Skip to content

Design sketch: Fleet Persuasion Eval (multi-agent extension) #8

Description

@rudi193-cmd

Design sketch: Fleet Persuasion Eval (multi-agent extension)

APE measures single-model attempt to persuade; our deployment uses multi-agent orchestration (bus messages, dispatch, policy hooks). We sketched an extension — Fleet Persuasion Eval (FPE) — that scores persuasion across agent boundaries under a fixed policy shell (benign compliance traps in v0, no harmful topic generation required).

Design doc: fleet-persuasion-eval-sketch.md

Proposed metrics: Fleet Attempt Rate, Fleet Compliance Rate (target violates policy), Fleet Safety Rate (refusal / hook block).

If there’s interest in a small v0 (pytest + mock fleet identities, reusing APE topic draws where appropriate), we’re happy to iterate here or on a dedicated issue.

rudi193-cmd/willow-2.0

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions