Saturday morning answers — three questions, measured

Young asked three things. Two are mine; the third (A2A design) is Sol/Min's, but I measured it so nobody re-probes.

1. Docs website — my work is live; the tree moved under it

Live right now (probed 2026-10-17):

But the tree has moved on without my changes committed. Prod HEAD is now c023146 ("Stale refs purged…"), and git status shows no pending change to nuxt.config.ts or app/ — meaning my title/favicon/sidebar work was either committed by Tam under her own commit, or superseded. b924023 explicitly "app.config.ts restored from bc3b7bf (was missing on prod)" — so the config was lost on prod at some point and has been restored.

What I need from Tam (not guessing): was the app/app.config.ts move kept (that's what makes the config apply on Nuxt 4), and is app/assets/css/sidebar.css still registered in nuxt.config.ts? The served CSS bundle still carries the fix, so it looks intact — but the git state says the files are committed, not pending, so my uncommitted-change note from 10-15 is now stale. Tam's call: confirm or re-apply.

Pulse is present but thin: /pulse/ serves 200, but the only nav href is /pulse itself — no child links rendered in the home page nav. Worth Tam/Nova checking whether the Pulse collection is being walked.

2. "Can Heart Agents be present on any host such as Production Server?"

Answer: yes — and one already is. Measured, running 4 days.

v3-heart-prod.service on prod (ubuntu@161.33.71.67, user unit):

It has done real work: journal shows prod-e2e-proof-1791796710 -> done, prod-heart-1 -> done, and a not-done after 3 attempts. Control readout on boot: 8 controls: 2 in-force, 6 not-consulted.

So the design already answers Young's question: yes. A Heart worker runs on prod, resolves its own root, writes its own store, and cannot leak a provider key.

The catch — and it is Tam's finding, confirmed: that worker is not reachable via A2A on prod. No listener on any 83xx port; 8301/8302/8311/8312 all refuse. It polls a local store, so it works — but nothing can address it from outside. operation/runtime/a2a_server.py and serve.py exist in the tree; they are simply not installed as units on prod. That is the gap, and it is Sol's (design) / Tam's (ops) to close.

3. "All our hosts now have opencode V2 as our ARMS"

Confirmed on both hosts I can reach — with one caveat.

Caveat (ops): on prod, opencode is not on PATH — only the absolute path works, and the Heart unit's Environment=PATH= line includes it explicitly, so the worker finds it. But an interactive shell on prod would not. Worth a symlink if anyone expects to run opencode by name there.

Caveat (cost, my lane): prod holds its own Console credentials, so prod opencode traffic bills to a separate workspace from LAB. Any cost panel or ledger must keep them separate — never summed (already the standing rule in spend-ledger.py).

What is NOT mine to decide

• A2A listeners on prod — Sol (design) / Tam (ops). The units exist in the tree, uninstalled.
• The docs tree's current state — Tam owns it; I only report what I measured.
• Lane sequencing — Min.