Async interruption response misattribution

open

1 independent reportacross 1 sourceon 1 stacklast reported 2026-07-26

What this is

A delayed decision about whether an earlier overlap counted as an interruption gets wrongly applied to a later, unrelated moment, cutting the agent off for speech that was never actually judged an interruption.

Status open: reported and still open where it was reported.

Evidence

toubatbrian2026-07-26livekitGitHub issuesgithub.com/…/2119

...is accepted and attributed to that later overlap — which can emit a false isInterruption: true and cut off the agent for a user utterance the model never scored that...

First seen 2026-07-26 · our copy fetched 2026-07-26T03:05:24Z · SHA-256 of that copy begins a5091d4d64d4c61e… · profile

Quotes run to 30 words around the reported failure and link to the post.Add your call to this entry Take a quote down within 24 hours.

What is measurable

Signal	`hotato autopsy ./call.wav` maps to this mode as detector kind `overlap_while_agent_talking`. Printed as BARGE-IN: the caller took the floor and the agent kept speaking, reported with the overlap in seconds. Needs the caller and agent on separate channels.
Establishes	The seconds of overlap where the caller had taken the floor and the agent kept speaking, with the timestamp in the call where it started. What else it establishes, and what your own trace settles The same recording produces the same overlap on every run, so the number is comparable across builds. What your trace settles Your pipeline trace at that timestamp names what held the floor: the interrupt handler, the VAD threshold, or the realtime API's own turn logic. Whether the caller meant to interrupt is a judgement about the conversation; the measurement covers who held the floor and for how long.
Tier	Tier 1 · dual-channel audio: caller and agent separated at capture, so overlap and timing are facts of the recording, exact to the sample. hotato is MIT and runs locally, getting started.

Status history

The record has carried one status since it opened: open, reported and still open where it was reported. A move lands here with the link, the quote and the date that show it.

Mode	Status	Shares
Barge-in teardown race	acknowledged	reported on livekit; measured by the same detector (overlap_while_agent_talking); reported on github
Tool call mid-utterance audio truncation	open	reported on livekit; reported on github
Unanswered tool call blocks mic input	open	reported on livekit; reported on github

Discussion

Reader discussion of this mode: what fixed it and what did not. The reports above are attributed and linked; a comment is a comment.

Open this discussion on GitHub

The record

Ids, digests, every link this entry stands on, and how to take a quote down

Atlas id	`VFA-2026-0016`
Mode id	`pat_e056fac7e6cf3eb1621eff88`
Content digest	`sha256:e056fac7e6cf3eb1621eff88027176e4a8f38be862961b8bf306717887b52646`
First reported	`2026-07-26`
Last reported	`2026-07-26`
First seen by the atlas	`2026-07-26T02:09:49Z`
Last re-checked	`2026-07-26T02:09:49Z`
Reported on	livekit
Sources	GitHub issues
Scope	1 independent public report across 1 source. This entry states where the mode has been reported; where else it can occur is what the next report settles.
References	reporthttps://github.com/livekit/agents-js/issues/2119 · toubatbrian · GitHub issues · 2026-07-26
Machine surfaces	feed.json feed.atom feed.schema.json atlas.llms.txt
Removal	If you wrote something quoted on this page and want it gone, one email to [email protected] does it. The removal path takes the quote, the handle and the link off this page, out of the feeds and out of the llms.txt within 24 hours, and tombstones the id so the entry cannot quietly return at a new address.