research(abr): Apple's published H.264/HEVC ladder bitrates & resolutions vs our 1.5x-gap assumption

[jonathaneoliver]

Motivation

Our synthetic ABR ladder (go-proxy/pkg/ladder/ladder.go) is built on near-uniform geometric gaps:

• DefaultMaxStep = 1.15 (max rung-to-rung ratio ~1.15×)
• DefaultBumpPct = 5.0 (5% peak-anchor bump)
• DefaultTopHeadroomPct = 50.0

A lot of our characterization conclusions about AVPlayer behavior — startup over-selection, over-downshift (drop ≥2 rungs then correct), and ladder-density hunting — are measured against this gap structure. See findings:

• .claude/findings/47d16786-over-downshift-2026-06-12.md
• .claude/findings/ladder-density-over-downshift-2026-06-12.md
• .claude/findings/avplayer-startup-variant-selection-2026-06-07.md

If real-world / Apple-recommended ladders use non-uniform, wider gaps (the suspicion is they do NOT conform to a clean ~1.5× geometric step), then our characterization ladder may not represent how AVPlayer's estimator actually behaves in production — which would change how we read over-selection and over-downshift severity.

Hypothesis to check

Apple's published HLS reference ladders (H.264/AVC SDR and HEVC) do not follow uniform ~1.5× bitrate gaps between adjacent rungs.

Tasks

• Pull Apple's published reference tiers from the HLS Authoring Specification for Apple devices — both the H.264/AVC SDR table and the HEVC table (bitrate + resolution + frame rate per rung).
  • https://developer.apple.com/documentation/http-live-streaming/hls-authoring-specification-for-apple-devices
• Tabulate the actual rung-to-rung bitrate ratio for each adjacent pair, for both codecs. Confirm/refute the "~1.5× uniform gap" assumption.
• Compare Apple's gap structure against our synthetic ladder (1.15× max step, 5% bump) — note where they diverge (gap magnitude, uniformity, resolution/bitrate pairing, HEVC vs H.264 offset).
• Decide whether the characterization rig should offer an "Apple-reference" ladder profile alongside the synthetic one, so over-selection/over-downshift tests can be run against a production-shaped ladder.

Deliverable

A short comparison table (Apple H.264, Apple HEVC, our synthetic) of bitrate/resolution per rung + adjacent ratios, plus a note in .claude/standards/ (abr-decision-model.md) on whether our gap assumption holds. If they diverge materially, a follow-up issue to add the Apple-reference ladder profile.

[jonathaneoliver]

Agile triage

• Story points: 3
• Effort: ~1 day (research + tabulation + a standards note)
• Priority: P2 (next sprint)
• Value: medium

Rationale

• 3 points / ~1 day: Bounded analysis work — pull two published Apple reference tables, compute adjacent rung ratios, and compare against our ladder.go defaults. No production code path changes for the core deliverable; the only unknown is whether the spec tables are as clean as expected, hence 3 not 2.
• P2: Informs the accuracy of our characterization conclusions (over-selection / over-downshift severity are read against ladder gap structure) but isn't blocking a shipping feature, so it can wait for the next sprint.
• value:medium: A meaningful correctness check on the rig's representativeness; payoff is conditional on the gaps actually diverging. If they do, the follow-up Apple-reference-ladder profile would carry the higher value.