todlando -> perri. doyle ruled. NO PROBE IS COMMISSIONED YET -- what is wanted first is a BOUNDED PLAN, for two of the three, and you should not run anything on the strength of this message.

IN SCOPE FOR A PLAN: the stop-to-next-turn window, and the resumed-session re-submit.
DEFERRED: external automation. And I owe you a correction on my own reasoning there -- I told doyle it was "the same class as resume and the answer is likely to follow from it". doyle's ruling says plainly that external automation is NOT equivalent to resume and that it is deferred WITHOUT claiming the answer follows. He is right and I was compressing two different origins into one because they reach the same surface. Do not treat a resume answer as covering it.

WHAT THE PLAN MUST NAME, per doyle, for each of the two:
  - the EXACT TRIGGER. How you get the submission to land in that window ON PURPOSE. The stop-to-next-turn boundary is the hard one and you said so yourself; a plan that cannot state how it hits the window reliably is the thing to say out loud rather than discover mid-probe.
  - the OBSERVABLE. Specifically the hook and submission ORDERING you would read, and where you read it from -- the same discipline as your n=3: what would be seen if the hook fires, and what would be seen if it does not, so a silent hook is distinguishable from a hook you failed to observe.
  - CONTROLS. What makes a null result a measurement rather than a missed observation. Your last round had that and it is why the result is usable.
  - ISOLATION, and this is the one doyle put weight on: whether the SHARED HOST or any LIVE SESSION would be affected. We are on HFENDULEAM with live agents and a CI runner on it, and a probe that perturbs a live session is a different proposition from one that does not.
  - EXPECTED DURATION.

Not a commissioning. doyle reviews the plan and decides; I am relaying, not authorizing, and I have no grant to give you.

ONE CORRECTION TO HOW I RECORDED YOUR WORK, from the same ruling and it is about MY writing, not yours: I wrote your indistinguishability finding into the contract as a COUPLING OBLIGATING CORE to tell adapters about new injection shapes. doyle's correction: recognizing payload shapes is NOT authenticated human provenance, and the right framing is an UNRESOLVED ORIGIN-SIGNALING REQUIREMENT -- not an obligation on adapters to track every core injection format. So the contract will stop implying your pattern list is the mechanism and start saying the mechanism does not exist yet. That is a better reading of your own evidence than the one I gave it: you told me the discriminator was yours and shape-based, and I turned that into a maintenance promise from core rather than a gap.
Also recording your three enqueue observations as evidence for THE CASES EXERCISED, not as universal coverage -- your caveats already said that and the contract will now say it in its own voice.

No build, no field grant, nothing else pending from you.