How we grade claims
PeptIQ scores statements, not people. Leaderboards aggregate claim-level evidence grades so the public can see patterns — with sources, limitations, and a path to correct mistakes.
Pipeline
- Ingest a public Instagram, Reddit, or caption link (transcript + caption when available).
- Extract discrete peptide-related claims with an LLM.
- Retrieve supporting or contradicting literature (PubMed, Europe PMC, Semantic Scholar, ClinicalTrials.gov).
- Assign a verdict per claim and reconcile an overall post verdict.
- Publish a shareable report with citations. Account scores update from claim tallies.
Three narrative buckets
Charts group the four verdicts into a story people can share: what is backed, what is wrong, and what is grey — like animal data presented as human fact.
Evidence-based
Supported claims that match published evidence at the stated strength.
Factually incorrect
Misleading claims that conflict with evidence or invent certainty.
Grey / overstated
Overstated or no-evidence claims — e.g. “BPC-157 fixes Achilles” from rat tendon models, not proven human Achilles repair trials.
Verdict definitions
✅ Supported
Claim aligns with published evidence at the stated strength.
⚠️ Overstated
Direction may be plausible, but magnitude, certainty, or audience framing exceeds the evidence.
🚩 Misleading
Claim conflicts with evidence or omits critical safety/context in a way that misleads.
❌ No Evidence
We could not find adequate published support for the claim as stated.
Evidence sources
PubMed / MEDLINE
35M+ biomedical articles
Europe PMC
Preprints + open access
Semantic Scholar
AI academic graph
ClinicalTrials.gov
Human trial registry
Leaderboard math
Science credibility score = (Supported + 0.5 × Overstated) ÷ total claims × 100. Accounts need at least 3 audited claims to appear on public boards. Labels use evidence-grade language (“High evidence grade”), never insults or fraud accusations.
Inclusion on the priority watchlist is not an endorsement or accusation — it is a backlog of high-reach peptide voices we want to audit fairly.
Limitations
- Science evolves; older citations may be superseded.
- Transcripts and captions can miss context, sarcasm, or disclaimers.
- AI extraction can err — humans can request review via Right of Reply.
- Not medical advice. Educational use only.