# Public release compatibility review

Review date: **2026-10-07**. Release: **0.3.0-beta**. Scope: source and documentation review, official platform requirements, and existing synthetic test evidence. No private profile, account key, account form, real capture source, or external publishing destination was used in this review.

**Disposition:** present this as a Windows experimental/beta encoder frontend with explicit prerequisites. The evidence does not support universal platform compatibility, a guaranteed replacement for TikTok LIVE Studio, or “the most lightweight streaming app.” A public download may accurately explain its narrow design and measured test scenarios; it must keep the limitations visible before download.

## Claim ladder

| Claim | Evidence status | Appropriate public wording |
| --- | --- | --- |
| Small native frontend with no embedded browser/chat panel | Supported by source design | Native Windows controls; optional preview; deliberately limited feature set. |
| Reuses encoding for matching outputs | Supported by source and prior native encoder-pointer/loopback tests | Matching widescreen destinations share one video encoder; portrait needs a second; AAC is shared. |
| RTMP transport and independently failing outputs work locally | Prior synthetic loopback evidence | Local RTMP receiver accepted synthetic streams; other routes continued during an intentional portrait-route disconnect/reconnect. |
| H.264/AAC files are decodable | Prior decode evidence, with scope noted below | Synthetic encoded media decoded in local checks. |
| Connects to YouTube, Twitch, Kick and TikTok | Design compatibility only; no authenticated live ingest verification | Supports user-supplied RTMP/RTMPS destinations subject to account eligibility and actual platform acceptance. |
| Uses less RAM/CPU than OBS or LIVE Studio | Requires matched, repeated measurements with version/workload/method | Publish only the measured comparisons and their exact limits; not a general promise. |
| Most lightweight, no lag, works on every PC | Unsupported | Do not use these claims. |

## Release-critical boundaries

1. **Installed OBS dependency and ABI.** The reviewed engine checks for 64-bit OBS **30.x**, tested with **30.1.2**, and rejects other major versions. It imports `avutil-58.dll` and specific native layouts/functions. Installing a newer OBS version does not automatically make this build compatible. The app package alone is not a standalone media runtime. Do not request that users downgrade their working installation; a future isolated supported runtime/version strategy needs its own validation.
2. **TikTok account eligibility.** Direct server/key access is required. No transferable keyless integration exists in the source. Do not promise that a TikTok login, LIVE Studio approval, follower threshold, or third-party application approval unlocks this app. [Platform guide](PLATFORMS.md).
3. **Platform-specific settings.** Generic engine limits are broader than individual services. Kick's official maximums are 8 Mbps, 1920x1080 and 60 FPS. Public preflight must apply the selected-platform constraints before sending. YouTube's published recommended H.264 bitrate and stereo audio settings differ from the shared defaults. That is a compatibility tradeoff, not tested service rejection.
4. **TLS is not covered by cleartext loopback.** RTMPS transport is delegated to installed OBS modules; no live-platform TLS/SNI/authentication test was performed. Prefer secure official endpoints. Twitch publishes a secure default ingest template. [Official ingest list](https://ingest.twitch.tv/ingests).
5. **Capture and hardware coverage.** The implementation supports explicit window/display selection and optional cameras. Synthetic tests do not establish practical game capture, fullscreen transitions, camera driver compatibility, microphone quality, audiovisual synchronization, GPU contention, or long-session behavior. NVENC is preferred; x264 is a CPU option with a different resource cost. Only tested encoder paths should be described as validated.
6. **Distribution hygiene.** Use fresh allowlisted staging and ship matching source, build scripts, assets, license and notices with each executable. Exclude profiles, credential-helper state, account/session notes, personal captures and personal benchmark machine identifiers. Audit the final archive, not just the source list.

## Endpoint and credential audit

The UI checks absolute `rtmp`/`rtmps` URLs, a nonempty host/key, single-line input and duplicate server/key pairs. The engine independently checks scheme, host, key, count and readiness before encoder/output creation. Those checks establish syntax and execution guards, not domain ownership, TLS authenticity, account access or broadcast readiness. Custom endpoints are intentional; users must copy trusted values from the intended service.

The service is OBS `rtmp_custom` with `use_auth=false`; only server and key are supplied. Separate username/password authentication, OAuth login, token refresh and platform-side broadcast creation are not implemented. Consequently, a service requiring one of those flows is not supported just because its name is a streaming platform. Server URLs containing username/password user-info or fragments are unsupported; token-bearing query URLs must remain encrypted with the rest of the profile.

The whole saved profile, including ingest URLs, uses Windows DPAPI CurrentUser encryption with atomic replacement; public error messages are sanitized. Native OBS logs are dropped and FFmpeg logging is disabled to avoid inadvertent key disclosure. UI exceptions redact configured server/key strings. This reduces accidental logging but is not a security certification: managed strings, native service copies, clipboard contents, same-user processes and crash dumps can expose active credentials. Neither DPAPI nor masked fields encrypt a plain RTMP connection.

Cameras are disabled in new and migrated profiles. A saved scene may use a camera only after the user explicitly selects and enables it; applying that scene starts its selected inputs. Enumerating source/device metadata does not start capture. Display capture is an explicit choice and includes everything visible on the chosen display. Desktop audio captures the selected playback device and can include other applications. The layout editor renders placement guides without opening devices.

Scenes keep independent landscape and portrait placements and background image paths. Image files must remain readable at their saved local paths. Scene application stages both compositions before committing a successful change. Pause sends a break slate while muting effective broadcast audio; it keeps connections active and retains logical mixer choices for resume. The release validation document identifies the exact tests that exercised these behaviors.

Global hotkeys are disabled by default. When users enable and save them, normal launches register their bindings and report conflicts. Diagnostics never register keys. Saved profiles from the earlier dedicated-game prototype preserve credentials and mixer choices but do not infer window, display, camera, or hotkey consent.

## Test evidence and its limits

- `engine-research/dual-loopback-probe-result.txt` (2026-10-04): synthetic input, silent mixer, fake keys and IPv4 loopback. It checked two distinct H.264 encoders, three matching outputs sharing one encoder, shared AAC, continued landscape frames while the portrait route disconnected/reconnected, and a local recording surviving network-output stop.
- `engine-research/dual-loopback-sink-result.json`: three landscape connection records and three portrait connection/reconnection records. Platform labels are fixture names, not real platform endpoints.
- `engine-research/dual-loopback-decoder-result.txt`: the recorded commands decoded **one video frame per connection**. This establishes media headers and initial frame decode, not a complete-stream soak test. Do not describe that artifact alone as a full decode.
- `tests/launch-guard-result.txt` and `tests/profile-probe-result.txt`: previous controlled fixtures cover startup/profile handling, not real account activation. Re-run relevant guards if public platform-selection behavior changes.
- New benchmark results, if added elsewhere, should state exact OBS/app versions, GPU/encoder settings, source complexity, frame rate, duration, preview state, process accounting and how many video encoders/outputs each side used. A single-frame static source, idle process, or different feature workload is not a representative gameplay comparison.

### Independent full-file decoding, 2026-10-07

All six existing synthetic loopback FLV fixtures were decoded completely, including both video and audio, using FFmpeg `n6.0-12-ga6dc92968a` with software decoding capped at one thread. Every decode and frame-count probe exited 0 with no error messages. This was scheduled before the separate resource benchmark, so it did not run during that measurement. It used no native capture, microphone, preview, encoding or network connection.

| Fixture | Video | Decoded video frames | Decoded AAC frames |
| --- | --- | ---: | ---: |
| `dual-loopback-1.flv` | H.264 High, 1280x720, 30 FPS | 89 | 136 |
| `dual-loopback-2.flv` | H.264 High, 1280x720, 30 FPS | 89 | 136 |
| `dual-loopback-3.flv` | H.264 High, 1280x720, 30 FPS | 89 | 136 |
| `dual-loopback-4.flv` | H.264 High, 720x1280, 30 FPS | 77 | 118 |
| `dual-loopback-5.flv` | H.264 High, 720x1280, 30 FPS | 77 | 118 |
| `dual-loopback-6.flv` | H.264 High, 720x1280, 30 FPS | 61 | 93 |

All audio was AAC-LC, 48 kHz, stereo. The first three preserved files were byte-identical, consistent with the shared widescreen encoder. These are short preserved samples, approximately 2–3 seconds each; decoding them fully does not turn the earlier loopback test into a long-duration stability test.

Commands, run against each existing fixture under `engine-research/`:

```text
ffmpeg -hide_banner -loglevel error -nostdin -threads 1 -i <fixture.flv> -map 0:v:0 -map 0:a:0 -threads 1 -f null NUL
ffprobe -v error -count_frames -show_entries stream=codec_name,profile,width,height,r_frame_rate,sample_rate,channels,nb_read_frames -of json <fixture.flv>
```

SHA-256 identification of tested samples:

```text
161d8b7e20e8bb54e72e8b6c0c197aa72504544a8e4ec72834c4be6b0fe3b469  dual-loopback-1.flv / -2.flv / -3.flv
216e0396164f292460686fa64afbb4576b565bcef00a48f5cb19346a71330707  dual-loopback-4.flv
47e77a5f7422e1589d6637dce5fe701a5c0026dc3bde3736ad5626e4a4cf9857  dual-loopback-5.flv
277388920a48f38b9d259e1fd50243d76e4c45820b3a62624932dda1941102c8  dual-loopback-6.flv
```

## License and public distribution

The repository declares GPL-2.0-or-later and includes the GPLv2 text. It links to libobs. Preserve that license grant, supply the complete corresponding app source and build assets for each binary, and retain upstream notices in any copied upstream code. OBS's official guidance requires GPL-compliant distribution for works linking against its code. Separately installed OBS binaries are dependencies, not bundled in the reviewed archive; adding them later needs a separate third-party/source review. [OBS licensing overview](https://obsproject.com/help), [OBS source license](https://github.com/obsproject/obs-studio/blob/master/COPYING), [OBS project licensing guidance](https://obsproject.com/forum/threads/forum-resource-and-ip-policy.178569/).

The executable and source archive must correspond to the same release. File hashes verify an exact artifact after download; they do not establish publisher identity or replace code signing. Avoid official-looking platform endorsements, compatibility certification badges or reused platform logos implying affiliation. State that LeanCast is independent of OBS and the streaming services.

This review is a bounded engineering/distribution audit, not independent security certification or a completed live-platform acceptance test.
