The Response Ladder

Every answer the advisor gives sits on one level of the Response Ladder. The level says how much the answer rests on: your own words, a lab certificate, a cited source, or nothing at all. The advisor starts at the bottom and goes up only as far as the evidence earns. When the science is thin, it says so, and that is the product working.

The seven levels

1. Answers you write yourself. Store questions such as hours, shipping, returns and how to order. You write the answers, and the rules check them when you save. No model writes them.
2. What is in the product. Potencies, ingredients and test results, read from the product’s lab certificate. A product without a current certificate is held to the category education level for composition and never reaches a cited recommendation.
3. A medication check. A shopper names a medicine and the advisor checks it against cited interaction sources. When a source has nothing, the advisor says there is no data. It never turns no data into no interaction.
4. What the research shows about the category. What is known and what is not, with graded citations. This is education about the category, not a claim about your product.
5. A fit to what the shopper says they want. Products matched on verified attributes and on the shopper’s stated preferences, phrased as fit and never as therapy.
6. A cited recommendation. Said only where graded human evidence supports the statement, with the citations beside it. It is the earned exception, not the default.
7. A refusal with the reason. A claim the rules prohibit, or one nothing supports, is declined with a fixed explanation. Where the rules allow, the refusal comes with what the research shows in general.

The floor value in the API

Developers see the level as the floor field on every chat response and as the ladder_cap rule action in the API reference (Widget endpoints). The field holds one of seven values.

Level	floor
Answers you write yourself	0
What is in the product	1
A medication check	2
What the research shows about the category	3
A fit to what the shopper says they want	4
A cited recommendation	5
A refusal with the reason	"R"

A ladder_cap rule sets max_floor: the highest level an answer may reach for that product, jurisdiction or profile. When the answer would land higher, it is lowered to the cap. It is not turned into a refusal.

Two rules that hold at every level

The refusal rule. A refusal is the only response that carries floor: "R", and its wording is fixed by the rule that caused it. The advisor does not rephrase it, soften it or write a new one.

The no data rule. When a source holds nothing on a question, the answer says there is no data on file. It does not fill the gap, and it never reads silence as safety.