---
name: science-claim-verifier
description: >
  Scientific literature verification for health and wellness products — two
  modes that bookend every product build. Use this skill BEFORE building any
  health product (supplements, medical test prep, nutrition, wellness) to
  discover what the evidence actually says and generate a science-accurate
  content brief. Use it BEFORE shipping (before or during /shipcheck) to verify
  every specific claim in a finished product against peer-reviewed literature
  and flag anything wrong, overstated, or contested. Triggers automatically
  when the user starts a new health/wellness product ("I want to make a guide
  about X supplements", "let's build a X prep guide"), when /shipcheck runs on
  a health product, or when the user asks "is this accurate?", "check the
  science", "verify these claims", "what does the research say about X". Also
  activates if adversarial-critic flags scientific issues on a health product.
  Do not skip this skill for health products — one wrong claim in a
  science-branded product destroys the PhD differentiator.
---

# science-claim-verifier

Scientific literature verification for health and wellness digital products.
Two modes, same tool. The goal is always the same: make sure every claim
in the product is something a peer reviewer would be comfortable with.

---

## When to activate

### Mode 1 — Pre-build Research
Activate when a new health/wellness product is being scoped. The user is
about to build something and needs to know what content to include — and
what to avoid.

Trigger phrases: "I want to make a [supplement / prep / condition] guide",
"let's build a chart about X", "what should I cover in a guide about X",
"what does the science say about X" (when a product is the context).

### Mode 2 — Pre-ship Verification
Activate when a health product is nearly done. The user is about to list it,
run /shipcheck, or has just finished building. Check every specific claim.

Trigger phrases: "is this accurate?", "check the claims", "verify the
science", /shipcheck on any health product, adversarial-critic flags
scientific issues, or user says a product is done/ready.

---

## The core problem this skill solves

Health content has two failure modes:
1. **Overstated claims** — technically based on evidence but presented more
   strongly than the evidence supports ("increases absorption by 3×" when
   that's a single-meal fasting study result, not a dietary context result)
2. **Outdated or wrong claims** — based on what was believed to be true but
   has since been revised ("probiotics survive better on an empty stomach"
   — contradicted by 2024–2025 research)

Both failure modes damage credibility, invite refund requests, and — most
importantly — give buyers incorrect health information. This skill catches
both before they reach a customer.

---

## Mode 1: Pre-Build Research Brief

### Purpose
Discover what the evidence actually says about a topic before writing a
single word of product content. The output is a content brief that the
product build can be structured around.

### Process

**Step 1 — Understand the scope**
Identify the health topic and the type of product being built. What specific
claims will this product need to make? What are the core questions a buyer
would have?

Example: "Supplement timing chart" → questions like: Does timing of fat-
solubles matter? What's the evidence for empty-stomach absorption? Is there
actually a best time for probiotics?

**Step 2 — Run targeted literature searches**
Use PubMed (peer-reviewed literature). Run 3–6 searches covering the key
claims the product will need to make. Search for the specific mechanism,
not just the general topic.

Good search strategy:
- Search the specific claim mechanism ("probiotic survival gastric acid food
  matrix") not just the supplement name
- Use PubMed filters to narrow to RCTs, meta-analyses, or systematic reviews
  when available (use the Article type filter or add "[Filter]" terms)
- Run a separate search for any claim that feels intuitive but unverified
- Search for contested areas ("X controversy" or "X conflicting evidence")
- Prefer papers with DOIs — record the PubMed URL (pubmed.ncbi.nlm.nih.gov/PMID)
  for each paper used

Rate each search result by evidence strength:
- **Strong**: Meta-analysis, systematic review, or multiple RCTs with
  consistent findings, high citation count
- **Moderate**: Single RCTs or well-designed observational studies
- **Weak**: Small studies, animal/in vitro only, or conflicting results
- **Contested**: Active disagreement in the literature

**Step 3 — Produce a content brief**

Format:
```
## Science Brief: [Topic]

### Well-supported claims (safe to state directly)
- [Claim] — [Evidence basis, study type, citation count context]

### Claims that need qualification (true but needs nuance)
- [Claim as often stated] → [How to state it accurately]
- Reason: [What the evidence actually shows vs. the common framing]

### Claims to avoid (not supported or contradicted)
- [Claim] — [Why: what the evidence actually shows]

### Evidence gaps (honest "we don't fully know yet")
- [Area where evidence is insufficient to make a strong claim]

### Recommended language
Specific phrasings that are accurate and still useful for a buyer:
- Instead of "[overstated version]" → "[accurate version]"
```

---

## Mode 2: Pre-Ship Claim Verification

### Purpose
Go through a finished product claim by claim and verify each one against
the literature. The output is a verification report with a clear status
for each claim and corrected phrasing where needed.

### Process

**Step 1 — Extract all specific claims**
Read the product and list every factual claim that could be verified:
- Mechanism claims ("X absorbs better with fat")
- Quantity claims ("0.5–3 mg is sufficient", "separate by 2+ hours")
- Interaction claims ("Iron + Calcium compete for the same transporters")
- Comparative claims ("more does not improve sleep quality")
- Categorical claims ("fat-solubles need dietary fat to absorb at all")

Do not list general statements or disclaimers — only specific, verifiable
factual claims. Each claim gets its own verification.

**Step 2 — Search and verify**
For each claim (or related cluster of claims), run a targeted PubMed
search. Focus on the specific mechanism or number being claimed.

Batch related claims where possible (e.g., all iron-related claims in one
search pass) to avoid redundant searches.

**Step 3 — Rate and report**

For each claim:

| Status | Meaning |
|--------|---------|
| ✅ Verified | Well-supported by current evidence (meta-analysis, multiple RCTs, or strong consensus) |
| ⚠️ Qualified | Supported but needs nuance — often true in specific conditions, overstated as written |
| ❌ Contradicted | Current evidence contradicts the claim |
| 🔍 Unverified | Could not find direct evidence — needs caution or removal |
| ⚙️ Mechanism-only | Supported by mechanism/in vitro but limited human trial data |

Output format:
```
## Claim Verification Report: [Product Name]

### ✅ Verified Claims
- "[Exact claim as written in product]"
  Evidence: [Source type, e.g., "Meta-analysis, 2017, 26 studies"]

### ⚠️ Claims Needing Qualification
- "[Claim as written]"
  Issue: [What's overstated or missing]
  → Suggested fix: "[More accurate phrasing]"

### ❌ Contradicted Claims — Fix Before Publishing
- "[Claim as written]"
  Evidence: [What the literature actually shows]
  → Required fix: "[Corrected claim or removal recommendation]"

### 🔍 Unverified Claims — Verify or Soften
- "[Claim as written]"
  Action: [Recommend softening with "may", "some research suggests", or removal]

### Summary
Total claims checked: N
Verified: N | Qualified: N | Contradicted: N | Unverified: N

Ship-blocking issues: [list or "none"]
```

---

## Evidence quality hierarchy

When interpreting search results, weight evidence in this order:

1. **Systematic reviews and meta-analyses** — strongest; reflects pooled
   evidence across multiple studies
2. **RCTs (randomized controlled trials)** — strong; especially with high
   citation counts and large sample sizes
3. **Observational/cohort studies** — moderate; good for prevalence, weaker
   for causation
4. **In vitro / animal studies** — weak for human claims; can inform mechanism
   but not outcomes
5. **Expert consensus / clinical guidelines** (NIH, WHO, EFSA) — strong for
   established practice even when RCT data is thin

Always note when evidence is from a specific population (clinical patients,
athletes, burned patients) and whether it generalizes to the target audience
of the product (healthy adults managing their supplement routine).

---

## Critical rules

**Never fabricate or assume.** If a search returns no relevant results for
a claim, mark it 🔍 Unverified. Do not assume the claim is correct because
it "sounds right."

**Context matters.** A claim supported in fasting/single-meal studies may
not hold in a complete diet context (e.g., Vitamin C + iron absorption).
Always note when the evidence context differs from the product's real-world
context.

**Distinguish mechanism from outcome.** "Fat-solubles need fat" (mechanism,
well-established) is different from "taking D3 with olive oil doubles
absorption" (specific outcome, needs specific evidence).

**Population specificity.** Evidence from clinical/hospital populations
(burns, bariatric surgery, athletes) does not automatically generalize to
healthy adults. Flag this when it applies.

**Flag safety-critical issues first.** Drug interactions (5-HTP + MAOIs),
toxicity warnings (Vitamin A doses), and contraindications in pregnancy
are always flagged as ❌ if missing or understated, regardless of how
minor they seem in context.

---

## Bibliography documentation (always required)

Every product must have a bibliography file. This serves two purposes:
1. **Internal accountability** — if a claim is ever questioned, there is
   a paper trail showing exactly where every claim came from
2. **Etsy bonus download** — the bibliography can be offered as a separate
   downloadable file in the listing, a genuine differentiator no competitor
   offers ("Sources & Scientific References included")

### Where to save it
```
products/[product-folder]/bibliography/
  sources.md          ← human-readable citations (internal + Etsy download)
  sources-internal.md ← extended notes, evidence ratings, contested claims
                        (internal only, not for customers)
```

### When to update it
- After **Mode 1 (pre-build)**: add all papers consulted during research
- After **Mode 2 (pre-ship)**: add any additional papers used in verification
  that weren't already in the file
- Any paper that informed a claim, even if it was used to decide NOT to
  include something, goes in `sources-internal.md`

### Citation format for sources.md (customer-facing)
```
## Sources & Scientific References
*[Product name] — Salvia Silvestre Co.*

All claims in this guide are based on peer-reviewed scientific literature.

### [Category, e.g., "Supplement Absorption & Timing"]
1. [Authors] ([Year]). [Title]. *[Journal]*. [DOI or URL]

### [Next category]
...

---
*This reference list reflects the literature available at the time of
publication. Science evolves — always consult your doctor for personal
medical guidance.*
```

### Citation format for sources-internal.md (internal records)
Same as above but with additional fields:
```
1. [Authors] ([Year]). [Title]. *[Journal]*. [DOI or URL]
   Citations: [N] | Evidence level: [Strong/Moderate/Weak/Contested]
   Used for: [Which claim this supports or refuted]
   Notes: [Any caveats — population specificity, context limits, etc.]
```

### What counts as a citable source
- Peer-reviewed papers found via PubMed (record the PubMed URL)
- NIH/WHO/EFSA official guidelines or position papers
- Cochrane reviews
- Do NOT cite: blog posts, supplement brand websites, news articles,
  Wikipedia, or anything without a DOI or institutional URL

---

## Integration with other skills

- Runs **before** `product-brief-generator` on health products — supplies
  the evidence base that informs what goes in the brief
- Runs **before or alongside** `/shipcheck` on health products — feeds
  findings into the ship/no-ship decision
- Pairs with `adversarial-critic` (Scientific Peer Reviewer persona) —
  adversarial-critic identifies what to challenge, this skill verifies
  whether the challenge is valid
- Results feed directly into fixing the product or updating the content brief
- Always produces or updates the bibliography files — this step is not
  optional even when PubMed returns limited results (use what was found)

---

## Example search queries (reference)

These show the level of specificity needed. Vague queries return vague results.

| Claim to check | Good search query |
|---|---|
| "Probiotics survive better before food" | "probiotic survival gastric acid food matrix timing" |
| "Vitamin C increases iron absorption 3×" | "vitamin C non-heme iron absorption bioavailability meta-analysis" |
| "Melatonin: 0.5–3mg is sufficient" | "melatonin optimal dose low dose sleep 0.5mg" |
| "Zinc depletes copper" | "zinc supplementation copper depletion deficiency" |
| "Ashwagandha lowers cortisol" | "ashwagandha cortisol reduction RCT" |
| "Magnesium glycinate best for sleep" | "magnesium glycinate sleep quality comparison forms" |