[miguel-heygen]

fails without the fix: shows the first step's frame when seeking to 0 from %s s (three cases) in packages/core/src/runtime/adapters/gsap.steps.test.ts

What

A film built from repeated same-time sets (tl.set(all, hidden, t) then tl.set(one, visible, t) at every step) showed nothing on its first frame: at time 0 both sets are applied, but in reverse order, so the "hide all" set lands last. Every later step was right. The timeline re-render near 0 renders just above 0 and then steps back to 0, and the step back replays the sets at 0 in reverse; the code that puts them back in authored order only ran when a structure key was passed, which a render capture deliberately does not pass. It now derives the owner from the timeline's own children and always reapplies the steps at the landing time.

What I measured

• End to end on a Linux box, 12 sprites, each step set(all, opacity 0) then set(one, opacity 1) every 0.1 s, frame at 0 shows s0 when correct:

  	before (hyperframes 0.8.221)	after (CLI built from head caf9b4ce)
  snapshot --at 0	no sprite	s0
  render frame at 0 s	no sprite	s0
  render frame at 0.4 s	s4	s4

  Steps after 0 (0.1 to 1.0 s) were already correct before.

• The three new cases are red with gsap.ts at main, green with the change. packages/core runtime suite: 109 files, 2658 tests pass. oxfmt --check and oxlint clean.

What I did NOT execute

• The reporter's own project; the repro is my reduction of their description, and their sprites hid at 0.4, where my reduction already rendered correctly. The first frame is the case I could reproduce.
• Hooks skipped on commit (no node_modules in the worktree); the checks above ran on a Linux box.

Size

Lone fix nothing can carry: a few lines in the GSAP seek adapter and one parameterized test; it cannot be folded into another PR.

[github-actions]

Edit accuracy: accurate 271 (base branch 271), smooth 112 of those

The gate passes.
Smoothness is reported in the artifact, not gated. A case fails only if it fails 2 of 3 runs.

Quarantined, measured but not gated (0)

[heygengenesis]

Verdict: APPROVE · 0 blocker · 0 important · 0 nits · head caf9b4ce

Scope

Read gsap.ts and how init.ts drives it with structure undefined during render capture. The owner comes from getChildren(false, …)[0].parent, which is the real timeline even behind the bound wrapper. version: "" takes the uncached placeLandingMarks, so the per-revision cache can't go stale, and that path only runs for t < REDRAW_STEP. Capture and the adapter now reapply steps at 0 the same way preview already did. Red/green confirmed locally: all three shows the first step's frame when seeking to 0 from %s s cases fail with main's gsap.ts and pass at head, and the file passes 77/77. Every required check is green. The one red job, the optional Catalog payloads, failed with "Could not download the pinned catalog embedding model" (network, unrelated); I re-ran it.