Grail Atlas principles

Five principles govern how Grail Atlas is built. They're here so you can hold us to them, and so we can hold ourselves to them.

1. Trust over engagement

Where trust and engagement contradict, trust wins. We don't run growth tactics that compromise the signal we publish. We don't make headlines more clickable by making them less true. We don't hide information that would lose us a sale (for us or for an affiliate partner) but would help the reader make a better decision.

The product-design test is “would I give this to my dad, my kid, or my best friend?” If the answer is no, the feature isn't the right shape yet.

2. Graded signals, never accusations

The risk read is a four-level grade (clear, caution, elevated, high), built from factual non-accusatory signals. We never label a specific seller a fraud or a scammer. The word “fraud” is banned from published copy by a code-enforced test.

This is two things at once. It's a legal posture — defaming a seller carries real exposure. It's also a harassment posture — “Grail Atlas called this seller a fraud” would carry through to social media and create a vector for real-world harm against people who may simply have a new account or a thin sales history.

See counterfeit awareness for the buyer-side rule that pairs with the editorial discipline.

3. Deterministic, auditable scoring

Every published score on Grail Atlas is a deterministic function of public inputs. There's no hidden weight, no “our proprietary algorithm,” no model that adjusts based on who the seller is or what they pay us.

Every score writes a record of its inputs (canonical-JSON SHA-256 hash), the engine version that scored them, and the outputs. The audit trail is the legal-defense evidence and the reproducibility check. See how it works for the pipeline.

4. Monetization separated from ranking

The trust composite, the valuation engine, and the risk read take no monetization input. Affiliate state is invisible to the scoring layer; an affiliate-tagged outbound link is decorated cosmetically at render time and never enters a database row that any score reads.

We say this here so it's a falsifiable claim. If we ever wire monetization into a published score, the disclosure page will say so explicitly, in language a hostile reviewer would write.

5. Honest absence beats false precision

When the data isn't there, we say so. A reference with fewer than five recent comparable sales gets a “low confidence” label and no deal tier. A listing without a clear platform tier gets a hedged trust read. A passed-in auction gets “appears delisted (last checked X)” not “unsold.”

Precision we can't defend isn't precision — it's confidence theater. The market punishes confidence theater faster than it rewards it.

How to hold us accountable

• Report a violation: Acceptable Use Policy contact
• See past mistakes: /postmortems
• Catalog state today: /stats
• The full methodology: /how-it-works