name

ONEVO System Audit

overview

Complete end-to-end audit of the ONEVO Phase 1A Camera Loss Prevention System. The core pipeline (Connector → S3 → Cloud AI → Backend → Dashboard) is largely implemented, but critical connector bugs, security gaps, missing production hardening, and absent customer-facing features block a safe customer release.

todos

id	content	status
fix-p0-connector	Fix P0 connector bugs: wizard key (B-01), duplicate cameras (B-02), admin client 503 (B-03), installer RTSP skip (B-04)	completed

id	content	status
security-hardening	Rotate default secrets, secure connector admin :8099, fix zones/SSE tenant filtering (S-01 through S-06)	completed

id	content	status
production-infra	TLS for API/dashboard/MinIO, GPU VM provisioning, replace EnsureCreated with EF migrations	completed

id	content	status
dashboard-integration	Fix remote connector snapshot (B-05), storeId query params (B-08), SSE reconnect (B-13)	completed

id	content	status
missing-features	Build Analytics, Reports, Logs pages; password reset; connector auto-update; monitoring	completed

isProject

false

ONEVO Phase 1A — Complete System Audit

Repository: camera-phase-1
Product: ONEVO Phase 1A — retail camera loss-prevention (evidence-based risk indicators, no auto-theft confirmation)
Audit date: July 29, 2026
Production readiness score: 58 / 100 — functional MVP for controlled pilots; not ready for broad customer release without fixes below.

1. Architecture Explanation

ONEVO is a four-tier edge-to-cloud pipeline with a Windows edge agent and a cloud stack deployed via Docker on a single Azure VM (MVP).

flowchart TB
  subgraph shopPC [Shop PC - Windows]
    Installer[Inno Setup Installer]
    Connector[Connector Python Service]
    SQLite[(SQLite Upload Queue)]
    AdminUI[Admin UI :8099]
    Tray[System Tray]
    Installer --> Connector
    Connector --> SQLite
    Connector --> AdminUI
    Connector --> Tray
  end

  subgraph cameras [Cameras]
    RTSP[IP Camera RTSP/ONVIF]
    TestMP4[Test MP4 File]
  end

  RTSP --> Connector
  TestMP4 --> Connector

  subgraph cloudVM [Azure VM - Docker Compose]
    Backend[.NET 8 Backend :8080]
    Postgres[(PostgreSQL)]
    Redis[(Redis Job Queue)]
    MinIO[(MinIO S3)]
    CloudAI[Cloud AI Worker YOLOE]
    Dashboard[Angular Dashboard nginx :4200]
  end

  Connector -->|"register, heartbeat, claim, cameras"| Backend
  Connector -->|"presigned PUT clips"| MinIO
  Connector -->|"complete clip"| Backend
  Backend -->|"enqueue clip job"| Redis
  CloudAI -->|"BRPOP jobs"| Redis
  CloudAI -->|"download clip"| MinIO
  CloudAI -->|"GET zones, POST ai-events"| Backend
  Backend --> Postgres
  Backend -->|"SSE alerts, email"| Dashboard
  Dashboard -->|"JWT REST /api/*"| Backend

  subgraph distribution [Distribution]
    Vercel[installer-site Vercel]
    GHA[GitHub Actions CI/CD]
  end

  GHA --> cloudVM
  Vercel -->|"EXE + latest.json"| Installer
  Dashboard -->|"download installer"| Vercel

Module responsibilities

Module	Path	Role
Installer	connector/installer/	Inno Setup + PyInstaller + WinSW; bakes backend URL; writes `%ProgramData%\ONEVO\Connector\config.json`; registers Windows service
Connector	connector/app/	RTSP/ONVIF capture, motion clip cutting, SQLite durable queue, S3 upload, heartbeat, local admin UI
Cloud AI	cloud-ai/app/	Redis consumer; YOLOE/YOLO26/RF-DETR detection; ByteTrack; zone mapping; Re-ID; posts events
Backend	backend/	JWT auth, multi-tenant RBAC, connector pairing, clip lifecycle, Risk Engine V4, alerts, SSE, email
Dashboard	dashboard/	Angular 19 SPA — onboarding, alerts, clips, zones, tuning, health, admin
Infra	infra/mvp/, docker-compose*.yml	Azure VM, ACR, nginx, prod overrides

Data flow (happy path)

Admin generates setup code in dashboard → connector installer claims it via POST /api/connectors/claim
Connector registers cameras → starts RTSP capture with MOG2 motion + optional HOG person filter
Motion triggers 10–20s H.264 clip → SQLite queue → presigned S3 PUT → POST /api/clips/{id}/complete
Backend verifies S3 object → pushes job to Redis onevo:clip-jobs
Cloud AI downloads clip → YOLOE + ByteTrack → zone events → POST /api/ai-events
Risk Engine scores (0–39 none, 40–69 analytics, 70–89 medium, 90+ high) → creates alert → SSE + optional email
Staff reviews alert in dashboard → PUT /api/alerts/{id}/review

Important: YOLO runs only in cloud-ai, not on the edge connector. Edge uses motion detection only.

2. Module Dependency Diagram

flowchart LR
  Installer --> Connector
  Connector --> Backend
  Connector --> MinIO
  Backend --> Postgres
  Backend --> Redis
  Backend --> MinIO
  CloudAI --> Redis
  CloudAI --> MinIO
  CloudAI --> Backend
  Dashboard --> Backend
  Dashboard -.->|"broken: localhost only"| ConnectorAdmin
  ConnectorAdmin[Connector Admin :8099]
  Connector --> ConnectorAdmin

External dependencies: Azure ACR, Azure VM, Vercel (installer hosting), Gmail SMTP (optional), shop LAN cameras, ffmpeg (bundled in installer), NVIDIA GPU (recommended but MVP VM is CPU-only per infra/mvp/PROVISIONED.md).

3. End-to-End Workflow Validation

Step	Exists?	Status	Gap
Customer installs Installer	Yes	Partial	RTSP skip broken; stale docs; no health wait
Connector starts (WinSW service)	Yes	Complete	Auto-restart via WinSW
Setup code claim	Yes	Broken	wizard.py L286 — undefined `key` after claim
Camera connects (RTSP)	Yes	Complete	Reconnect unbounded; no backend offline signal
RTSP stream received	Yes	Complete	TCP transport, buffer=1
Frames → clip on motion	Yes	Complete	Not sent frame-by-frame to AI; clips only
Clip uploaded to S3	Yes	Complete	Requires port 9000 reachable from shop
Cloud AI processes clip	Yes	Complete	CPU fallback; MVP VM has no GPU
YOLO detects event	Yes	Complete	Synthetic test.mp4 produces no retail cues (documented)
Backend stores event	Yes	Complete	AiEvents + RiskEvents tables
Dashboard updates	Yes	Partial	SSE works; no reconnect; no store filter on SSE
Alert generated	Yes	Partial	Silent pilot mode hides from reviewers; cross-cam alerts broken clip URL
User acknowledges alert	Yes	Complete	Confirm/Dismiss/FalsePositive/NeedsFollowUp
History updated	Yes	Complete	AlertReviews persisted

Missing links: Dashboard zone snapshot → connector admin (hardcoded localhost); admin source management → backend (503); wizard duplicate camera creation; cross-camera theft → no SSE/email notification.

4. Broken Functionality List

Critical (P0 — blocks core workflows)

ID	Module	File	Description	Severity	Why it matters	Fix	Deps	Effort
B-01	Connector	connector/app/wizard.py:286	`wizard_claim` references undefined `key` after `claim_setup()` already persisted creds	Critical	`/setup` retry crashes with `NameError`; breaks recovery after failed activation	Remove duplicate cred writes or capture `key` from `claim_setup` return	provisioning.py	1h
B-02	Connector	connector/app/wizard.py:349-370	Duplicate `create_camera` loop after `provision_sources()` + `finalize_setup()`	Critical	Duplicate cameras or 502 on wizard source save	Delete second loop; rely on `provision_sources` only	provisioning.py	2h
B-03	Connector	connector/app/admin.py:854-868	`start_admin()` never passes `BackendClient` to `build_app()`	Critical	POST/DELETE `/sources` returns 503 — add/remove cameras broken in service mode	Instantiate client from store creds; pass to `build_app`	backend_client.py	3h
B-04	Installer	connector/installer/onevo-connector.iss:676-679	RTSP "Skip camera setup" blocked by validation on empty field	Critical	Installers cannot skip camera config as documented	Skip validation when user chose skip	Inno Setup	1h

High (P1 — broken features / bad UX)

ID	Module	File	Description	Severity	Why	Fix	Effort
B-05	Dashboard	dashboard/src/app/pages/setup/setup.component.ts:352	`connectorAdminHost = 'localhost'` hardcoded	High	Zone snapshot/live preview broken when dashboard is remote (production)	Resolve connector IP from backend connector record or configurable host	1d
B-06	Backend	backend/Services/TheftOrchestrator.cs:79	Cross-camera alerts use `ClipUrl = "multi-camera-event"`	High	Alert detail shows unplayable video	Link to exit clip or composite evidence	4h
B-07	Backend	backend/Controllers/CamerasController.cs:118-132	`TestStream` is ack-only stub	High	Setup "test stream" button does not validate RTSP	Proxy to connector or run ffprobe server-side	1d
B-08	Dashboard	Alerts/Clips pages	`storeId` query param ignored (only Setup reads it)	High	Admin deep-links from Get Started don't filter	Read `queryParamMap` in alerts/clips components	2h
B-09	Connector	connector/app/workers.py:63-70	Heartbeat overwrites setup `degraded_reason`	High	Admin UI loses activation error context	Merge reasons or don't overwrite setup errors	2h
B-10	Backend	backend/Controllers/AlertsController.cs:112+	SSE broadcasts all alerts to all authenticated users	High	Cross-tenant data leak in multi-store deployments	Filter by `storeId` claim; per-store channels	1d

Medium (P2)

ID	Module	File	Description	Fix	Effort
B-11	Connector	connector/app/main.py:186-192	`_preflight_rtsp()` defined but never called in native install	Wire into `_provision_native_installer`	2h
B-12	Connector	connector/app/orchestrator.py	Multi-camera shares single `RuntimeState`	Per-pipeline state or namespaced metrics	1d
B-13	Dashboard	dashboard/src/app/pages/alerts/alerts.component.ts	SSE no reconnect on error	Exponential backoff reconnect	4h
B-14	Installer	connector/installer/onevo-connector.iss:222-238	`WaitAndOpenStatus()` never called	Hook to post-install or remove dead code	2h
B-15	Connector	connector/app/workers.py:46-49	Failed uploads leave clip files on disk	Delete or quarantine after max retries	4h

5. Missing Implementation List

Critical missing features

ID	Module	Description	Why it matters	Effort
M-01	Infra	Production GPU VM (MVP is `Standard_D2s_v5` CPU)	YOLO inference too slow at scale	2d + Azure quota
M-02	Infra	TLS for API, dashboard, MinIO	Credentials and clips traverse HTTP	2d
M-03	Backend	EF `Database.Migrate()` instead of `EnsureCreated` + ad-hoc SQL	Schema drift, no migration history in prod	3d
M-04	Security	Connector admin authentication	LAN attacker can control capture	2d
M-05	Security	Production secrets management (no default JWT/bootstrap keys)	Trivial compromise	1d
M-06	Connector	Auto-update mechanism	Manual EXE redeploy for every fix	1w
M-07	Backend	Rate limiting on auth/claim/register	Brute force setup codes	1d

High priority missing features

ID	Area	Description	Effort
M-08	Dashboard	Analytics page — not implemented	1w
M-09	Dashboard	Reports page — not implemented	1w
M-10	Dashboard	Logs viewer — not implemented	3d
M-11	Dashboard	Detection History as dedicated view (clips partially cover this)	3d
M-12	Dashboard	AI Settings (model backend, confidence thresholds) — only risk tuning exists	3d
M-13	Auth	Password reset, change password, user profile	3d
M-14	Auth	Sign-up / self-service onboarding	1w
M-15	Backend	Camera delete API (connector removes locally only)	1d
M-16	Connector	Remote log shipping to backend	3d
M-17	Connector	Retry failed upload jobs from admin UI	1d
M-18	Cloud AI	Cross-camera theft SSE/email notification path	1d
M-19	Backend	Horizontal scaling (in-memory SSE AlertChannel)	1w

Medium / low priority

ID	Description	Effort
M-20	POS integration (Phase 1B — documented deferral)	Phase 1B
M-21	Staff cross-check patterns (Phase 1B)	Phase 1B
M-22	RF-DETR real tracking (pseudo track ID -1 today)	3d
M-23	`camera-module/` knowledge base folder — empty	N/A
M-24	Backup/restore strategy for Postgres + MinIO	2d
M-25	Monitoring/APM (App Insights, Prometheus)	1w
M-26	Multi-store connector fleet management dashboard	1w
M-27	False positive feedback loop → model tuning	Phase 2

Frontend pages requested vs actual

Requested page	Status	Location
Dashboard	Partial	Alerts + Get Started serve as hub; no unified KPI dashboard
Camera Management	Complete	setup.component.ts
Connector Management	Partial	health.component.ts + Get Started
Alerts	Complete	alerts.component.ts
Detection History	Partial	clips.component.ts
AI Settings	Partial	tuning.component.ts — risk weights only
Installer Management	Partial	Admin + Get Started download
User Management	Partial	admin-stores.component.ts — admin only
Analytics	Missing	—
System Health	Complete	health.component.ts
Reports	Missing	—
Settings	Partial	Scattered across Setup/Tuning/Admin
Logs	Missing	—

6. Security Issues

ID	Severity	Module	File / Area	Issue	Recommended fix
S-01	Critical	Backend	Program.cs:63, .env.example	Default JWT signing key in repo	Require strong env var; fail startup if default in Production
S-02	Critical	Backend	DbSeeder.cs	Default admin `admin@onevo.local` / `Admin123!`	Force password change on first login; disable seed in prod
S-03	Critical	All	Bootstrap key `dev-connector-bootstrap-key` shared by connector register + cloud-ai ingest	Separate service keys with scoped permissions
S-04	Critical	Connector	admin.py binds `0.0.0.0:8099` with no auth	LAN can pause capture, add sources, read ONVIF snapshots	Localhost-only or API token
S-05	High	Backend	ZonesController.cs	No tenant check on zone CRUD — any JWT with camera UUID	Add `TenantAccess.CanAccessStore` via camera join
S-06	High	Backend	AlertsController Stream	SSE leaks alerts across stores	Filter by tenant/store
S-07	High	Infra	PROVISIONED.md	MinIO port 9000 on public IP without TLS	Private endpoint or HTTPS reverse proxy
S-08	High	Database	`Cameras.RtspUrl`	RTSP credentials in plaintext	Encrypt at rest or use secret vault
S-09	Medium	Backend	JWT via `?access_token=` for SSE	Token in logs/referrer	Short-lived SSE tokens
S-10	Medium	Backend	No rate limiting	Brute force login/setup codes	Add middleware
S-11	Medium	Backend	Swagger enabled when `EnableSwagger=true`	API surface exposure	Disable in prod (prod compose does)
S-12	Medium	Dashboard	JWT in localStorage, no expiry handling	XSS → token theft	httpOnly cookies + refresh; 401 interceptor
S-13	Low	Connector	baked_config.py	Hardcoded pilot VM IP	Build-time only; document rotation

7. Performance Issues

ID	Area	Issue	Impact	Fix	Effort
P-01	Infra	MVP VM is CPU-only (`Standard_D2s_v5`)	YOLO inference bottleneck, queue backlog	GPU VM + docker-compose.gpu.yml	2d
P-02	Cloud AI	Sequential `brpop` single worker	Throughput limited to ~1 clip at a time	Horizontal cloud-ai replicas + job locking	3d
P-03	Cloud AI	Full clip frame processing	Latency scales with clip length	Frame sampling / keyframe-only option	3d
P-04	Connector	Per-frame MOG2 + optional HOG on edge	CPU on shop PC	Tune thresholds; disable HOG by default on weak hardware	1d
P-05	Connector	Unbounded RTSP reconnect	CPU/network churn on dead cameras	Cap retries; report offline to heartbeat	1d
P-06	Backend	`EnsureCreated` + raw SQL patches at startup	Slow/fragile boot on large DB	Proper migrations	3d
P-07	Backend	In-memory SSE	Doesn't scale; memory grows with subscribers	Redis pub/sub or SignalR backplane	1w
P-08	Network	Clips upload over HTTP to public MinIO:9000	Large bandwidth; no CDN	TLS + dedicated upload endpoint	2d
P-09	Database	Alerts list capped at 500, no pagination API tuning	Slow as data grows	Cursor pagination, indexes	2d

8. Technical Debt

Schema management: EF migration exists (20260718070711_AddReIDEmbedding.cs) but runtime uses DbSeeder.EnsureCreatedAsync + manual SQL patches
README inaccuracies: Claims ".NET 10" but targets net8.0; backend port 8080 in README vs 8081 in compose
Duplicate activation logic in connector/app/main.py (overlapping provisioning paths)
Dead code: _preflight_rtsp, WaitAndOpenStatus(), stale INSTALL.md v1.1.0 vs code v1.1.5
Demo data in production path: DbSeeder creates Demo Store with file://samples/test.mp4
Single shared service key for multiple trust boundaries
No structured logging in connector (print + 200-line deque)
Build deps not in repo: ffmpeg.exe, WinSW-x64.exe manual placement
Empty camera-module/ referenced in README as design knowledge base

9. UI/UX Improvements

Priority	Issue	Recommendation
High	Remote zone snapshot broken	Show connector-reachable URL or "open on shop PC" guidance
High	No global 401 redirect	Interceptor → login on token expiry
High	SSE "Offline" with no retry	Auto-reconnect + user notification
Medium	Reviewer role same nav as Manager	Role-specific landing pages
Medium	Tuning page save 403 for Reviewer	Hide save button for unauthorized roles
Medium	No loading/error states on some admin actions	Consistent skeleton + toast pattern (Alerts already good)
Medium	Dev credentials shown on login page	Remove in production builds
Low	Wildcard route → welcome not login	Consider `/app/*` unknown → 404 in app shell
Low	Mobile responsiveness	Zone canvas editor likely poor on tablet

10. AI Pipeline Improvements

Current state (solid): Pluggable backends (detector.py), ByteTrack, 11 Phase 1A event types (events.py), Re-ID embeddings (reid.py), dead-letter queue (main.py).

Improvement	Priority	Details
GPU production deployment	Critical	`CLOUD_AI_DEVICE=cuda` useless without GPU VM
Confidence threshold tuning UI	High	Env-only today (`CLOUD_AI_*`)
False positive handling	High	Review workflow exists; no feedback to model/thresholds
Batch processing	Medium	One clip at a time
RF-DETR tracking	Medium	Track ID always -1
Model hot-reload	Low	Requires worker restart
Evaluation harness	Low	cloud-ai/eval/ exists but not wired to CI
Phase 1B POS proxies	Future	Documented in README and events.py comments

11. Connector Improvements

Area	Current	Needed
Camera discovery	ONVIF in admin only	Add to installer wizard
Offline mode	Implicit via SQLite queue	Explicit UI indicator + queue depth alerts
Config sync	Poll backend cameras every 10s	Conflict resolution when dashboard edits RTSP
Log upload	None	Ship to backend or S3
Version management	Hardcoded 1.1.5	Check `latest.json` from Vercel
Memory/CPU monitoring	Disk % in heartbeat only	CPU/RAM metrics
Encryption	HTTPS to backend; local SQLite unencrypted	Encrypt creds at rest
Camera removal	Local only	Call backend delete

12. Installer Improvements

Issue	Fix
RTSP skip broken (B-04)	Fix Inno validation
No post-install health wait	Call `WaitAndOpenStatus` or document actual behavior
Stale INSTALL.md (v1.1.0)	Sync to 1.1.5
Manual ffmpeg/WinSW	Document clearly; consider submodule or download script
Uninstall wipes all ProgramData	Offer "keep configuration" option
No downgrade path	Document; consider migration backup
Baked backend URL rotation	Rebuild required — document update process

13. Backend Improvements

Implemented well: 40+ REST endpoints, BCrypt connector keys, atomic setup code claim, tenant scoping on stores/cameras/alerts/clips, presigned URLs, Risk Engine V4.

Needed:

Fix zones + SSE authorization (S-05, S-06)
Replace EnsureCreated with migrations (M-03)
Implement real TestStream (B-07)
Input validation on DTOs ([Required], polygon schema)
Separate cloud-ai service key from connector bootstrap key
Cross-camera alert notifications + valid clip URL (B-06, M-18)
Pagination on list endpoints
Health endpoints for cloud-ai worker (currently only pipeline queue depth)

14. Frontend Improvements

Implemented well: 12 routes, lazy loading, auth guards, SSE live alerts, zone canvas editor, onboarding checklist, installer download.

Needed:

Fix localhost connector admin integration (B-05)
Query param deep linking (B-08)
SSE reconnect (B-13)
Missing pages: Analytics, Reports, Logs (M-08–M-10)
Auth: password reset, token expiry (M-13)
ng serve API proxy config for local dev without Docker
Production build: strip dev login hints

15. Production Readiness Score Breakdown

Category	Score	Notes
Core pipeline E2E	85/100	Works with real footage; synthetic test validates plumbing only
Connector reliability	55/100	Critical wizard/admin bugs; no auto-update
AI accuracy/ops	70/100	Good architecture; CPU MVP limits throughput
Backend API	75/100	Feature-complete; auth gaps on zones/SSE
Frontend UX	65/100	Main flows work; missing pages; remote snapshot broken
Security	35/100	Default secrets, unauthenticated admin, HTTP MinIO
Deployment/ops	50/100	CI/CD exists; no GPU, no TLS, ad-hoc schema, no monitoring
Documentation	60/100	Good README; stale installer docs, port mismatches
Testing	45/100	Connector unit tests; no E2E integration tests
Scalability	40/100	Single VM, in-memory SSE, single cloud-ai worker

Overall: 58 / 100

Verdict: Suitable for controlled pilot with manual ops support. Not ready for unattended multi-store customer rollout until P0 bugs and security items are resolved.

16. Prioritized Roadmap

Phase 0 — Release blockers (1–2 weeks, high impact)

Task	Issues	Effort	Impact
Fix wizard claim `key` bug	B-01	1h	Unblocks setup recovery
Fix duplicate camera creation	B-02	2h	Prevents bad DB state
Pass BackendClient to admin	B-03	3h	Unblocks camera management
Fix installer RTSP skip	B-04	1h	Unblocks headless install
Rotate all default secrets	S-01–S-03	1d	Prevents trivial compromise
Secure connector admin (localhost or token)	S-04	2d	Prevents LAN takeover
Fix zones + SSE tenant filtering	S-05, S-06, B-10	2d	Multi-store safety
Fix remote zone snapshot	B-05	1d	Production zone setup

Phase 1 — Production hardening (2–4 weeks)

Task	Issues	Effort	Impact
TLS + domain setup (API, app, MinIO)	S-07, M-02	2d	Security + compliance
GPU VM provisioning	P-01, M-01	2d	AI throughput
EF migrations replace EnsureCreated	M-03, P-06	3d	Schema reliability
Rate limiting + input validation	S-10	2d	Abuse prevention
SSE reconnect + 401 handling	B-13, S-12	1d	UX + security
Cross-camera alert fix + notifications	B-06, M-18	1d	Feature completeness
Failed upload cleanup + heartbeat fix	B-09, B-15	1d	Edge reliability

Phase 2 — Customer feature completeness (4–8 weeks)

Task	Effort	Impact
Analytics + Reports pages	2w	Customer visibility
Logs viewer + remote log shipping	1w	Supportability
Password reset + user profile	3d	Enterprise auth
Connector auto-update	1w	Ops at scale
AI settings UI (thresholds, backend)	3d	Tunability without redeploy
Monitoring (App Insights / Prometheus)	1w	Observability
E2E integration test suite	1w	Regression safety

Phase 3 — Scale & Phase 1B (future)

Horizontal cloud-ai workers
Redis-backed SSE / SignalR
POS integration
Model feedback loop from false positive reviews
Multi-region deployment

17. Integration Audit Summary

Integration	Auth	Retries	Error handling	Health	Offline recovery	Grade
Installer → Connector	Config file	WinSW auto-restart	Partial	`/health` exists	WinSW restart	B
Connector → Cameras	RTSP creds in URL	Unbounded reconnect	Good logging	Frame read failures	Reconnect loop	B-
Connector → Backend	API key header	Upload exp backoff	SQLite queue	Heartbeat 10s	Queue persists	A-
Connector → AI	Indirect via S3+Redis	N/A	Upload retry	N/A	Clips buffered	A-
Backend → Database	Connection string	EF retry limited	EnsureCreated patches	`/api/health`	Postgres persistence	C+
Backend → Frontend	JWT Bearer	HTTP client default	Partial	Health page	Manual re-login	B
Frontend → Connector admin	None	N/A	Broken remote	N/A	N/A	F
Cloud AI → Backend	Service key	3x job retry + DLQ	Good	Pipeline health only	DLQ inspection	B+

18. What Works Well (do not regress)

End-to-end clip pipeline architecture is sound and documented
Durable SQLite upload queue with exponential backoff
Atomic setup code pairing (BCrypt hashed, 30-min expiry, single-use)
Risk Engine V4 with evidence-only language guardrails
Pluggable YOLO backends with evaluation harness scaffold
Angular dashboard covers core staff workflows (alerts, review, zones, tuning)
GitHub Actions CI + Azure deploy pipeline with Windows installer artifact
Connector ONVIF discovery and multi-camera orchestrator
Presigned S3 uploads (clips never pass through backend body)

Recommended immediate actions before any customer release

Fix all four P0 connector/installer bugs (B-01 through B-04) — estimated 1 day
Rotate secrets and disable default admin in production — 1 day
Lock connector admin to localhost or add auth — 2 days
Fix SSE/zones tenant leaks — 2 days
Enable TLS and move MinIO off public HTTP — 2 days
Provision GPU VM or set expectations that AI processing will lag — 2 days

Minimum viable pilot timeline: ~2 weeks after P0 + security fixes, with ops runbook (docs/PILOT_ONBOARDING.md) and manual support.

FilesExpand file tree

onevo_system_audit_ebd9d4dc.plan.md

Latest commit

History