# Clarity Trust

> Stage 4 of the stranger test — audit the trust layer. Proof proximity (evidence adjacent to the claim it supports), specificity over adjectives, credibility signals, and risk reversal at the moment of doubt. Scored 0-10.

- Skill: `modiqo/clarity-trust` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add modiqo/clarity-trust`
- Raw SKILL.md: https://api.skillmd.com/api/skills/modiqo/clarity-trust/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: modiqo (https://skillmd.com/u/modiqo)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/modiqo/clarity-trust

---


# clarity-trust — "Says who? What's the catch?"

Trust is not a section; it is a *property of claims*. A testimonial wall the
visitor scrolls past is furniture. What converts skepticism is evidence
**adjacent to the claim it supports, at the moment the doubt occurs**. This
stage audits every claim on the page for its accompanying proof, and the
page as a whole for the credibility signals that form before any claim is
read.

Consume the inventory's number census and section copy.

## Rubric

| Check | Weight | What to verify |
|---|---|---|
| **Proof proximity** | ×3 | For each material claim: is its evidence within the same viewport moment? A claim in section 2 proven in section 7 is unproven where it matters. Map claim → proof distance. |
| **Specificity** (Hopkins, Caples) | ×2.5 | Specifics beat superlatives: "2,562,207 tokens" outranks "massive savings"; "312 people ran this in July" outranks "loved by thousands". Flag every adjective doing a number's job. Precise numbers also read as *measured* — round ones as invented. |
| **Verifiability** | ×2 | Can the visitor check the proof? Named sources, linked evidence, live data beat anonymous quotes and logos. Anonymous or unverifiable proof scores as decoration. Fabricated-looking proof (stock faces, "John D.") is negative trust. |
| **Prominence–interpretation** (Fogg) | ×1.5 | What's prominent gets interpreted. Do the most visually prominent elements carry credibility or undermine it (ads, hype-words, dead links, lorem ipsum, stale dates)? |
| **Risk reversal at doubt** | ×1.5 | Where the visitor's cost/commitment fear peaks (near price, near signup), is there an honest reversal — guarantee, free tier, no-card trial, cancel-anytime, data-export promise? And is it *stated at that spot*? |
| **Consistency** | ×1 | Numbers must agree across the page (the inventory's number census makes contradictions mechanical to find). One contradiction costs more trust than ten proofs earn. |

## Anchored bands

- **9–10** — Every material claim carries adjacent, verifiable, specific
  proof; zero contradictions; risk reversed exactly at the doubt points.
- **7–8** — Claims proven but one or two at a distance; proof specific but
  one unverifiable item.
- **5–6** — Proof exists but pooled in a "trust section" far from claims; or
  specificity is patchy (adjectives carrying key claims).
- **3–4** — Superlatives largely unbacked; proof anonymous or generic.
- **0–2** — Claims contradict each other or the proof looks fabricated.

## Findings

Directions: `proof-adjacent`, `number-for-adjective`, `verifiable-source`,
`reverse-risk-here`, `reconcile-numbers`. Evidence = the claim quoted + the
location (or absence) of its proof. For specificity findings, advice must
name the *actual number the site could truthfully use* if discoverable —
never invent one; if the number doesn't exist yet, the advice is
"instrument, then state".

**The fabrication line (hard rule)**: never advise adding proof that doesn't
exist — no invented testimonials, counts, or logos. If the site lacks real
proof, the finding is "earn/instrument proof", and interim advice is honest
specificity about what *is* true (mechanism transparency, live demos,
open source, founder's name).

Doctrine suppression: sites with an execution-verified-numbers-only doctrine
(no testimonials, no logos) are audited for proof *within* that doctrine —
the question becomes whether the verified numbers sit adjacent to the claims
they prove.

References: `references/trust-experts.md`.