Open research questions
What would change my mind about PEA?
A framework for examining AI work should face the same scrutiny it asks agents to face.
Activation before attribution
The first author-run pilot did not demonstrate the benefit I hoped to examine. All four agents passed the functional checks, but neither PEA-labeled session applied the trace workflow. That is a reason to test first-use activation, not to advertise a comparative advantage. A stronger experiment must establish that the agent actually reads and applies the relevant skills before attributing an outcome to PEA. Corrected host conditions should be recorded as a new experiment, with the unsuccessful earlier attempt still visible.
Evidence that could strengthen or weaken the proposal
I would learn from independent practitioners finding that the decision record genuinely helps them challenge a consequential choice, and equally from reports that it becomes ceremonial overhead. Cross-session continuity needs tests in which a later agent acts appropriately on earlier context rather than merely retrieving a file. Resource effectiveness requires measured work and cost on comparable tasks, not an impressive anecdote. A persuasive explanation that conflicts with the observable diff would weaken confidence. None of these questions is settled by passing the pilot’s cart tests or producing a long audit trail.
An invitation to five practitioners
I am seeking five voluntary, independent attempts on bounded, nonconfidential tasks. You can use the small public fixture or choose a task from your own permitted work. The useful questions are simple: what did you expect, what actually happened, and did the record help you inspect or challenge a decision? Would you choose to use it again? A failed installation, an abandoned attempt or a clear account of unnecessary overhead is a useful report. This invitation is not a claim that participants have already been recruited or that any organization endorses the framework.
Participation without promotional obligations
The linked report form accepts useful results and failures. Public reporting is optional: never share credentials, employer code, private transcripts or personal data merely to supply evidence. Submitting a report does not grant permission to quote you, use your name or imply employer endorsement; those decisions require separate consent. I do not equate stars, views, downloads or automated mirrors with voluntary practitioner use. The next step is to learn whether the workflow earns its place in real work, and to publish what that evidence actually supports.