DN Procurement Agent Savings Calculator and Quote Benchmark v1.0
Authored and evaluated: 22 October 2026 (Africa/Johannesburg)

SCOPE
Eight authored requests for quotation (RFQs), three fictional supplier quotes per RFQ and three deterministic comparison policies produce 24 policy decisions. The candidate catalogue is fixed. This does not test open-web supplier discovery, an AI model, a commercial procurement product, genuine negotiation, supplier performance, contract formation or a purchase. The conditional discount case evaluates supplied revised terms; no conversation with a supplier occurred.

The units CU, QCU-A and QCU-B, conversion inputs, prices, approval flags, delivery promises and baseline orders are fictional. Each RFQ represents a separate one-time order. The eight cases are deliberately chosen contrasts, not a random or representative sample. Do not aggregate their savings or turn counts of passing fixtures into a real-world success rate.

PURCHASE QUANTITY AND COST
Packs = max(ceil(required units / units per pack), minimum packs).
Purchased units = packs * units per pack.
Excess units = purchased units - required units.
Item subtotal in CU = packs * quoted price per pack * (1 - discount percent / 100) * CU per quote-currency unit.
Landed purchase cost = item subtotal + total freight + nonrecoverable tax + one-off onboarding, all in CU.

All purchased packs are charged, including minimum-order excess. No residual value or future consumption credit is assigned to excess inventory. No intermediate result is rounded. The model uses exact rational arithmetic. CSV/JSON decimals are exported to eight decimal places, with trailing zeroes removed; recurring percentages are rounded half up. Displayed currency values normally use two decimal places. A missing required cost produces an unknown total, not a zero charge.

The subtotal is not lifecycle total cost of ownership. Financing, holding costs, inspection, maintenance, operating costs, quality losses, returns and disposal are excluded unless explicitly included in an input. Nonrecoverable tax is an authored amount; this model makes no tax classification or legal determination. A payment-period gate is a policy check, not a valuation of financing terms.

ELIGIBILITY
A quote is eligible only when all of the following supplied facts pass: approved supplier; matching specification; valid_until on or after the RFQ evaluation date; lead days no greater than the maximum; excess units no greater than demand * permitted overbuy percent / 100; payment days no less than the authorized minimum; negotiated terms authorized; and all five required cost inputs present. Dates use YYYY-MM-DD. The inputs are a declared policy envelope, not evidence gathered by the model. Real identity, quality, authorization and commercial conditions require verification outside this calculation.

POLICIES
1. sticker_price: choose the smallest raw numeric quoted price per pack. It ignores pack size, currency, discounts, charges and eligibility. This deliberately limited control is not a model of any named vendor.
2. landed_cost: choose the smallest complete normalized landed cost. It ignores non-cost eligibility conditions.
3. eligible_landed_cost: exclude every ineligible offer, then choose the smallest complete landed cost. If no candidate qualifies, return HOLD.
All ties are resolved by lexicographic offer_id; tie breaking is a reproducibility rule, not commercial judgment. The reported decision_status distinguishes ELIGIBLE_SELECTION, INELIGIBLE_SELECTION and HOLD.

COMPARABLE BASELINE AND SAVINGS
Each authored baseline is a compliant incumbent order for the same required quantity, specification, delivery deadline and permitted 30-day payment basis, including the modeled landed purchase costs. A different specification or unauthorized condition cannot establish a comparable saving. Even an eligible selection is only conditional arithmetic; no saving has been realized.

For eligible selections only:
Gross purchase saving = comparable baseline landed cost - selected landed cost.
Cash net saving = gross purchase saving - incremental workflow cash cost.
Cash net saving percent = cash net saving / comparable baseline * 100.
Owner-time valuation = owner review hours * owner-hour value.
Net after owner-time valuation = cash net saving - owner-time valuation.

Workflow cash cost includes the fictional incremental software/external verification cost for the comparison. Owner time is a separate noncash valuation, not a second cash invoice. No avoided baseline labor benefit is assumed. Do not enter the same labor both as cash cost and as noncash valuation. Negative results remain negative. An ineligible selection or HOLD suppresses all savings figures; the workflow and owner-time inputs remain recorded because withholding an award does not erase assessment costs.

EDITABLE OFFLINE CALCULATOR
The HTML file compares one reader-entered quote with one baseline. It reproduces the quantity, landed-cost, gate and savings rules but does not choose among multiple suppliers. Quotation validity is an explicit evidence checkbox rather than a date calculation. The default is fictional case P02-B. Baseline and conversion input must be positive. Costs and time inputs must be known nonnegative ordinary decimals with at most six decimal places; scientific notation is rejected. Integer quantity/pack inputs are bounded to 1–1,000,000 and day inputs to 0–3,650. All checked evidence conditions, delivery, payment and overbuy gates must pass for savings to appear. The file sends and stores no data and does not enforce account permissions or authorize an award.

REVIEW
The separate assistant review report records independent recomputation of all 24 normalized offers and 24 policy choices. The calculator verification runs the actual shipped JavaScript with minimal DOM stubs and compares all 23 fully costed quotes, plus boundary and invalid-input cases. This checks arithmetic and validation, not a browser layout or a live procurement operation. See the respective reports for exact coverage. The article's frontend checks are separate from this public model package.

CHANGE LOG
v1.0 — initial authored fixture set, reproducible comparison policies, editable single-quote calculator and independent arithmetic review. Correction route: https://decentralised.news/contact.
