KasiStock Demo workspace

How decisions work

Clear authority at every step

KasiStock deliberately separates reading evidence, calculating a plan and authorising a purchase. This keeps AI useful without letting it control money or final quantities.

3
authority layersInterpret · Calculate · Approve
01

AI responsibilities

Extract messy evidence, propose product matches, and explain deterministic results.

Proposes
02

Deterministic responsibilities

Currency arithmetic, stock coverage, demand constraints, optimisation, and approval state.

Enforces
03

Human responsibilities

Resolve uncertain evidence, approve product matches, and authorise purchase orders.

Authorises

End-to-end control

From source evidence to an approved order

  1. 1
    Evidence is uploaded

    Shelf images, supplier lists and sales history remain tied to their original source.

    Retailer
  2. 2
    Structured facts are proposed

    AI extracts product details and uncertainty through validated contracts.

    AI
  3. 3
    Uncertainty is resolved

    A person accepts, corrects, rejects or leaves unsupported values unknown.

    Retailer
  4. 4
    The plan is calculated

    Deterministic code calculates demand, pack constraints, supplier costs and budget allocation.

    System
  5. 5
    The draft is locked

    The server rebuilds the order from accepted evidence and signs the exact result.

    System
  6. 6
    The purchase is approved

    An authorised person reviews the total and explicitly approves supplier orders.

    Retailer

Why it matters

The browser cannot invent an approved order

Edited quantities are treated as requests. Before approval, the server checks accepted evidence, confirmed mappings, demand ceilings, pack sizes, budget and signatures again. Any material change requires a new draft.