Clear Path's five DynamX articles contain clinically accurate performance data. None of them contain U.S. investigational boundary language in the article body. On Signal's May XAA guidance was precisely calibrated. Okta's June 23 ecosystem announcement shifted the positioning surface, and nothing in Anchor's architecture noticed. The failure mode is identical across both deployments: a well-sourced claim detaches from the boundary condition that makes it safe to use. This is a category 1 structural pattern. Anchor needs boundary metadata attached to high-stakes claims, because the same sentence can be educational content in one jurisdiction and a compliance liability in another.
The signal
DynamX is Elixir Medical's coronary bioadaptor, a stent-like device designed to restore vessel motion after implantation. The clinical evidence is strong. The INFINITY-SWEDEHEART trial enrolled 2,400 patients and showed a 48% reduction in target lesion failure (TLF, the rate at which the treated artery segment fails) through two years versus permanent drug-eluting stents. Eight peer-reviewed papers on DynamX or bioadaptor technology have been indexed on PubMed since February 2026.
DynamX is also investigational in the United States. Elixir's own homepage carries the required disclosure: "Limited by Federal (or United States) law to investigational use." The openFDA PMA database returns no matches for Elixir Medical Corporation. Under 21 CFR 812.7, a sponsor shall not:
"promote or test market an investigational device" before FDA approval, and shall not "represent that an investigational device is safe or effective for the purposes for which it is being investigated."
These two facts coexist without tension as long as they stay bound together. The clinical data is real. The regulatory boundary is real.
The stress point
Clear Path published five articles presenting DynamX performance data in January 2026. I checked each article body as rendered publicly. "Five Competitive Advantages" presents 0.3% vs. 1.7% TLF at 6–12 months and zero definite thrombosis through 36 months. "Five Questions That Drive Platform Selection" states that INFINITY-SWEDEHEART documented zero target lesion failures in small vessels and says "this translates evidence into procedure." A fictionalized adoption scene presents three-year results through a cardiologist character.
None of these article bodies contain the words "investigational," "not available for sale," "United States," "U.S.," or "FDA."
I want to be precise about what I can and cannot see. A disclaimer may exist in a template wrapper or private artifact the public URL doesn't expose. But that possibility makes the problem worse, not better. If the boundary lives in a template while the claim lives in the article body, any act of extraction strips them apart: a cardiologist copying a performance figure into a presentation, a sales team pulling a paragraph into an email, a social post quoting the headline number. The boundary survives only if it is architecturally attached to the claim.
On Signal's May 2026 issue shows the same vulnerability from a different direction. The publication correctly marked Cross App Access (XAA, Okta's protocol for letting AI agents access multiple applications under a single identity authorization) as "Early Access" and told sellers: "Do not promise it to buyers, period." On June 23, Okta announced 25+ XAA integrations and described XAA as "formally incorporated as an official MCP authorization extension." MCP (Model Context Protocol) is the open standard for connecting AI agents to external tools and data sources. The MCP specification itself says enterprise-managed authorization extensions are "opt-in" and support "varies by client." Neither On Signal nor The Context Window has published a second issue. On Signal's guidance is five weeks old and was overtaken by Okta's June 23 announcement eight days ago. A seller carrying that May guidance into a July conversation holds a boundary calibrated to a world where XAA had no named partners and no public timeline.
Anchor treats a claim as a unit of text with a source and a date. It has no concept of a claim's permitted-use boundary, and no invalidation trigger for the external event that would require the boundary to be re-evaluated.
The implication
This is a category 1 finding: the same structural gap across two unrelated deployments with different regulatory and commercial contexts. In Elixir's domain, the stakes are compliance under federal device regulation. In Okta's domain, the stakes are seller credibility in enterprise procurement. The mechanism is the same.
The pharmaceutical industry solved this problem decades ago. A drug's clinical efficacy data and its prescribing information are physically bound in the package insert. You cannot legally distribute one without the other. The binding is regulatory architecture, enforced by law.
Anchor needs the equivalent. Concretely, three fields attached to any high-stakes claim:
Permitted-use context. Where this claim can safely appear. "Educational content for non-U.S. clinicians" carries different permissions than "sales enablement for U.S. enterprise buyers." The same sentence about DynamX's TLF reduction is appropriate in the first context and potentially violates 21 CFR 812.7 in the second.
Boundary statement. What must accompany the claim when it travels. For DynamX: "Investigational device. Not available for sale in the United States." For XAA: "Early Access. Support varies by client and is not active by default." These are load-bearing components of the claim's meaning. They belong inline, attached to the performance figure or status designation, visible in every context where the claim appears.
Invalidation trigger. The external event that requires re-evaluation. For DynamX: an FDA PMA filing or approval. For XAA: an Okta announcement changing GA status or publishing a deployment timeline.
This goes beyond the claim-lifecycle gap identified in the June 27 analysis. Claim-lifecycle tracking monitors whether a source has moved. Boundary metadata covers a different axis: whether the contexts in which a claim can safely appear have shifted, even when the underlying data remains formally unchanged. The DynamX performance data and the XAA Early Access status are both formally intact. The boundaries around both have shifted, and Anchor has no architecture to notice.
The test
Two events within the next publication cycle. First: if Clear Path publishes issue 2 and the DynamX performance claims still carry no inline U.S. investigational boundary in the article body, the gap is architectural. An editor who publishes twice without the boundary is following the system as built. The system doesn't require it. Second: if On Signal publishes issue 2 and the XAA status language reflects Okta's June 23 announcement without also carrying the MCP specification's "opt-in, varies by client" qualifier, the invalidation-trigger gap is confirmed in the opposite direction. The boundary will have been updated on one axis while losing precision on another.
If both publications update and both boundaries appear correctly, the founders can conclude that editorial discipline is currently compensating for architectural absence. That is survivable at six customers. It does not survive at sixty.
- Hysteresis figure custody failures: The Hysteresis issue 2 reportedly rendered AeroVironment's P550 contract as $132M instead of the actual $13.2M base/$42M total, a 10x inflation that shows boundary loss extending from permitted-use context to source-custodied figures.
- Eight post-January DynamX publications: PubMed indexes eight bioadaptor papers since February 2026, including the DYNAMITE complex-lesion study and a meta-analysis of bioadaptor versus drug-eluting stent trials, meaning Clear Path's evidence surface has moved substantially while the publication has not.
- TinyFish standards-clock parallel: W3C published a WebDriver BiDi working draft on June 29, 2026, four days after TinyFish issue 35, suggesting that browser-agent publications face a version of the same boundary problem where spec-level changes can shift what claims about agent control surfaces remain current.
- Docket as the counterexample: CEP reports that over 75% of nonprofit leaders want more transparency about foundation impact, which suggests Docket's dependency object may be circulation evidence rather than claim custody, making it the deployment where this boundary-metadata architecture would not apply.