openclaw - 💡(How to fix) Fix [Ballast] Initial SKU research — 15–20 candidates across 3 categories [1 participants]

Context

Before any product can be listed on Ballast, it must pass the vetting rubric (data/scripts/PRODUCT_VETTING_PROTOCOL.md). This is the first manual vetting pass: research 15–20 candidate products, apply the rubric, publish the pass/fail results.

This is also the source of the real numbers the homepage will display ("We reviewed N products. M made the list.").

Categories to cover

For each candidate

Apply the four-gate rubric (G1–G4) and assign a tier (T1–T4):

G1 — Mechanism of action: Can it be stated in one sentence with a PubMed citation? G2 — Threshold compliance: Does the product's dose/spec meet the studied therapeutic threshold? G3 — Claim safety: Can it be described without "treats/cures/prevents"? Can we write 3 non-claims? G4 — Supplier quality: Does CJ Dropshipping carry it with a spec sheet? US warehouse available?

Example candidates to evaluate

Perimenopause: magnesium glycinate capsules, ashwagandha (KSM-66 extract), cooling weighted blanket (menopause-specific), perimenopause symptom journal/tracker, compression socks (hormonal edema)

Sleep: silk sleep mask (mulberry), mouth tape (myofunctional), sleep-environment thermometer, magnesium glycinate (sleep dose), melatonin-free sleep supplement

Nervous system: vagus nerve stimulation device (e.g. Pulsetto/Nurosym class), weighted lap pad, breath-pacer card deck, TENS-adjacent device, cold-exposure tools

Output format

For each evaluated SKU, produce:

{
  "name": "...",
  "category": "perimenopause|sleep|nervous-system",
  "tier": "T1|T2|T3|T4",
  "gates": { "G1": true, "G2": true, "G3": true, "G4": true },
  "mechanism_one_sentence": "...",
  "evidence_refs": [{ "text": "...", "source_url": "https://pubmed.ncbi.nlm.nih.gov/..." }],
  "nonclaims": ["...", "...", "..."],
  "verdict": "approve|hitl_required|auto_reject",
  "rationale": "..."
}

Save results to data/scripts/lib/shopify/vetting_results/initial_pass_2026.json.

Acceptance criteria

• ≥ 15 candidates evaluated
• At least one T4 auto-reject per category (proves the rubric has teeth)
• At least 8 approved (T1 or T2) across all three categories
• All PubMed URLs verified live (not hallucinated)

Related

• data/scripts/PRODUCT_VETTING_PROTOCOL.md — full rubric spec
• data/scripts/PLAN_AGENT_B_WELLNESS.md §3 — rubric summary
• Feeds: [How We Vet page] and [vetting.mjs implementation]