researchpublished findings

a local model for product selection

testing Bitrix, 1C and a local language model on a small catalogue.

a local model for product selection
actual interface of the checked testbed; November 2026 capture

01

question

can a model interpret a request while the server remains responsible for prices, stock and orders?

02

method

the local testbed uses Qwen 2.5 1.5B, Bitrix and a training 1C setup. i checked product selection, repeat submission and model failure. the server validates prices and quantities independently of model output.

03

findings

the server calculates the offer and checks stock. repeating a request does not create a duplicate order. model unavailability visibly switches the interface to rules. these scenarios were checked in the local testbed on 5–6 November 2026.

04

limitations

this is a local testbed with three demo products. it establishes neither quality on a large catalogue nor production readiness. availability of the public demo is not evidence of a load test.

sources and materials

findings apply to the stated version and dates. original reports are preserved; the linked evidence summary records scope and limitations.

← all research