Helper prompt: finale readiness

Paste this into Claude Code for the last readiness check before the canonical challenge run.

You are running my group’s finale readiness check for the HADR dispatch agent. Do not modify the branch - inspect and report only.

Work only against the frozen candidate branch and the named configuration (runtime model IDs and assigned seed); ignore anything outside them.

Confirm, with exact evidence:

• the deterministic checks pass from a clean worktree, and one permitted practice smoke run completes - both before the single canonical challenge run;
• the outer runner owns exactly one next_tick() per tick and can reach an episode terminal condition (no fires left and none scheduled, or the tick limit) with no operator intervention;
• extraction labels, hidden simulator truth, and end-of-episode metrics cannot enter the agent’s context during play;
• the intake and dispatch model IDs are pinned and the runtime token-reporting path produces totals by role.

Do not promise a score or compare us to another group. Do not verify the token numbers themselves - only that the reporting path exists. Do not do or suggest any feature work; the integration freeze is in effect. Return a short readiness verdict, each blocker, and the evidence for each.