OFFLINE — two models, one brief

Unzip this entire folder and open index.html. All four films play locally.
Original and Revised switch between versions. Play both starts both players
from the beginning. Browser scheduling can introduce a small playback offset;
use individual controls for detailed inspection.

GPT-6.1 Sol (Codex) and Claude Opus 5.5 (Claude Code), both at High effort.
The same brief, local fonts and Remotion starter were supplied to each.
These are two provider workflows on one task, not a general model benchmark.
An earlier pair at Max effort was abandoned for both models after Opus was
still planning at about 24 minutes. Both then received a fresh High-effort
attempt with the same brief and a reminder to keep planning brief. The pair
was restarted for time, without selecting a visual winner. Models reviewed
and revised their own work within each attempt; no human hand-edits were
made to either model's original or revised films.

For a fresh comparison, copy source/starter twice into separate directories,
then give each agent BRIEF.md with the same effort setting and tools.

Editable source:
  source/sol/src/index.tsx       original Sol
  source/sol/src/index-v2.tsx    revised Sol
  source/opus/src/index.tsx      original Opus
  source/opus/src/index-v2.tsx   revised Opus

To render from either source directory (requires Node.js and internet to install):
  npm ci
  npx remotion render src/index.tsx Offline out/v1.mp4 --concurrency=2
  npx remotion render src/index-v2.tsx Offline out/v2.mp4 --concurrency=2

Local font files and their licenses are included. The brief describes a
fictional event. All four full films are 15 seconds, 1280 × 720, silent.
