About this article

• Audience: Operators running deploy, redeploy, soak, and optional video-understanding workflows
• Goal: After reading, you can look up the standard verification steps without rereading full guides
• Type: Reference

Summary

Use these checklists as lookup tables alongside Quickstart, Recovery, and Layer 3 guide. Commands use placeholders; substitute your device IP and password at runtime.

Prerequisites

• Phantom host stack up (MQTT 1883, proxy 8788)
• PHANTOM_PROXY=https://<phantom-proxy>:8788
• SSH: root@<device-ip> with YOUR_DEVICE_PASSWORD
• Scripts in [repo-root]/oak-vms-firmware/

Pre-deploy / bake

Step	Action	Pass signal
1	Prune if storage is CRITICAL: dry-run then real run with retention policy	Utilization drops to WARNING or OK
2	Confirm proxy health and MQTT broker respond	curl -s "$PHANTOM_PROXY/fleet/health" returns 200
3	Confirm RF-DETR models present on device or push models script	Model paths reported in bake log
4	Run deploy script	./scripts/deploy_oakapp.sh <device-ip> YOUR_DEVICE_PASSWORD completes
5	Verify registry and core pipeline	New CID in /fleet/registry; ghost_core reaches steady state in logs
6	Confirm edge app running	oakctl list or equivalent runc inspection shows active container

Post-deploy / redeploy

Step	Action	Pass signal
1	Probe registry, MQTT cache, and MCP tools	get_recording_status, get_device_identity with FULL_KEYS fields present
2	H.265 first-chunk check	Fresh chunk >1 MB mtime; fps verifier PASS at expected rate
3	Short soak or harness batch	--prune-first --visual --tap with normal_short or recovery scenario
4	Thermals and storage	Die temp within policy; storage not CRITICAL
5	Federation	Hermes peer ack; A2A precomp query returns 200 with rf_detr_precomputed
6	Handoff anchor	Update private handoff only—do not paste live tables into public docs

Detailed redeploy patterns also live in internal POST_REDEPLOY_VERIFICATION_CHECKLIST (not public-synced).

Soak / gate / visual

Step	Action	Pass signal
1	Prune-first on secondary shadows when disk is tight	Orchestrator or prune script completes
2	Run batch or soak_gate duration	60×1 min or policy duration
3	Inspect artifacts	.pass marker, summary.json, handoff_snippet in v3-runs/ batch folder
4	Visual mode	Checkpoints and annotated frames generated
5	Optional e2e	Playwright visual suite collected when enabled
6	Grade (optional)	Production soak grader: A ≥90, production ≥85 with zero critical findings

Mobile / actor (historical)

Phase-one mobile and actor checklists moved to appendices/internal-plans/. Current production focus is sovereign edge + Phantom proxy federation—not mobile-primary testing.

Video understanding gate (~5 min Phantom)

Step	Action	Pass signal
1	Stack up + readiness modules	gemini_video_understanding, video_intelligence_annotate, Hermes validate PASS
2	Pull justified clip via proxy	Playback-ready URL or pull-recording succeeds
3	Cloud annotate	Timestamps in prompts; Video Intelligence labels/objects returned
4	Fuse results	Structured output on MQTT / A2A / MCP without overriding precomp live path
5	TAP + handoff	Evidence in tap log; private handoff updated

See Video cloud architecture and Video samples.

Command quick reference

export PHANTOM_PROXY=https://<phantom-proxy>:8788

# Deploy
./scripts/deploy_oakapp.sh <device-ip> YOUR_DEVICE_PASSWORD

# Short soak
python3 tools/v3_shadow_orchestrator.py \
  --shadow <device-ip> \
  --duration 1 \
  --prune-first --visual --tap \
  --scenarios normal_short \
  --batch-id checklist-$(date +%s)

Next steps

• Recovery — when a checklist step fails
• Soak grading — longer production soaks
• Release docs — gate tiers before promote

Operator depth

Live probe commands, recovery playbooks, and exact checklist variants live in private operator handoff (not published) (private).