How do you know we're not making this up?
The Skeptic Protocol
The Skeptic Protocol is our adversarial verification system: every factual claim we publish is checked, character by character, against a primary source document before it goes live. Claims that cannot be proven from the evidence are deleted, and the completed check is stored beside the product as its truth map.
The Hallucination Gap
"AI-powered" has become a warning label. Most gift sites use AI to generate content faster. They hide the process because there is no process worth showing. Prompt in, listicle out, publish.
The problem: AI makes things up. It sounds confident. It cites sources that don't exist. It says "hand-poured in Brooklyn" when the source only says "designed in Brooklyn." Small drifts compound across a pipeline until the "fact" has no connection to reality.
Our solution: Don't trust. Verify. Every factual claim on SleekNova is traced to a primary source with verbatim quotes. If we can't prove it, we remove it.
The 3-Step Verification
Every product page we publish runs the Skeptic Protocol: adversarial verification passes with one job, to break our own claims before you ever read them.
Claim Extraction
Every factual statement is identified in the product narrative. Founding dates. Material claims. Origin stories. Award mentions. If it states a fact, it gets flagged.
Adversarial Challenge
The sources come first: locked and human confirmed before any check begins. The verification pass is isolated. It receives only the claims and the source evidence, never our drafts and never our reasoning, and it attempts to break the match between each claim and its source. The goal is destruction, not confirmation.
Truth Mapping
Claims that survive are locked into our sn_truth_map with: the exact claim text, the source URL, and a verbatim quote from that source. Claims that fail are removed from the narrative. No exceptions.
Anatomy of a Grounded Fact
Every verified claim looks like this in our database:
{
"claim_id": "TM_001",
"claim_text": "Founded in 1851",
"source_ref": 1,
"verbatim_fact": "DENSHIRO has been working with the natural material of cherry bark, and making cherry bark ware since its foundation in 1851"
}This entry is copied byte for byte from the live database record of the Cherry Bark Candy Box. Nothing retyped, nothing trimmed.
See this claim live on the product page →
This isn't "we checked it." This is character-level verification. Our claim must match what the source actually says. Not close. Not paraphrased. Matched.
What Does "Grounded" Mean?
When you see the ✓ Grounded badge on a product page, it means:
- Every factual claim passed the Skeptic Protocol's adversarial verification
- Each claim is mapped to a primary source with verbatim evidence
- The product's
sn_grounded_statusis set to "grounded" - The audit trail is stored in our database
Products that fail verification don't get published. We'd rather have fewer products than false claims.
The Proof
We don't just claim rigor. We show our work.
The Lab
Products we rejected and why. If we never say no, our yes means nothing.
See rejections →The Methodology
Our full 152-point evaluation system. Scoring formula. Confidence Audit. Everything.
Read methodology →The Bar
The score floor a product must clear before a page exists at all. Most of what we look at never clears it.
See the bar →System Evolution
This protocol is not frozen. When a pass gets stricter, a law gets added, or a claim we published turns out to be wrong, we publish the change. The list below is read live from our three public records at the moment you load this page: corrections to what we published, rules of ours that changed, and capabilities the system gained. Nothing on it is hand-maintained.
- The Audit That Disagreed With Itself Eight product pages printed a confidence label their own score contradicted, and one rejection record gave a single audit factor 25 points out of a possible 20.
- We Rejected Them Correctly And Explained It Wrong Our rejection records said products are turned down for confidence below 60, gave a floor of 40 where it is 35, and in one case printed no reason at all.
- The Dataset That Said It Changed Every Day The Bar's machine readable dataset description stamped itself as revised today on every day, publishing a daily revision nobody made.
- The Gate We Applied and Never Published Below a Confidence Score of 35 we publish no verdict on a product. That floor decided real rejections and appeared on no page, and it is not the same rule as the 60 mark for the Labs Verified badge.
- The Corrected Page Now Says So Corrected pages showed no trace of having been corrected, corrections carried no name, and the busiest record took every slot on all four pages that display them.
45 entries published as of 10 August 2026: 19 corrections, 5 rule changes and 21 system updates.
Every correction → Every rule change → Every system update →Shop Grounded Gifts
386 products. 3,684 verified facts. And a public record of every time we were wrong.