superstition-debunker

Use when a forwarded message, viral post, chain letter, or "they don't want you to know" claim needs a quick credibility read — detects the rhetorical fingerprints of folklore (vague authority, suppression urgency, forwarding quotas, unfalsizable mechanisms, single-case causation, unsourced percentages) across 6 pattern families, scores myth-signature strength, and generates the exact questions that would settle it. Structural triage, not fact lookup.

Superstition Debunker — Folklore Fingerprinting

Myths don't carry a "false" label — but they do carry structural fingerprints. "Harvard doctors confirmed," "share before this is taken down," "my cousin's neighbor died after," "90% of people" — these aren't evidence, they're costumes. Real information looks structurally different: named sources, checkable mechanisms, denominators, and no forwarding quota. This skill scores any claim against a catalog of folklore patterns and hands you the exact questions that would settle it.

When to Use

• A family group chat just received "URGENT, forward to 10 people"
• Deciding whether a viral health/safety claim deserves a fact-check rabbit hole or a one-line reply
• Teaching kids/parents media literacy with concrete patterns instead of "don't believe everything online"
• Triaging your own feed before resharing
• Building a family "is this real?" routine that scales

Don't use for: actual fact-checking (go to primary sources / fact-checkers for verdicts), hate/harassment content moderation (different problem), or scoring satire (structure is deliberately myth-shaped — that's the joke).

The Six Pattern Families

1. Authority mimicry — borrows credibility instead of earning it: vague "doctors confirmed," big-institution flex ("NASA proves"), secret/classified research (uncheckable by design).
2. Urgency & suppression — rushes you past verification: "they don't want you to know," "about to be deleted," forwarding quotas, "act before it's too late."
3. Unfalsifiable structure — arranges to survive being wrong: energy/vibrations/toxins with no measurable mechanism, "only works if you don't test it," ancient-wisdom appeals.
4. Causal sleight of hand — dresses sequence as cause: correlation-as-cause, single vivid case ("my friend's cousin"), post-hoc timing ("days later doctors found").
5. Numeric theater — numbers as costumes: precise percentages with no denominator or source, "9 out of 10 experts," micro-dose scares with no reference dose.
6. Emotional amplifiers — hijack sharing instincts: children at risk, gruesome imagery, luck/curse chain mechanics.

Workflow

1. Get the full text of the claim (screenshot transcript or paste).
2. Score it:

   bash

   python3 scripts/myth_analyzer.py analyze "URGENT: Harvard doctors
   confirmed 90% of people who... share with 10 friends before this
   is taken down" --log
   

3. Read the verdict: STRONG (score ≥10 or ≥4 families — treat as folklore until primary sources appear), MODERATE (≥6 or ≥3 families — verify before sharing), WEAK (≥3 — sloppy writing or possibly real), NO signatures (structure can't judge it; fact-check normally).
4. Use the per-pattern "ask" lines as your reply script — each is the question that would settle the claim ("Which study? Named institutions publish named people.").
5. --log accumulates analyses; stats shows which families dominate your feed — a media-diet readout.

Calibration Examples (from tests)

• Chain-message style claim (Harvard + share-to-10 + energy frequencies + cousin anecdote): score 10 → STRONG
• Boring real-science report ("2014 study of 40,000 nurses, 23% lower mortality, adjusted for smoking"): score 0 → NO signatures
• The tool reads rhetoric, not truth: a badly-written true story can score MODERATE, and a carefully-worded lie can score 0.

Common Pitfalls

1. Treating the score as a truth verdict. It's a rhetoric fingerprint. High score = folklore-shaped, low score = check facts the normal way. Say this explicitly when replying.
2. Running it on satire. The Onion is structurally mythic on purpose; the skill will happily flag it.
3. Missing screenshots with little text. The patterns live in words; a three-word claim has nothing to fingerprint — ask for the full forwarded text.
4. Replying with the score alone. The persuasion is in the questions ("which doctor? what study?") — lead with those, not with "my analyzer says 10."
5. Forgetting survivorship in your own feed. stats counts what you analyze, not what you see; it's a triage log, not an epidemiology of your inbox.

Verification Checklist

• Full claim text captured (not a summary of it)
• Verdict reported as "myth signatures," never "false"
• At least one settling question included in any reply
• If STRONG + health/medical: point to a primary source or professional, not just the score
• Log reviewed occasionally (stats) for family patterns

One-Shot Recipe: the family-chat reply

bash

python3 scripts/myth_analyzer.py analyze "<pasted message>" --log

Reply template: "This has the classic chain-message shape — unnamed 'Harvard doctors,' a forwarding quota, and a scary stat with no source. [The matched patterns.] The questions that would settle it: which study? what's the denominator? Real warnings cite checkable sources and never ask you to forward. Happy to dig if you find where it came from."