Update review · November 19, 2026
Identity Verification Methods Values Update (-04)
Revision -04 adds one new ivm value, one new claim, and the draft's first example. Two items from the previous review were picked up: there is now a value for biometric binding, and there is now a way to say when verification happened. Everything else is still open.
What Changed
lfm(Live Facial Match to Identity Evidence) is a new method value. The registry now has fourteen values.idv_time(Identity Verification Time) is a new JWT claim, registered alongsideivm.- Section 3.4 is a new example showing
"ivm": ["phy", "lfm"]. - New informative references to ISO/IEC 30107-1 and NIST SP 800-63A-4.
Nothing was removed. The authors are the same.
The lfm Value
This is the fix for the biggest gap in the person-oriented values. Before -04, phy said a document was checked and nothing said the person holding it was the person on it. The new example says this directly: without lfm, the claim says "only that the document was verified, not that the person presenting it is the person it identifies."
The definition is the most precise one in the draft. It requires an automated comparison of a live capture against the portrait on the identity evidence, with presentation attack detection applied to the live capture. It points at ISO/IEC 30107-1 for PAD and at NIST SP 800-63A-4 for the method.
There are still problems.
It has no meaning on its own. The draft says lfm is "typically used together with" phy or dig. That's a description, not a rule. An array of just ["lfm"] says a face was matched to evidence that nobody verified.
It doesn't say which evidence. In ["phy", "dig", "lfm"], the face was matched to the passport, the mobile driver's license, or both. The array can't say.
It covers only automated matching. A clerk comparing a face to a passport at a counter is not lfm. Neither is an agent on a video call. So does inp or vid imply the person was matched to the document? The draft doesn't say. The binding gap is closed for the selfie flow and still open for the two human-operated flows.
There is no performance floor. ISO/IEC 30107-1 is a framework. It defines terms, not pass criteria. Two issuers emitting lfm can have very different false match rates and very different resistance to deepfake injection.
It makes the organization problem worse. In ["brv", "rep", "phy", "lfm"], whose face was matched? Probably the representative's. The flat array still can't say that.
The idv_time Claim
idv_time is an optional NumericDate, modeled on auth_time. It is the time the issuer "had completed all steps required to assert that result." The draft is explicit that it isn't the earliest or latest evidence-collection event.
This addresses the "when" part of the previous review's context complaint. "By whom" and "which attributes" are still missing.
One timestamp covers the whole array. An issuer that ran dbv in 2023 and added lfm last week has one idv_time for both. The draft's wording suggests it's last week. A relying party can't tell that the database check is three years old.
There is no guidance on using it. auth_time has max_age on the request side. idv_time has nothing. The draft doesn't say what a relying party should do with an old value, or whether an expired passport changes anything.
It doesn't work with KYAPay's structure. KYAPay token nests ivm inside hid and apd and bans it at the top level. idv_time is defined relative to "the accompanying ivm claim," so it has to be nested in the same object. This draft doesn't say that. KYAPay -04 doesn't mention idv_time at all.
Small things. The draft doesn't say whether idv_time can appear without ivm. The text refers to "the verifier's record," and "verifier" isn't defined anywhere. The new example doesn't include idv_time. The IANA section still says it registers "the following Claim," singular.
Was the Previous Feedback Addressed?
Two items, partly.
- No biometric binding value. Fixed for automated matching with
lfm. - No context.
idv_timecovers when. Who verified and which attributes were verified are still missing.
Everything else is unchanged.
- No mapping to OpenID IDA. IDA already has
verification.timeand a time on each piece of evidence.idv_timeis one more overlap with no mapping. dbv,dbv1, anddbvmare still three values for one method.- No combination rules.
lfmadds a new one to the list. - "Whose identity?" The abstract and introduction still say "the person's identity." Five values still describe an organization. The new example puts
ivmat the top level next tosub, which is the layout KYAPay prohibits. repstill mixes identity and authority.bovstill has no threshold.tinis still organization-only.- US-centric definitions. "Consumer reporting data sources" is unchanged.
- Template errors. The registration template still describes "the authentication method." RFC 5226 is still cited instead of RFC 8126. The change controller for
ivmis still Mike Jones personally, and the newidv_timeregistration copies that. - Security and privacy considerations. Each is still one sentence pointing at JWT.
Writing Quality
The lfm definition is good. It says what was compared, how it was captured, and what protection was applied. The older values should be rewritten to the same standard.
The example is welcome. It's the first one in the draft.
The privacy section needs attention now more than before. lfm in a token tells every recipient that the issuer processed the subject's face. Biometric data is a special category under GDPR, and it's regulated by laws like Illinois BIPA. The draft adds a biometric method and says nothing new about privacy.
Bottom Line
This is the first revision that responds to review feedback, and lfm is a real improvement. idv_time is a reasonable start that needs rules for nesting and for mixed-age verifications.
Feedback should focus on three things. Say whether inp and vid include matching the person to the document, so binding is covered in every flow. Define where idv_time goes when ivm is nested, and coordinate that with KYAPay. Then return to the open items from the last two reviews: one subject per ivm array, a mapping to OpenID IDA, and privacy considerations that cover biometrics.