Skip to content

agent_simulation: sample scenarios within a run for pass@k / pass^k - #1799

Open
u9g wants to merge 4 commits into
mainfrom
jason/simulation-repeats
Open

u9g wants to merge 4 commits into
mainfrom
jason/simulation-repeats

Conversation

@u9g

@u9g u9g commented Sep 14, 2026

Copy link
Copy Markdown
Contributor

A run today holds exactly one job per scenario, so a single flaky attempt is indistinguishable from a broken scenario.

  • SimulationRun.Create.Request.samples: attempts per scenario (0/unset = 1), requires scenario_group. Mirrored on SimulationRun.samples.
  • SimulationRun.Job.scenario_id / sample: every attempt of one scenario shares the id, so a run's jobs group back into scenarios. New field numbers; 9 stays reserved.

Consumers: livekit/agents-private (agent-service fan-out and aggregation) and livekit/livekit-cli (lk agent simulate --samples), PRs to follow.

@changeset-bot

changeset-bot Bot commented Sep 14, 2026

Copy link
Copy Markdown

🦋 Changeset detected

Latest commit: 644a4c0

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 2 packages
Name Type
github.com/livekit/protocol Patch
@livekit/protocol Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

Comment thread protobufs/livekit_agent_simulation.proto Outdated
Comment thread protobufs/livekit_agent_simulation.proto Outdated
u9g added a commit to livekit/livekit-cli that referenced this pull request Sep 15, 2026
@u9g u9g changed the title agent_simulation: repeat scenarios within a run for pass@k / pass^k agent_simulation: sample scenarios within a run for pass@k / pass^k Sep 15, 2026
u9g added a commit to livekit/livekit-cli that referenced this pull request Sep 15, 2026
@u9g
u9g force-pushed the jason/simulation-repeats branch 3 times, most recently from 5c16cbf to 0ef9b88 Compare September 15, 2026 19:23
u9g added 3 commits September 15, 2026 15:34
A run today holds exactly one job per scenario, so a single flaky attempt
is indistinguishable from a broken scenario. `Create.Request.repeats` asks
for k attempts of every scenario in one run; each Job carries the
`scenario_id` and 1-based `attempt` it belongs to so jobs group back into
scenarios. `RunMetrics.pass_at_k` (at least one attempt passed) and
`pass_pow_k` (every attempt passed) summarize the run over scenarios.
@u9g
u9g force-pushed the jason/simulation-repeats branch from 66effb8 to be5b074 Compare September 15, 2026 19:34
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants