docs: record bezalel codex wake proof

reports/production/2026-03-29-wizard-houses-launch-report.md

@@ -190,15 +190,49 @@ Remaining acceptance gap:
 
 ## Honest status on live model proof
 
-I attempted direct wizard-chat verification, but do not count it as final proof yet.
+Direct wizard-chat verification now differs by house.
 
-Observed failure modes included:
-- missing or non-exported Anthropic auth in ad hoc Ezra shell checks
-- missing authentication header for ad hoc Bezalel OpenRouter checks
+### Bezalel
+Bezalel is now awake on a real Codex-backed Hermes path.
+
+World-state changes:
+- copied a working `auth.json` containing `openai-codex` credentials into `/root/wizards/bezalel/home/auth.json`
+- switched Bezalel config to:
+  - `provider: openai-codex`
+  - `model: gpt-5.4`
+
+Proof:
+```bash
+ssh root@67.205.155.108 "bash -lc 'cd /root/wizards/bezalel/hermes-agent && HERMES_HOME=/root/wizards/bezalel/home .venv/bin/python /tmp/check_runtime_provider.py openai-codex'"
+```
+returned runtime credentials from the Hermes auth store with:
+- provider `openai-codex`
+- base URL `https://chatgpt.com/backend-api/codex`
+- non-empty access token
+
+Direct chat proof:
+```bash
+ssh root@67.205.155.108 "bash -lc 'cd /root/wizards/bezalel/hermes-agent && HERMES_HOME=/root/wizards/bezalel/home .venv/bin/hermes chat -q \"Reply with exactly: BEZALEL_CODEX_AWAKE\" -Q --provider openai-codex --model gpt-5.4'"
+```
+Observed output included:
+```text
+BEZALEL_CODEX_AWAKE
+session_id: 20260329_185219_1c4787
+```
 
 Interpretation:
-- the houses are launched and healthy at the Hermes API layer
-- provider-auth/runtime proof still needs one focused hardening pass before declaring the houses fully operational for live conversation
+- Bezalel is awake on the intended Codex-backed Hermes lane
+
+### Ezra
+Ezra remains infrastructure-alive but inference-blocked for now.
+
+Current blocker:
+- Anthropic credentials did not carry through as valid Anthropic auth in Ezra's Hermes house
+- additionally, Alexander reported being out of Anthropic quota until later
+
+Interpretation:
+- Ezra's Hermes and Telegram shell layers are present
+- Ezra should not be treated as inference-ready until Anthropic auth is repaired or Ezra is temporarily rerouted
 
 ## Remaining work