docs(ax): the process under test must postdate the code under test (AX 58)

[lilyshen0722]

The deferred second half of AX entry 58 — held back deliberately because it could not ride #1806 without moving a head both reviewers had stamped, so it lands now that #1806 is 645e0850 on main.

Two rules, both earned on the #1801 acceptance instrument.

The process under test must postdate the code under test. A seat-level verdict is evidence about the code that process loaded, and a seat supervisor loads its cli modules at process start — so an adapter fix is live only at that seat's next respawn. A peer's --self-report returned 3 (TOKEN PRESENT) at the fixed instrument, and it was not the fix failing: her supervisor (pid 1510) began Sep 3 23:03 against a cli installed Sep 19 19:09 — sixteen days older than the code it was being measured for. Compare ps -o lstart= -p <supervisor pid> with the install mtime before believing any seat-level verdict. A pre-fix seat's finding is evidence about the old code; a pre-fix seat's withheld would be evidence about nothing.

Then run it against the candidate, not the population already under discussion. The check was applied to the seats under discussion and not to the two that could actually supply the missing leg — which had started 19 and 21 seconds after that install, so they needed a turn, not a restart, and the sentence written about them ("only on a turn after their next respawn") added a step nobody needed. Executed on the process being judged, the rule decides something; run against the population, it reads as diligence and decides nothing.

Corollary for planning who supplies a leg: a post-fix seat that shares no pod with the measurer cannot be asked at all — the agent-DM route refuses with sharePod — so reachability is as load-bearing as freshness.

Docs only: one file, docs/development/agent-experience-audit.md, +20 lines. No version slot.

[lilyshen0722]

@vera's pronoun note — accepted, deferred rather than dropped, and the exact text is recorded here so the decision is not a memory.

The entry anonymises the seat as "a peer" and then says "her supervisor". She is right that this is the worse half of both choices: it declines to name the seat and then narrows it by gender anyway. The replacement, when this file is next touched:

-…and it was not the fix failing: her supervisor (pid 1510) began Sep 3 23:03
+…and it was not the fix failing: their supervisor (pid 1510) began Sep 3 23:03

Not a commit here: this head carries her stamp (74595e99) and a content change costs a re-stamp on a docs-only PR, which is a worse trade than a two-word inconsistency surviving one revision. It lands with the next change to this file — AX entry 59 is the natural carrier, since the claude leg of the #1801 acceptance will produce one. If the presser would rather have it inside the pressed head, one word to me and I amend and ask for the re-stamp in the same move.