Observed
Service state, nginx routing, systemd configuration, source excerpts, database backend, effective runtime flags.
A technical view of the deployed system and the work still ahead.
This presentation focuses on the Real Video Creator architecture observed in a read-only September 24, 2026 VPS/source audit. Real Gamers is a separate native media effort, disclosed as a technical capability and possible future upsell: not a shipped RVC integration.
Audit provenance: founder-supplied RVC eight-phase read-only VPS/source audit and separate Real Gamers static/source audit. Neither is a third-party certification.
Do not mistake implementation for production-scale proof.
Service state, nginx routing, systemd configuration, source excerpts, database backend, effective runtime flags.
Project row locking, realtime authorization, private media delivery, credit reservation and payment webhook verification.
Independent pentest, sustained throughput/load profile, universal pass of detected tests, third-party attribution of creator revenue.
Account and export screenshots are dated samples; external activation, repeated use and paid conversion still need instrumentation.
Scope boundaries are inherited from the September audit itself, not inferred from newer live checks.
Explicit separation of interactive requests, realtime collaboration and media jobs.
nginx/TLS → Gunicorn/Django HTTP application; Daphne/Channels for Studio WebSockets.
PostgreSQL for app and transaction state; Redis for Channels and background task transport.
Celery Studio export queue, FFmpeg/FFprobe render/validation, private media delivery.
video generation worker, separate CPU preprocessing and GPU voice inference/generation queues; recovery scheduler.
PayPal live billing integration, subscriptions, credit reservations and ledger records.
Systemd-managed services and retention timers; shared host with other projects.
Observed on a shared Ubuntu/Plesk VPS in September 2026. Production and “staging” hostname mapped to the same Gunicorn runtime.
Features are meaningful when the state transitions are correct.
Authenticated workspace/project lookup and project revision checks.
Row-lock and expected-revision strategy for HTTP/WebSocket saves; stale revision returns a conflict.
Size limits, media type/signature/probe validation, SHA-256 and private delivery.
FFmpeg argument arrays without observed shell invocation; stream/container validation post-export.
Source inspection and targeted media-access HTTP checks; not a full cross-browser performance benchmark.
Access is rechecked when saving, not only at socket connection.
Daphne/Channels connection requires an authenticated user.
Project is scoped through the workspace-visible project query.
Transaction and database row lock guard state updates.
Expected revision, operation count, payload size and cursor bounds are validated.
Messages are distributed through Redis Channels layer.
Audit cleared a suspected connection-only authorization issue in inspected save paths. No external hostile WebSocket matrix was run.
CPU preprocessing, GPU generation and scheduler have distinct roles.
Enrollment/reference preprocessing with consent/version checks.
GPU-backed Chatterbox voice processing and recorded outputs.
Stale-job detection, requeue logic and bounded attempts.
Private output/project asset handling; failure states kept in the job record.
Models/tests and workers were observed in September audit; independent throughput, quality benchmarking and universal job success were not established.
Production services were observed on a shared multi-application VPS.
Gunicorn :8010 and Daphne :8011 bound to loopback behind nginx.
video generation worker, Studio export worker, voice CPU, voice GPU and recovery scheduler.
Support maintenance and backup/retention timers.
Worker units had stronger systemd sandboxing than Gunicorn/Daphne; resource limits were not configured.
Services and ports are historical observations from September audit, not an assertion of present configuration.
Observed controls are encouraging; an audit is not a certificate.
Actual existing private-media file requests returned HTTP 404 on both production and staging hostnames.
PayPal signature-verification path, idempotent webhook handling, transactional reservation and refund logic.
File signature checks, probed media formats, private project asset streams and bounded message/state sizes.
Project user scope, realtime write re-authorization and owner-controlled paid-generation approvals.
No confirmed critical application-layer vulnerability was identified in inspected paths; complete authentication/pentest coverage was NOT performed.
The source contains substantial automated testing. A full passing run was not captured in these audits.
Payments, promotions, concurrency, project revision, uploads, voice state recovery and support rate limiting.
GPU mailbox, AV timing, H.264 queue behavior, recording recovery, replay and RTMP transport.
Neither September audit captured execution of every detected test. An executed, timestamped test report is the next diligence artifact.
774 and 357 are independent project inventories; not one combined product’s passing suite. Results have not been re-executed in this package.
The current product records do not yet show independent adoption at scale.
Independent registration, first usable export, recurring weekly creators and cohort retention.
Completed real checkout, paid customer interviews, net revenue and per-export compute margin.
Snapshots do not establish successful public downloads, activated creators, or revenue. Baseline from Oct. 21 materials.
Native capture and streaming work is a future option, not an RVC feature today.
Rust/Tauri, D3D11/DXGI capture, frame mailbox and compositor.
WASAPI, timing/mixing, Media Foundation H.264/AAC.
Custom RTMP/RTMPS transport, MKV recording/recovery, replay and remux.
Replay memory pressure, unbounded queues, WebView CSP and local-service hardening were flagged.
September Real Gamers static/source audit; runtime fuzzing, GPU device-loss and stress tests not performed. Do not imply general availability or RVC integration.
Specific engineering work identified in the September review.
Shared writable voice-runtime files under the web-tier identity weakened separation.
High-load media services lacked effective systemd memory/CPU quotas.
Public Gunicorn/Daphne had weaker sandbox settings than voice/export workers.
Production and staging hostnames shared the same app process and live payment runtime.
RVC-BOUNDARY-001 · RVC-RESOURCE-001 · RVC-SVC-001 · RVC-DEPLOY-001 from Sep. 24 audit. Remediation status requires a fresh re-audit.
Current controls should be measured, retested and strengthened as use grows.
Independent user classification, first-export funnel, stable tracking and dated executed test reports.
Reduce time-to-export, introduce external reviewer, prioritized sandbox/resource hardening.
Retained creators, compute cost per export, recovery and workload profiling.
Subscription conversion, churn, acquisition unit economics and operational continuity runbooks.
Proposed gates align with the 12-month investor budget; outcomes require actual users and dated financial records.
Architecture, source and dated records can be shared for diligence.
We can review architecture, dated audit reports, redacted test output, a guided product demo and measurement plans. Outside technical review, production load metrics and external customer cohorts remain part of the financing work.
Engineering record based on founder-supplied September 2026 read-only and static audits; October 21, 2026 customer snapshot. No new production audit conducted by this update.