code-rl-briefing

Primary-source-driven briefing site for:

“What is the latest in code RL environments and human data?”

Dataset files

• public/data/accounts.json — seed + secondary account links
• public/data/papers.json — paper list + source links
• public/data/tweets.json — thread/post records with post/account links and quoted excerpts
• public/data/signals.json — structured trend synthesis from supplied notes

Site

• public/index.html renders all dataset files directly (link-driven view)

Run

pnpm install
pnpm dev

Run briefing pipeline (wired process)

# default: today + default topic
npm run run:briefing

# explicit date/topic
./scripts/run-briefing.sh 2026-03-04 "What is the latest in code RL environments and human data?"

This pipeline now does, in order (scripted step prompts):

1. step-01-papers-alphaxiv (AlphaXiv-first paper discovery)
2. step-02-supergrok-topic
3. step-03-supergrok-paper-discussion (exact query: "what are people saying about this exact paper title?")
4. step-04-supergrok-signals
5. step-05-supergrok-history-updates
6. step-06-exa-people (deep, category=people)
7. step-07-youtube-search (YouTube Data API)
8. step-08-synthesis

Step-07 uses youtube-transcript-api when available to attach per-video transcripts.

Separate YouTube channel-search workflow

python3 scripts/youtube_channel_search.py --query "RL and human data" --min-subs 5000

Outputs ranked channels with stats to a JSON file (supports --out).

Definition of "run briefing" is stored in briefing-process.md. Per-step prompts are stored in prompts/. Per-run artifacts are stored in data/runs/YYYY-MM-DD/steps/step-XX-* with prompt + story + raw + normalized outputs.

Deploy

pnpm run deploy