Scene · VISITOR 3 validated

VISITOR 3 — Waves down what looks like a small boat, but it's huge

VISITOR 3 running on PS1: Johnny on the island waving down a passing boat. The perspective gag reveals the boat is much larger than it first appeared.
VISITOR 3 on PS1 hardware. The perspective gag — Johnny waves what he thinks is a small nearby boat; it turns out to be a huge one farther out. The wide multi-view stitch this scene needs to render is also why the two perf-battle-card rows for VISITOR 3 are the lowest in the yellow band on /perf/ at the current release (around 97% target speed after the v0.8.2 + v0.8.6 release-tagged passes).

Validated on 2026-05-04.

Pack identifiers

  • ADS dispatch: VISITOR.ADS scene 3
  • Slug: visitor3
  • High-tide pack: VISITOR3.FG2
  • Low-tide pack: VIST3LOW.FG2

What this scene is

Johnny spots a boat in the distance and starts waving it down, thinking it’s a small one nearby. The perspective gag reveals it’s actually a huge boat much further out — and the size flip plays for the laugh. Confirmed by direct on-PS1 playback observation while capturing the chapter-select thumbnail; the earlier “yacht couple, photos” caption-mapping guess was wrong (no couple or photo-taking in the on-PS1 pack).

Validation notes

VISITOR 3 is not a plain base-diff scene. The validated high/low packs use the standard normal, far-left, and far-right foreground-only capture set, then a VISITOR3-specific synthesis helper builds the foreground source:

  • Foreground-only views keep the moving sprite pixels clean.
  • Full-host frames contribute the red ship hull only during live crash frames.
  • FGP3 temporal cleanup clears the post-crash blank rows instead of holding stale red.
  • Hold timing keeps the real source frame-158 splash visible without replaying stale right-side splash residue.

Production island placement remains variable. The capture/test island positions were evidence-gathering positions, not runtime pins.

Notable runtime history

VISITOR 3 remains one of the high-leverage yellow-band rows on the performance battle card at v0.8.12-ps1. After the compact metadata work, motion-copy payloads, high-only frame 117 target-hull promotion, the v214 high setup-prime pass, the v216 guarded low second setup segment, the v227 low frame-125/frame-126 resident re-anchor, the v234/v237 low frame-118/frame-127 resident copies, the v238 high frame-127/frame-130 resident-copy compaction, and the v248 low frame-114/frame-117 no-op residual compaction, the v327 low resident-slot swap, and the v338 low tail pack-only compaction, visitor3 high and low now run around 97.1% and 97.0% target speed instead of sitting in the red/orange bands.

Two of those experiment-version steps map cleanly to public release tags: v0.8.2-ps1 was the first release tagged specifically at VISITOR 3 (guarded-read performance pass), and v0.8.6-ps1 landed the setup-segment resident copies for frames 131 / 128 on both tides. The internal vNNN versions above are the performance-experiment-log sequence; the release tags are when batches of them shipped publicly. The wide multi-view stitch (the red ship crossing the full scene width) hits the prefetch window harder than most scenes; the remaining timing gap is concentrated in the ship’s live crash window and same-frame cleanup/restore work.

The arc that moved the rest of the matrix from 87.1% to 99.5% target speed is at /lab/from-87-to-99-5/. The named-experiment queue for visitor3 lives in docs/ps1/performance-experiment-log.md; recent rejected probes include several read-group, slack-gated, and setup-prime variants that didn’t beat the canary. The current source baseline has also removed unused noncompact motion dispatch without moving timings; that headroom is reserved for a future direct/precomposed VISITOR3 data path. Earlier read-plan clusters are closed for local runtime changes too: grouped-read, setup-owned persistent-segment, and data-only sector-alignment probes either measured exact-flat or shifted CD phase and regressed both tide variants.