Open Feed Recsys Lab: Verify, Audit, and Pretest X Algorithm Claims

Why this exists

The public X algorithm repo is useful, but a lightweight clone is not the same thing as a complete Phoenix model run, a live For You feed clone, or a reach predictor. This lab makes that boundary explicit before anyone writes an overconfident thread, tutorial, claim, or draft-post recommendation.

SourceCommit-pinned checkout and component inventory.

ArtifactsChecks for extracted Phoenix retrieval/ranker files.

DocsDetects README and config mismatches.

ClaimsVerdicts with file citations and share cards.

DraftsSynthetic audience reactions with explicit prediction boundaries.

Quick start

git clone https://github.com/zack-dev-cm/open-feed-recsys-lab
cd open-feed-recsys-lab
python3 skill/open-feed-recsys-lab/scripts/open_feed_recsys_lab.py

The script clones xai-org/x-algorithm with Git LFS smudge disabled, so the first report is lightweight and does not silently download multi-GB artifacts.

Audit a viral claim

python3 skill/x-algo-claim-auditor/scripts/x_algo_claim_auditor.py \
  --repo-dir /path/to/x-algorithm \
  --claim "Replies are worth 150x a like in the new X algorithm" \
  --output-dir /tmp/x-algo-claim-audit

The claim auditor returns supported, misleading, unsupported, or not_public_repo, then writes a Markdown report, JSON ledger, and SVG share card.

Pretest draft posts

python3 skill/tinytroupe-feed-research-lab/scripts/tinytroupe_feed_research_lab.py \
  --audience "AI builders and creator-operators interested in source-backed algorithm research" \
  --draft "I audited a viral X algorithm claim against public source. Verdict: misleading. What should I check next?" \
  --draft "Replies are king. Here is what the public repo actually proves." \
  --output-dir /tmp/tinytroupe-feed-research

The TinyTroupe-inspired lab compares drafts with deterministic synthetic personas and writes a report, CSV reaction table, JSON ledger, share card, and optional TinyTroupe experiment plan. It is qualitative research, not reach prediction.

Use an existing checkout

python3 skill/open-feed-recsys-lab/scripts/open_feed_recsys_lab.py \
  --repo-dir /path/to/x-algorithm \
  --output-dir /tmp/open-feed-recsys-lab-report

Pin a known commit or ref when you want a stable report:

python3 skill/open-feed-recsys-lab/scripts/open_feed_recsys_lab.py \
  --ref e414c17 \
  --output-dir /tmp/open-feed-recsys-lab-report

What the report tells you

run_report.md: commit, repo status, components, warnings, and optional command checks.
artifact_check.md: Git LFS pointer, expected Phoenix files, and missing artifacts.
architecture_map.html: Home Mixer, Phoenix, Thunder, Grox, Candidate Pipeline, and feed response flow.
manifest.json: machine-readable data for follow-up automation.

Run Phoenix only after artifacts exist

python3 skill/open-feed-recsys-lab/scripts/open_feed_recsys_lab.py \
  --repo-dir /path/to/x-algorithm \
  --artifacts-dir /path/to/oss-phoenix-artifacts \
  --run-phoenix

This is a reproducibility and research tool, not a growth hack. It does not promise reach, ranking, virality, revenue, or live-platform equivalence.