DN Procurement Quote Benchmark and Savings Calculator v1.0
22 October 2026

Start here
Open dn-procurement-agent-savings-calculator-v1.0.html locally in a current browser to compare one complete quote with a comparable baseline. Inputs remain local. No data is sent or saved. The default is fictional case P02-B. The HTML file does not contact suppliers, negotiate or place an order.

Evidence scope
Eight authored RFQ cases contain 24 fictional offers. Three deterministic comparison policies produce 24 decisions. No AI model, commercial procurement product, live supplier, supplier conversation or real purchase was tested. These are controlled quote fixtures, not evidence of realized savings or a vendor ranking. All cost units, conversion inputs and baseline orders are fictional.

Files
- fixtures.json: authoritative authored RFQ and quote inputs, baseline basis and policy definitions.
- build_model.py: Python 3 standard-library implementation using exact fractions; no network access.
- offer-normalization.csv: all 24 normalized quotes and declared eligibility failures.
- decisions.csv: all 24 policy decisions and conditional savings fields.
- benchmark-results.json: nested inputs-to-results presentation data.
- summary.json: compact results and scope statement.
- methodology.txt: equations, cost/eligibility boundaries, comparison basis and limitations.
- dn-procurement-agent-savings-calculator-v1.0.html: editable, self-contained single-quote worksheet.
- verify_calculator.cjs: executable JavaScript calculation/validation checks using Node's built-in vm; no external packages.
- calculator-verification.json: coverage and outcome of those checks.
- independent-review-report.json: separate assistant recomputation of the benchmark.
- REPRODUCTION-VERIFICATION.txt: byte-reproduction and archive verification record.
- SHA256SUMS: file checksums.

Reproduce the benchmark
From the extracted directory, run:
    python3 build_model.py --output reproduced-results

The four generated files are offer-normalization.csv, decisions.csv, benchmark-results.json and summary.json. Compare their hashes with the originals. No dependency installation or API key is required.

Check the shipped calculator arithmetic and validation
Optional, with Node.js installed:
    node verify_calculator.cjs
This executes the actual HTML JavaScript with minimal DOM stubs. It is not a browser rendering test.

Interpretation
Use landed purchase costs only on a comparable basis. The modeled subtotal excludes lifecycle costs that are not entered. Ineligible and missing-cost offers do not support a savings claim. HOLD does not mean evaluation costs disappear. Owner-time valuation is noncash and separate from incremental software/external-verification spending. Do not sum savings across unrelated cases or treat hand-designed fixture counts as a measured field success rate.

Article: https://decentralised.news/agentic-procurement-benchmark-2027
Corrections: https://decentralised.news/contact
Author: Heath Muchena, with AI assistance. The separate review is assistant arithmetic/consistency review, not external expert certification.
