The headline: this build can now measure your device and your link, tell you the right settings, explain every recommendation, and apply the lot with one press — revertibly.
New: Device Settings Assessment
Settings → Playback → Advanced → Device assessment
One flow that works out what this device, this display chain, and your link can actually do, then turns it into per-setting recommendations. Every row shows why — the measurement or device fact behind it — and your current value whenever it differs. Recommendations are grouped by how they were derived, because not all confidence is equal:
- Measured — tested on your stream. Runs the throughput sweep against your real last-played file and recommends the cheapest parallel-connection configuration that sustains it. Not a guess; a measurement.
- Calculated for this device. Derived from queryable facts: your RAM tier's memory limits, the display's real mode list, the Dolby Vision decode path your hardware actually resolves to, your audio route. Target buffer, max buffer, VOD cache sizing, AFR, the DV toggles and more — each with its arithmetic shown.
- Choose your priority — buffer profiles. Fast start / Balanced / Stall-resistant. This is a genuine trade-off with no right answer, so you pick — and if the sweep saw your link wobble, it marks a suggestion (a link-stability score is shown; steadier links suggest Balanced, variable links suggest Stall-resistant).
- Detected — verify yourself. Things no API can honestly decide (tunnelled playback quirks, unreadable HDR capabilities). The assessment says what it found and leaves them alone — these rows are never written by Apply.
Apply & Revert. After choosing a profile, Apply recommended settings (press twice — first press shows how many settings will change) writes the Measured and Calculated rows plus your profile. A full snapshot of your previous values is kept: Revert to pre-assessment settings restores everything, survives app restarts, and always undoes the most recent apply. Settings are per-profile.
Tip for best results: play your most demanding typical title before running — settings tuned for the heaviest case play lighter content fine, not the reverse. No last-played stream? The device rows still run; only the measured rows wait.
Improved: the stream speed test
The Advanced speed test and the assessment now share one engine, and it got smarter:
- Buffer-aware recommendations. Connection overhead and the playback buffer share one memory budget. On memory-constrained devices, the sweep will now prefer a cheaper configuration that still comfortably exceeds the title's bitrate when it buys meaningfully more buffer — and the verdict spells out the trade (e.g. "leaves 200 MB ≈ 16 s of buffer instead of 100 MB ≈ 8 s").
- Link-stability measurement. Throughput is sampled in sub-windows during every pass; the variability score feeds the profile suggestion above.
- No more knife-edge verdicts. A pass within ~5% of the sufficiency target counts — a 1% shortfall on a single sample was never a real fail, and previously could promote a configuration costing 64 MB more for identical throughput.
Changed: Filters & Sorting defaults reset (one-time) — please read
Debrid Filters & Sorting is re-baselined to the current shipped defaults once on first launch of this build, replacing stored preferences — including your customisations. This was deliberate: field configurations had drifted in ways that couldn't be reconstructed remotely, and the previous defaults logic had a bug where stored-equals-default state silently fell through to a legacy path with the wrong sort order. After the one-time reset, your edits stick and are version-stamped. If you had custom release-group tiers, sort chains or filters, you'll need to redo them. Max results now defaults to 8, and Best quality is the default sort everywhere.
Two quick actions at the end of Filters & Sorting make recovery painless: Reset to recommended defaults restores the shipped baseline in one tap, and Show everything clears all filters and the result cap while keeping the quality ordering — nothing hidden, still sorted best-first.
Also in this build
- Quality rules shortcut — Playback → Auto Stream Selection now links straight to Connected Services → Filters & Sorting, so auto-play and the quality rules it consumes are one press apart.
- Rate-limit recovery (unreleased nt3 fix, first shipped here): when a debrid CDN rate-limits the parallel downloader (HTTP 429), throughput previously stayed clamped for the rest of playback; it now restores itself after a quiet cooldown. If you've seen speeds collapse mid-title and never come back, this is that fix.
- Assessment result cards redesigned for TV: focusable, individually scrollable, squared-off corners — nothing clips or truncates.
Honest notes
- The assessment's profile values and its stability/sufficiency thresholds are first-generation numbers, marked as derived in the code, and will be tuned on field data — this build already includes two such tunings from real runs.
- All assessed settings apply to the ExoPlayer path. On the MVP player the assessment says so instead of pretending.
- The sweep measures raw download throughput; real playback also depends on decode and server load. The verdict is a strong recommendation, not a provable optimum.