Summary

  • The IETF's vCon working group has discussed moving from recording-centred references toward a session timeline with discrete events and optional nested sessions. Its 9 September minutes call this a direction to be confirmed, not an approved revision.
  • A separate 11 September proposal by Jonathan Rosenberg illustrates calls, conference sidebars and AI subagents. Datatracker identifies it as an individual Internet-Draft without formal IETF standing.
  • The issue is whether a portable record can show who acted during gaps between media fragments, and whether that history remains intelligible after redaction, without breaking existing vCon readers.

A caller is placed on hold; two support agents consult while separate audio tracks run; a voice agent invokes a tool between recorded utterances. If the portable file is built mainly around pieces of media, where does the whole episode live? That is the practical question underneath the Virtualized Conversations working group's current data-model discussion. It also sets a limit on an attractive governance claim: a container can preserve a trace, but it cannot promise completeness merely because it contains recordings.

The group's approved charter seeks interoperable movement of conversational data across systems while protecting personal information. Its active core draft, revision 04, describes parties, dialogs, analysis and attachments, with mechanisms for signing, encryption, redaction and amendment. It is still an Internet-Draft, not an RFC. The charter's tension is deliberate: the same record useful for migration or analysis may expose the person who spoke, the content they supplied and the later interpretation attached to it.

The 9 September interim minutes identify a concrete defect in a recording-led structure. Separate tracks may not tell a reader who else was present; short talkspurts may not span the beginning or end of a call. A join, a transfer or a keypad action can occur in a silence gap with no dialog to carry the event. Participants considered a recording set, then a more general session as the master timeline. In the reported direction, events would no longer have to borrow a recording's duration. A session could hold dialogs, events and child sessions, while parties would be defined once at the top level.

This is not an adopted schema. The minutes label the direction “to be confirmed on the list.” They leave open whether an event may also belong to a dialog. They record a second problem: once sessions nest, an array index does not identify a dialog unambiguously across the hierarchy. The proposal to assign document-wide object identifiers could make a reference survive removal of sensitive material, but the minutes call it a breaking change. Existing implementations already use vCon, and the group explicitly wants backward compatibility. Those facts belong in the same sentence as the design's appeal.

Rosenberg's 11 September individual draft gives the idea an intelligible shape. A session has its own start, end and participants; events mark point-in-time actions such as tool requests, results and transfers; dialogs contain exchanged media or text; a child session can represent a meeting sidebar or an agent's delegated task. The draft says this is an illustrative information model, not an implementable specification. Datatracker says anyone may file such a draft and that this one has no IETF endorsement. Calling it the vCon standard would erase the very decision the group has yet to take.

Redaction makes the reference choice more than a tidy-data issue. Revision 04 says a redacted vCon should refer to its less-redacted predecessor and that access to a referenced unredacted copy must be restricted. It also recommends empty placeholders when removing array elements so later indexes do not move. The September minutes' identifier proposal would avoid rewriting references in some removals, but whether and how that becomes part of the core is unresolved. Neither representation proves that omitted content was lawfully collected, that all events were captured, or that a recipient is authorized to use it.

The group met again on 24 September; the reviewed public meeting listing has materials but no approved minutes from which to infer an outcome. The published agenda for 8 October promises continued discussion of core updates, no more. The useful public test between now and a revised draft is therefore not whether a diagram looks elegant. It is whether the record states its scope: which sessions and event types are represented, who can attest to capture, what was removed, how old readers behave, and which separate institution sets consent and retention rules.

That test is BTW's editorial recommendation, not a requirement already issued by the IETF.

Sources