Where automated estimation is used
| Step | What goes in | What comes out | Who decides |
|---|---|---|---|
| Packed-box estimate | The item description typed into the calculator or sent to the API, the quantity, and for a pasted store link the page title | A box length, width, height and weight with a confidence and the assumptions made | The visitor sees the box on the rate card and can replace it with their own measurements before paying; a session an agent created shows the same box to the human at checkout |
| Product-page recognition | The title and visible product text of a store page the visitor pasted | The item name and, when the page states it, its dimensions | Same as above: the human sees the result and corrects it |
| Pre-payment packaging check | The item description and the declared contents of the parcel | A second opinion on the box and a flag when the contents match a shipping restriction | A flag stops the checkout and shows the reason; the human can correct the description or ask for a manual review by the founder |
No model decides anything about the person: there is no profiling, no scoring of a customer, no automated pricing of an individual, and no decision on a refund or a dispute is automated. Carrier prices come from the carriers; the SMKlog fee is a published formula. Restricted-goods refusals are rule-based and the rules are stated on the terms.
Which models, run where
The models are hosted by the providers named on the subprocessor list with their purpose, location and privacy terms: DeepSeek, OpenRouter (hosted models), OpenAI and Google. Requests go to a primary provider with a fallback when it does not answer in time. SMKlog sends these requests under each provider’s API terms; whether a provider retains the text it received is governed by that provider’s own policy, linked beside its name on the subprocessor list.
What the models receive: the item description and quantity, a pasted page’s title and product text, the declared contents. What they never receive: names, addresses, emails, phone numbers, payment details, tracking numbers or order ids. Addresses are sent to the carriers only.
Training data
- SMKlog trains no model and fine-tunes none. There is no SMKlog model; the models are the providers’ general models used as they are, with instructions written by SMKlog.
- Customer data is not used to train anything by SMKlog. The item descriptions SMKlog keeps are a catalog of estimated boxes, used to answer the same item faster next time and to publish aggregate price indexes; that catalog carries no customer identity.
- The instructions given to the models are checked against a set of manufacturer-published carton sizes that SMKlog collected from the makers’ own sites; that set contains products, not people.
- Should SMKlog ever train or fine-tune a model, this page will say so before it happens, with the data categories involved; the commitments page states the notice periods.
How the estimate is kept honest
An estimate can be wrong: a box that measures larger at the carrier than declared is billed by the carrier as an adjustment, which the human consents to at checkout and which SMKlog passes through at cost (see terms). That is why the box is shown before payment and why exact measurements always override the estimate. Published prices on guide pages come from captured quotes of a described box, never from a model’s price. Questions: info@smklog.com.