All four questions answered: Q1-3 RISKY (credible paths), Q4 PROVEN. WebView-
BotGuard POT (YTDLnis-proven) collapses the JS-runtime problem; HLahwani/yt-
dlp-android solves Q1+Q2+Q3 together. One open question remains: does WebView-
POT recover anti-bot WITHOUT curl_cffi (broken on Android)? The Stage-1
prototype tests exactly that. Fallback: client-to-server app against existing web
API. Update ROADMAP 3.2 with pointer to findings.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Real build-proof run locally: vtt.rs cross-compiles to aarch64 + x86_64 Android
.so (NDK r26d, exported C-ABI symbol). cargo-ndk + uniffi recommended; ~5 days
to reuse 5-6 pure modules. On-device run-proof blocked by sandbox emulator
reaping, downgraded to symbol-export check per plan.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The pivotal gate does NOT block: POT is required on mobile, but WebView-based
BotGuard token generation (proven in YTDLnis) is the durable path and collapses
the JS-runtime + POT problems into one, killing the Deno/Node sidecar.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
External JS runtime now a hard yt-dlp dependency (Python jsinterp deprecated
for YouTube 2025.11.12); QuickJS-ng via NDK or HLahwani/yt-dlp-android is the
path. Also fixes a markdown line starting with '#'.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Six tasks (Q1–Q4 research + emulator repros, plus scaffold and go/no-go
synthesis) against the approved spec. Research-spike shape: investigate →
reproduce-where-cheap → record evidence → verdict, not TDD.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
First sub-project of the Android client: a research spike to decide go/no-go
on a standalone on-device download engine (yt-dlp + JS runtime + POT) before
building any UI. Four research questions, "reproduce where cheap" on the
emulator.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
