Industry · Health and wellness
A wellness store runs on the neutral pack.
No dedicated pack is built for this sector yet, so a shop selling supplements, vitamins, teas or recovery kit lands on the neutral one. Twelve tools exist in the code, nine are handed to the model from the first conversation, and the base rule set every store gets is merged in whatever the shop sells. This is the trade that set was written closest to, and the trade where the most useful thing Refik does is decline. This page counts what runs on day one, where the declining is wired, and the one class of sentence the base set still lets through.
- Products
- Variants & options
- Prices
- Live stock
- Collections
- Order status
- Returns
- Store policies
- Answers you write
- A code you choose
A pack changes what Refik asks; it leaves the ten sources an answer is built from exactly where they were. A wellness store reads the same ten as a placed one — what Refik reads.
What runs
Nine tools, handed over on the first day.
Ten of the twelve tools are present under every pack; a pack only decides whether shade matching and a remembered profile are offered. The ordered-routine tool is the third one that stays shut here, and it is shut by your catalogue rather than by the pack: it is offered once at least one chain carries a product role, and roles are the thing a pack supplies. Order status, guest order lookup, the return request, the shipping and returns lookup, the stock-back e-mail and the handover to your team all work from the first conversation, and the delivery answer comes out of your own policy pages rather than out of memory — as two legs plus a total, because quoting the transit leg alone understates the wait a shopper is told by about a week.
Your own words do the matching
The search tool runs before any recommendation and reads what the shopper actually typed: it resolves barcodes. That leg is code and the same under every pack; the Spanish and Turkish words — and the misspellings — it maps are hand-written entries in the pack’s synonym dictionary, which here is an empty object. Nothing is corrected by algorithm: matching is substring-based, so a word merely cut short still lands and a transposed one does not. An item that is out of stock never appears in what it returns, and when one is wanted anyway the stock-back e-mail is offered right after the closest alternative. A line carrying a brand, a form and a count is answered out of whatever your titles, tags and collections already say, in the language your catalogue is written in. What it will not do is produce a strength it was never given: a milligram figure that is not in your own product data is not in the answer either.
A usage question is read off the listing, not composed
When a shopper is unhappy, the model is told to reach first for the tool that hands back that product’s own step-by-step instructions along with what usually goes wrong — ahead of any talk of a return, a credit or a discount. It is then told to read those instructions, ask one specific diagnostic question, and only move on to a return if the product genuinely does not suit them. The “what usually goes wrong” examples belong to a trade and arrive with a pack; under the neutral pack the sentence ships without them on purpose, because writing this trade’s typical mistake on a pack’s behalf, having measured none, is the leak the design was built to avoid.
Counted, not written
The question that matters most here never reaches the model.
The base set carries nine patterns that mark a question as belonging to a professional rather than to a shop: pregnancy and breastfeeding, an allergy, a long-term condition — diabetes, asthma, blood pressure, a heart condition, epilepsy or the thyroid — a prescription or a medication already being taken, the shopper’s own doctor, a question about whether something suits a baby or a child, and surgery. Every one of those nine fires in this trade, which is true of no other sector on this site. When one matches, the request is not sent to the model at all: the reply is fixed text, it offers the product’s full ingredient list, and it points the shopper at a professional. That is the whole design — a model handed a health question will try, in good faith, to be helpful.
9
patterns in the base set that send a question to a professional, and every one of them fires in this trade.
0
model turns a matched question costs. The wording is fixed, so there is nothing left to improvise.
8
of the sixteen base claim rules are written in words a wellness listing actually uses.
Eight of sixteen rules, five families
Of the eight families the base set covers, five are written about this trade and account for eight of the sixteen rules. Three block wording that says a product treats, cures or heals a named condition, in English, Turkish and Spanish. One blocks wording that puts a product in place of a medication, a prescription or a doctor. One blocks wording about stopping an illness or killing a virus. One blocks an evidence claim — clinically, medically, scientifically or lab proven, and the miracle framing. Two block age-restricted wording a shop cannot lift with a sentence of its own.
The rules are read on the way out
They are not a filter on what a shopper may type; they are read on the answer Refik is about to send. When one matches, the answer is produced again with the rule named rather than deleted, so the shopper is not left without a reply. A pack’s own rules sit in front of the base ones and a pack rule is the first one reported; a base rule that would report the same words twice stays quiet.
A rule only catches the words it was written for
A rule is a pattern over words, so it catches the sentence it was written for and nothing else. The three languages are covered unevenly too: of those eight rules, five are in English, two in Turkish and one in Spanish, so the same wording is not stopped equally whichever language a shopper uses. That is why the next section counts what is missing instead of calling the set finished.
All twenty-five of them are merged in whatever the shop sells — what the base set carries.
What this page is a picture of
scripts/sektor-maket-yaz.py.Measured, and still open
Where the base set stops, counted.
Eight probe words were checked against all sixteen patterns, each with a boundary on both sides so a short word sitting inside a longer one could not be read as a match. The check was calibrated in both directions before it was trusted: a word known to be in the set was found, an invented one was not. Two of the eight are in the patterns. Six are not, and the six are listed here rather than left out.
The two that are there
disease and symptoms both appear, in the rules about treating a condition and about stopping an illness. That is the half of this trade the base set was written with in view: the vocabulary of illness, in a shop that does not sell medicine. A sentence that names a condition and offers a product against it is what those rules were built to catch, and it is caught in all three languages.
The six that are not
supplement, dose, dosage, therapeutic, detox and immune appear in none of the sixteen patterns. Wording about supporting immunity, about a detox effect, or about how many capsules a day something is taken at is matched by no rule in the base set, in any language. The gap is not a bug in the rules; it is the edge of what sixteen patterns written for every trade at once can reach.
Of those six, the quantity question is the quietest. A shopper can ask how much of something to take without naming a condition, an allergy, a medication or their doctor — and with none of the nine handover patterns matched, that question reaches the model like any other. The nine are what make this the trade the base set covers closest, and this is the gap they leave: they key on who is asking and about whom, not on what the answer would amount to. Writing it here is the only honest thing available today, because the rule that would close it belongs in a pack and no pack exists for this trade.
From the schema
Four things a pack for this trade would add.
A pack is a file of named fields, and the schema says which of them a pack has to carry. These four are the ones this trade would fill, in the order of what they unlock.
-
Claim rules in this vocabulary
A pack carries its own claim rules and they sit in front of the base ones, so this is where the six absent words would get a pattern. A pack rule can also be bound to a market, which matters here more than in most trades: what a listing may say is not the same sentence in every country. The base sixteen stay merged in underneath, and the pack rule is still reported first when both would fire.
-
An active-ingredient dictionary
The schema asks every pack for a list of active-ingredient keywords, searched in your own ingredient text. Without one, an ingredient is just another word in a title; with one, a question about an ingredient is matched against the products that carry it. The field is required rather than optional, so a pack for this trade could not skip it.
-
Replenishment intervals
A role maps to a lifetime in days, and a reminder candidate is only produced between twenty-one and two hundred and forty days after the order. A daily count against a pack size is arithmetic, which makes this one of the few trades where that lifetime is a measurement rather than an estimate. Under the neutral pack the candidate list comes back empty for two independent reasons: the switch is one of the three a pack opens, and the role-to-days table is where a pack would put the days.
-
Wording for what gets said instead
The limit on claims lives in the code and applies under every pack; what a pack adds is the other half — short examples of what gets said in place of the sentence a rule blocks. That field is split by store language and only Turkish and English are read today, so a Spanish slice can be written and nothing will ever read it. The same pack would fill the survey object, whose three members in the neutral file are blank.
Unchanged
What stays yours, under every pack.
These three are the same on every vertical page on this site, and that is the point: they describe the part of Refik that your store decides and a pack only reads.
You choose it, and the server measures again
Refik counts how many of your products carry any of a pack’s words and shows you the count, the ranked score, up to eight of the words it matched and up to three example products for each. A pack is only suggested when the share clears a threshold, at least three different words matched and at least five products matched. You confirm it, and the server runs the measurement again on its own before it writes anything. How a pack gets chosen.
Your catalogue supplies the products
A pack supplies names; your catalogue supplies the products. Every option Refik offers has passed two gates in your own store — sellable, and in stock — so the products a shopper sees are the ones you carry today. A pack tag that matches zero products in your catalogue is counted and shown in the console, with the number it matched.
The base set stays merged in
Sixteen claim rules and nine handover patterns are merged in when any pack loads, in English, Turkish and Spanish, and they sit behind the pack’s own rules so the pack’s rule is still the first one reported. A base rule that would report the same words twice stays quiet. An empty set makes the app refuse the pack outright. The base set.
A pack arrives for this trade the day a catalogue in it is measured, and until then the neutral file is the honest one — the neutral pack, counted. The industries page lists the packs that exist and the three gates a store has to clear for one of them; this sector sits among the stores those gates leave unplaced — how a pack gets chosen.