenario. Its existence is verified. It was not executed in this audit.
A codelab is useful but specific evidence. It shows a recommended path in a controlled environment. It does not guarantee the compatibility of every version, every SDK or every wire format. Nor does it constitute proof of merchant adoption.
The right reading is progressive. The documentation confirms that Google is investing in a public UCP surface. The codelab confirms that an AP2-UCP flow is being taught. The technical issues in Report 1 show that a specific compatibility still has to be tested with the actual versions and serializations.
For an integration program, these resources are enough to start a technical review. They are not enough to announce a deployment, a geographic coverage or a transaction volume.
11. Shopify: building with UCP
The Shopify documentation for agents invites developers to build commerce experiences with UCP and the Shopify MCP servers described as UCP-compliant. Shopify has also published an engineering article devoted to UCP.
On June 17, 2026, Shopify stated in its developer edition that developers could build end-to-end agentic commerce on Shopify. The announcement is accompanied by technical documentation. The claim therefore reaches level 3.
Why not level 4? Because this report did not observe a complete flow open to a defined merchant, with conditions, account and result. It verified the official resources, not the transactional experience.
The distinction also protects Shopify from excessive extrapolation. A developer environment and documented servers constitute real evidence. They do not say that every store, every theme, every payment method or every region has the same availability.
The recommendation to teams is to separate three questions: do the tools exist, does our account have access to them, and does our scenario succeed? The current sources answer the first solidly. The other two require authorized local checking.
12. UCP and AP2: compatibility as a bounded claim
The official UCP and AP2 page states that UCP is fully compatible with AP2. It describes AP2 as a trust layer for transactions carried out by agents and presents cryptographic binding benefits.
The claim reaches level 3. Official documentation explains the binding. It would be incorrect to reduce it to a marketing formula without an artifact.
It would be just as incorrect to present it as demonstrated end-to-end compatibility for every shape. UCP #599 reports that the checkout_mandate pattern rejects a string serialization produced by the reference AP2 SDK. The issue was open as of August 31, 2026.
This issue is a targeted contrary report, not a general refutation. It can be fixed through the schema, through the SDK or through a clarification of scope. As long as the public resolution is not observable, the report keeps both elements.
The defensible formulation is therefore: UCP documents compatibility with AP2 and provides a public binding; a specific serialization incompatibility is reported in the official tracker; no general conclusion about all versions is measurable here.
13. Time as an evidence variable
A product claim is not an eternal property. It is true or false within a scope and at a date. The OpenAI example shows three distinct states:
- launch of Instant Checkout for an Etsy and United States scope in September 2025;
- future announcement concerning Shopify in the same press release;
- August 2026 state favoring discovery and merchant checkout.
A register that keeps only the latest page loses history. An article that keeps only the launch loses the current state. The method must therefore record the claim date, the observation date and any replacement relationship.
Technical documentation poses another problem. It can be updated without a visible editorial date. In that case, the report must not invent a publication date. It uses the observation date and, if necessary, a repository hash or a displayed version.
Finally, temporal words must be coded. “Now”, “today”, “coming soon”, “later this year” and “over time” are not interchangeable. Their translation must preserve their force. A roadmap does not become a launch because it appears on a current page.
14. What level 3 really means
Level 3 is the center of gravity of the corpus. It attests that a reader can consult a public technical artifact. This openness has real value: schema review, prototyping, version comparison, security audit and discussion on the trackers.
But a public artifact can be incomplete, experimental or intended for a controlled environment. It can depend on partner access. It can also evolve faster than integrations. Level 3 must therefore not be rephrased as “available everywhere”.
For an executive, level 3 answers the question “is there a technical base that my team can evaluate?”. For a developer, it answers “can I read and start building?”. For a buyer, it does not yet answer “can my organization use this capability today with its accounts and its countries?”.
The internal quality of a level 3 varies. A versioned repository with tests, changelog and examples is more usable than an isolated page. This report does not create a sub-score, for lack of a sufficient sample and an independent codebook. It keeps this nuance in the comments.
The level is therefore an evidence floor, not a maturity label.
15. What level 4 really means
Level 4 adds a surface officially described as active or publicly observable. It brings the claim closer to usage, but it remains below a reproduction.
An official page can state that a service is live in the United States. That does not prove that it is visible for every user, every query or every account. Progressive rollouts, A/B tests, eligibility conditions and catalog constraints can create different experiences.
Level 4 must therefore preserve the exact scope. For the current state of ChatGPT, the claim is “shopping live in the United States, purchases completed on the merchant’s site or app”. It does not become “universal agentic checkout”.
The historical Instant Checkout case also shows that a level 4 can expire. The evidence that a surface was launched remains valid historically. It does not guarantee its continuity.
To reach level 5, a test protocol would have to be described: date, account, region, query, merchant, steps, results, errors and limitations. For level 6, authorized and redacted transactional proof would be needed in addition, with reconciliation of the result. None of these acts was authorized or necessary for this report.
16. What remains unmeasurable
The register does not measure the number of merchants active on ACP, AP2 or UCP. Announcements of potential population, integrated catalogs and a standard’s partners do not provide that number.
It does not measure transaction volume, conversion rate, failure rate or gross merchandise value. No public and comparable transactional dashboard was identified in the retained sources.
It does not measure actual coverage by country beyond the explicitly published scopes. A service described as live in the United States allows no conclusion about France, the European Union or other markets.
It does not measure the conformity of proprietary implementations to the public repositories. A company can use a different version, add controls or not expose its code.
It does not measure the overall quality of an organization. A technical report does not become a reputation grade.
Finally, the evidence level does not measure commercial value. A level 3 artifact can be strategically important. A level 4 surface can be limited. Only authorized usage data could answer these questions.
17. Reading guide for e-commerce executives
Before allocating a budget, ask for three separate proofs.
Proof of standard. Is there a public schema, repository, version and governance? The ACP, AP2, UCP and Shopify claims provide several positive answers at level 3.
Proof of access. Are our country, our platform and our account eligible today? This proof must come from an availability page or from the verified account, not from a general press release.
Proof of flow. Is our catalog discovered, is our checkout reached and is the result reconciled? This proof requires an authorized test.
The decision does not have to wait for a level 6 before beginning any exploration. It simply has to match the right investment to the right level. At level 3, funding a prototype or an architecture review can be rational. To promise a commercial launch, a level 4 or 5 check is preferable.
Finally, OpenAI’s change is a reminder that an architecture choice must remain reversible. An embedded flow can become a merchant flow again. Teams should preserve the URLs, the measurement and the experience of the proprietary checkout, even when a platform announces native conversion.
18. Reading guide for procurement
A purchasing file should atomize every vendor promise.
Instead of asking “do you support agentic commerce?”, ask which version, which region, which merchant type, which surface and which transactional state are covered. Require a link to the documentation and a test environment whenever the answer goes beyond an intention.
Answers can be classified according to the report’s scale. A public repository is worth level 3. A declared availability for our scope is worth level 4. A guided demonstration is not necessarily an independent reproduction. A successful payment still has to be reconciled.
The contract should distinguish protocol compatibility from operational responsibility. Who handles version changes, schema errors, keys, retries and proof of consent? Who observes the flow when the checkout is handed back to the merchant?
Finally, future formulations must become milestones, not purchased capabilities. “Coming soon” can support an option or a roadmap, but not a go-live date without a separate commitment.
This register does not replace legal or security due diligence. It provides a common grammar to avoid comparing an announcement to a deployment.
19. Reading guide for journalists and analysts
Each sentence should keep its original verb and its date. “Announced”, “documents”, “described as live”, “was reproduced” and “succeeded” correspond to different kinds of evidence.
Partner lists call for particular caution. They do not all say the same thing. Supporting a standard, taking part in a group, offering an SDK and operating a transaction are distinct roles.
Potential volumes must be separated from activated volumes. The announcement of more than one million Shopify merchants “coming soon” did not prove their activation. The current page states that Shopify and Etsy catalogs are integrated, but gives no count of visible products or of transactions.
Current pages must not erase history. When the product changes, publish an update linked to the original article and specify what was replaced.
Finally, an open issue can contradict a technical claim without refuting an organization. UCP #599 must be presented as a specific report in the tracker, with its status and its scope, not as proof that all AP2 compatibility is impossible.
20. Updates, corrections and right of reply
The register must evolve through versions. A factual correction that does not change the conclusion can produce a version 1.0.x. A change of surface or the addition of a group of claims justifies a minor version.
Each row has an observation date. When a page changes, the old row can remain as a historical state if its URL and its date are preserved. A new row describes the current state and references the replaced item.
Named organizations must have a correction channel after publication. A response can provide a stronger primary source, correct a scope or report that an issue is resolved. It must be evaluated according to the same codebook.
The right of reply is not a transfer of editorial control. An organization can explain its private system, but the report must distinguish between a statement received and public evidence that has been checked.
Before any publication that names organizations, a legal review is recommended. It must verify the paraphrases, the chronology and the absence of any conclusion about reputation. No external message was sent in the production of this report.
21. Limitations and threats to validity
Purposive sample. The twelve items were chosen to cover the main verifiable official claims, not to statistically represent the whole market.
Dependence on public channels. A private capability can exist without public evidence. The absence of a high level is not proof of absence.
No flow executed. No account, purchase or payment. Levels 5 and 6 remain unmeasured.
Evolving pages. Some documentation displays no publication date. The observation date is then used.
Geographic scope. US evidence is not extrapolated to Europe.
Grouped claims. Several organizations can publish converging formulations. The register deduplicates the claim but keeps the corroborating sources.
Contradictory reports. An open issue is public information, not a final resolution. It bounds a claim without establishing a global failure.
Ordinal scale. The difference between levels is not a mathematical distance. A level 4 is not “twice as good” as a level 2.
22. Conclusion
Agentic commerce already has a substantial public technical base. Nine of the twelve items reach level 3. Teams can read, compare and prototype ACP, AP2, UCP and the surfaces documented by Google and Shopify.
Public evidence of usage remains narrower. Two surfaces are officially described at level 4, but one is historical and the other reflects a change of model. No flow was reproduced in this report.
This asymmetry justifies neither blind enthusiasm nor cynical rejection. It invites every decision to be framed at the appropriate level. Documentation makes it possible to begin an evaluation. A bounded availability makes it possible to prepare a pilot. A reproduction makes it possible to plan a launch. Transactional evidence makes it possible to measure a result.
The register is complete within this scope. Its immediate usefulness lies in its temporal distinctions and its capacity for correction. The next update will have to take place if OpenAI changes its flow again, if AP2 governance evolves, if Shopify publishes a more precise access scope or if UCP #599 is resolved.
23. Primary references
- OpenAI, Buy it in ChatGPT
- ChatGPT, Power product discovery
- OpenAI Commerce documentation
- Stripe, Instant Checkout announcement
- ACP repository
- Google Cloud, AP2 announcement
- AP2 repository
- Google, AP2 and FIDO Alliance
- FIDO Alliance announcement
- Google UCP Guide
- Google codelab AP2 and UCP
- Shopify agents documentation
- Shopify engineering UCP
- Shopify Spring 2026 Edition
- UCP and AP2
- UCP #599
24. Glossary and machine-reading guide
Atomic claim: a formulation reduced to a verifiable subject, verb, object, scope and date.
Technical artifact: usable public documentation, repository, schema, SDK, example or tutorial.
Described surface: a feature that an official source presents as available within a given scope, without independent reproduction in this report.
Historical state: a claim valid at its date but replaced as a description of the present.
Contrary report: a specific public issue that bounds a claim, without automatically becoming a verdict on the whole organization or the whole protocol.
NOT MEASURABLE: information that this report’s sources, accounts and authorizations do not make it possible to establish. This status is not a zero.
For automated processing, use data/claims.csv. The fields claim_date, evidence_level, evidence_state, primary_url, corroborating_url and observed_at allow an update without extracting the PDF. The editorial conclusions and limitations remain in this Markdown.
End of report. Version 1.0.0, August 31, 2026.
Primary sources and review date
- OpenAI, Buy it in ChatGPT, reviewed 2026-10-22.
- ChatGPT, Power product discovery, reviewed 2026-10-22.
- OpenAI Commerce documentation, reviewed 2026-10-24.
- Stripe, Instant Checkout announcement, reviewed 2026-10-24.
- ACP repository, reviewed 2026-10-24.
- Google Cloud, AP2 announcement, reviewed 2026-10-24.
- AP2 repository, reviewed 2026-10-24.
- Google, AP2 and FIDO Alliance, reviewed 2026-10-24.
- FIDO Alliance announcement, reviewed 2026-10-24.
- Google UCP Guide, reviewed 2026-10-24.
- Google codelab AP2 and UCP, reviewed 2026-10-24.
- Shopify agents documentation, reviewed 2026-10-24.
- Shopify engineering UCP, reviewed 2026-10-24.
- Shopify Spring 2026 Edition, reviewed 2026-10-24.
- UCP and AP2, reviewed 2026-10-24.
- UCP #599, reviewed 2026-10-24.