Deterministic IPTV corpus and performance benchmarks

[MrFr3di]

Objective

Maintain the repository-owned deterministic IPTV measurement authority used to justify parser, database, UI and playback optimizations. Correctness/compatibility fixtures are owned separately by #186; this issue owns scale/performance evidence and provenance.

Current execution checkpoint

• deterministic M3U profiles 1k / 10k / 50k exist;
• measurement series/harness exists;
• D0/test(android): enforce exact two-AVD repository contract #181 established exactly two repository-owned Android TV AVD identities;
• fix(measurements): align search boundary evidence with published results #178 currently owns the known catalog-search measurement-boundary correctness defect and must be accepted before new DB performance conclusions are trusted.

Canonical virtual-device contract

Repository-owned Android correctness/measurement execution uses only:

• MuxTV_TV_OLD_API26;
• MuxTV_TV_CURRENT_API36.

There is no third low-RAM, benchmark, catalog, player, variance, 720p or representative-mainstream AVD. Low-memory/display/stress conditions are configurations/evidence dimensions of the two canonical devices where the platform allows it.

Emulator measurements validate repeatability/correctness and relative changes on the named environment. They do not establish weak-ARM/vendor MediaCodec/HDR/passthrough/thermal/absolute-TV-performance claims; those require physical-device evidence through #31.

Measurement families

Ingest / source refresh

• M3U parse wall time;
• allocated bytes / GC where measurable;
• bytes read;
• staging/activation transaction duration;
• DB/WAL delta;
• HTTP 200 full-publication vs Source refresh: conditional HTTP validators and 304 Not Modified #100 304 path when implemented.

Database

Only after #178:

• Channels/Search/Guide query p50/p95;
• EXPLAIN QUERY PLAN;
• transaction duration;
• DB/WAL size;
• connection contention / SQLITE_BUSY where observable;
• Database measurement: compare Room3 pool/FTS5/WITHOUT ROWID candidates only after #178 #196 candidate experiments remain measurement-only until separately accepted.

UI

• startup/frame p50/p95/p99 and jank for accepted CUJs;
• D-pad action-to-visible-result latency where causal measurement is feasible;
• Compose stability/recomposition evidence only when used for an actual decision.

Playback

• request/prepare -> first rendered frame;
• seek acceptance/apply/render latency;
• rebuffer count/duration;
• channel-zap time;
• observed safe transport/format/device metadata;
• bounded memory growth;
• Playback: measured adaptive live buffer policy for weak TV and high-bitrate streams #109 buffer-policy decisions only after these semantics are trustworthy.

Evidence infrastructure relationship

• Observability: adopt AndroidX Tracing 2.0.1 behind a secret-safe MuxTV trace boundary #192 AndroidX Tracing provides typed trace spans, not product truth by itself;
• Observability: add secret-safe OkHttp phase timings for source refresh and playback #193 OkHttp phase timing separates DNS/connect/TLS/TTFB/body from parser/DB cost;
• Media3 Analytics supplies player/decoder/rebuffer evidence;
• Release hardening, baseline profile and physical TV alpha gate #31 owns Baseline Profile, Macrobenchmark release CUJs and physical-device qualification;
• Catalog: add sanitized IPTV compatibility corpus and executable support contract #186 compatibility corpus is correctness-shaped and must not become a performance threshold source.

Methodology

• compare like-for-like environment/corpus/commit provenance;
• report distributions, not one-off timings;
• initial measurements are descriptive until repeated variance evidence supports a threshold;
• timed 50k work remains manual stress evidence rather than a mandatory PR/release gate unless separately promoted with data;
• do not optimize a component solely from static inspection when a measurement can falsify the hypothesis.

Acceptance

• deterministic corpus/profile seed remains reproducible;
• every report records exact source commit + environment metadata;
• same comparison uses the same corpus/profile and equivalent device/runtime configuration where practical;
• no structural/runtime optimization is accepted from a stale or known-invalid measurement boundary;
• fix(measurements): align search boundary evidence with published results #178 is accepted before DB-query conclusions are used;
• no performance claim generalizes emulator evidence to physical weak TVs;
• no additional AVD identity is created.

Non-goals

• IPTV dialect correctness taxonomy (Catalog: add sanitized IPTV compatibility corpus and executable support contract #186);
• universal codec/device certification (Release hardening, baseline profile and physical TV alpha gate #31);
• automatic FTS5/pool/cache/Rust/libmpv adoption;
• treating newest technology as an optimization without measured benefit.

[MrFr3di]

Progress after PR #44

Package A is complete and merged as 3e24cccb188b53652285929a11e3b50697aad5f7.

Delivered:

• deterministic provider-neutral M3U profiles: 1k / 10k / 50k;
• explicit seed and exact source commit;
• streaming generation into caller-owned OutputStream;
• stable SHA-256 and UTF-8 byte count;
• controlled duplicates, long metadata, mixed line endings, relative locators and optional header directives;
• parser-recognized malformed/skipped fixture and manifest expectations;
• real StreamingM3uParser agreement tests;
• permanent :core:testing:test inclusion in Fast/Full validation.

Review caught and fixed a false-positive validation gap: core:testing:test was originally absent from verify-local.ps1. The real gate then exposed an incorrect malformed fixture, which was fixed without changing the production parser.

Remaining issue #27 packages:

1. canonical serialized manifest artifact and repository-owned corpus generation entry point;
2. deterministic HLS/XMLTV starter fixtures;
3. descriptive parse/stage/activate/query/Player measurements with environment metadata;
4. repeated variance evidence before introducing a dedicated benchmark threshold gate.

Issue #27 remains open until those packages are completed.