2026-08-04: First-look — a new individual I-D adds a receiver→sender delivery-quality feedback mechanism to MOQT, the first MoQ feedback/quality-signalling draft the wiki has tracked.
draft-liu-moq-feedback-00“MoQ Feedback” was submitted July 31 2026 (41 pages) by Yanmei Liu and Minghui Jiang (Alibaba Inc.) with Ronghua Wu (Ant Group), and announced to the moq@ietf.org list Aug 3 (04:10 UTC) — drawing an immediate reply from Cullen Jennings (16:18 UTC). Individual draft, not adopted; logged as a first-look. See discussions-2026-08.
draft-liu-moq-feedback-00 | individual submission | submitted 2026-07-31, 41 pages | announced to moq@ietf.org 2026-08-03
Authors
- Yanmei Liu (Alibaba Inc.)
- Minghui Jiang (Alibaba Inc.)
- Ronghua Wu (Ant Group)
Abstract (summary)
Defines an MOQT extension that lets receivers report delivery-quality information back to senders. The proposal introduces Multimodal Feedback (MMF), which operates at the MoQ semantic layer to supplement — not replace — packet-level QUIC transport feedback. Rather than adding new control messages, MMF reuses the existing Track/Object data model: delivery-status reports are carried on a dedicated “Feedback Track” as per-Object status objects.
Key Ideas
- Feedback Track instead of new control messages. Per-Object delivery-status reports (e.g. received / late / lost / partially received) ride the normal Track/Object hierarchy, so no new wire message type is introduced and relays need no special handling.
- Fills a transport blind spot. QUIC’s transport-layer feedback (ACKs, loss, RTT) cannot perceive frame integrity, frame type, or deadline — MMF surfaces those application-visible facts to the sender for adaptation decisions.
- Three-layer architecture. The MoQ layer synthesizes MMF signals with the local congestion-control output and issues commands (target bitrate, pacing rate) down to the CC algorithm — a semantic-layer signal informing transport-layer pacing.
- Two target scenarios. Supports both unidirectional live streaming and bidirectional AI-inference sessions (the latter tying into the same author’s Live Agent Interaction work).
Why it matters
- First delivery-quality / feedback draft in the WG’s document space. MoQ has so far leaned on QUIC transport feedback plus application-level filters (SWITCH_FROM, Top Tracks Filter, SSTS); this proposes an explicit object-granular quality channel back to the publisher.
- Timely against live design threads. It lands the same window as the Slack DATAGRAM object-fragmentation / partial-reliability discussion (afrind, Mathis Engelbart, Luke Curley) and Ian Swett’s July “Top Tracks and SSTS (or ABR in general)” list thread — all circling how a MoQ sender should learn what actually arrived and adapt.
- Continued Alibaba/Ant MoQ investment. A second Alibaba-authored MoQ artifact from Yanmei Liu (after moq-live-agent-interaction), alongside the XQUIC implementation — reinforcing the “MoQ as a reusable real-time substrate” thesis, here extended toward AI-inference feedback.
Status & Caveats
- Individual draft at -00 (submitted July 31 2026) — not adopted, no WG call for adoption. First list traffic is the Aug-3 announcement + Cullen Jennings’ reply.
- The wiki tracks individual drafts that are actively discussed or referenced; this is logged as a first-look because it opens a new feedback/quality-signalling category for MoQ. Watch for WG-list reaction, a follow-on revision, or reference from an implementation.
Related
- moq-transport — the transport this extension augments (MMF reuses its Track/Object model as a Feedback Track)
- moq-live-agent-interaction — the same author’s AI-agent application profile; MMF’s bidirectional-inference scenario ties into it
- switch-abr / adaptive-bitrate — existing MoQ adaptation signals this complements
- xquic-moq — Alibaba’s MoQ implementation
- subgroups-and-objects — the Group/Subgroup/Object data model MMF’s Feedback Track rides on