EC · Playroll — raw session data

Multi-POV Sync

Three players in the same GTA V (Enhanced) lobby on 2026-06-11, each running the Playroll recorder. Below: a 105-second excerpt of that session, with every signal playing off one shared clock.

Three synchronized POVs — 720p excerpts, wall-clock aligned via each recording's VIDEO_START epoch.
Video + inputs — 60 fps NVENC capture and the raw keyboard / mouse stream, frame-stamped.
Live speech — a whisper transcript per player, merged across the three mics (click a line to seek).
VLM segmentation — motion-scored good / bad clips with captions and tags.
Session metadata — encoder settings, fps quality breakdown, privacy-scrubbed hardware.
Raw files — the unmodified bundle artifacts, linked at the bottom of the page.

Not structured intentionality (yet) — the speech is natural teammate chatter; quests add the declare-then-execute layer on top of this substrate.

notebook · IT · preview streams ~26 MB/POV (720p excerpt); full-res originals total 5.7 GB

Synchronized playback — one clock, three recorders

alextrinsic lead clock · session t₀ + 450 s

brondix95 rec started +0.613 s

zoinkwiz rec started +2.315 s

t 0.00 ssession 450.0 s · epoch 1781187788.52 sync —

VLM segments

speech · A

speech · B

speech · C

loading bundle…

Signals on the shared clock

Live transcript — all three mics, merged chronologically (click a line to seek)

Input events — raw keyboard / mouse stream, frame-stamped

keys held now

VLM pipeline manifest — motion-scored segmentation + captions (Qwen3-VL, full config in file)

Session metadata — per-recording quality & provenance

The raw bundle — what a lab receives, per session

file	what it is	this session (lead POV)	raw

The three 720p clips above are excerpts re-encoded for this page; every other linked file is the unmodified artifact produced by the recorder and pipeline. Full-resolution source videos (1.4–2.7 GB each, 1440p60) are available on request. Hardware identifiers in meta.json are privacy-scrubbed at capture time (excluded fields).

Next layer Declared intentionality via quests (declare-then-execute, game-state-verified), microphone/video alignment provenance (wall_slot, sync QA), and per-window uncertainty for exploration steering. This page shows the substrate those attach to.

← back to the notebook