What we tested in ForgeQuote /
The tests behind ForgeQuote: inputs, actions, observed results and the limits of that evidence.
Turn unstructured requirements into reviewed, precisely priced quotations with traceable catalogue versions.
Internal local testing, using synthetic inputs and owned assets. The results below establish the observed software behavior in those scenarios. They do not establish customer ROI, uptime or external deployment readiness.
Money and revision checks
Twelve ForgeQuote HTTP checks passed. The calculation fixture used three units priced at 1001 minor units with 1800 tax basis points: subtotal 3003, tax 541 and total 3544. Tests checked idempotent quote creation, rejected fractional quantities, stale versions and incorrect approval digests, plus revision invalidation after an approved quote's quantity changed.
Missing commercial facts remain blocking
An unknown catalogue SKU did not receive an invented price. The supported-money boundary either retains exact arithmetic or returns an explicit validation failure; the tested large boundary was covered by the harness. Separate workspace access could not read the quotation. These checks assess defined rules, not whether a user's catalogue rates or specifications are commercially correct.
One traceable model extraction
A consented synthetic request, Supply 12 BOLT-M8 each, completed through the application using Gemini 3.1 Flash-Lite. The requirement retained quantity 12, unit each and a verbatim source quotation. Consent false returned DATA_CONSENT. Replay and reload returned the same persisted execution data. This single short example does not establish general RFQ extraction accuracy or a latency guarantee.
A usable document through the browser
Separate browser QA imported a two-part catalogue, extracted the supplied requirements, saved and approved the exact revision, reloaded it and downloaded a real PDF. The inspected document showed INR 35.03 subtotal, INR 6.31 tax and INR 41.34 total for that separate fixture. The390 px layout had no document overflow and the authenticated flow had no runtime error.
What this evidence does not establish
Supported currencies are INR, USD, EUR and GBP with two-decimal money handling; quantities are positive whole-unit strings.
Rates, catalogue names, units, tax inputs and exception resolutions are supplied by the operator. The application does not validate their commercial correctness.
Extraction tests used owned synthetic text. Confidential customer enquiries require an appropriate provider/data-handling decision before any external model request.
No ERP inventory check, supplier availability, live price feed, outbound email or CRM write was verified in this local workflow.
An exported draft remains a draft. A downloaded document is not evidence that a recipient received or accepted a quotation.
A similar problem in your business?
Bring a real workflow, representative inputs and the result that needs to be reliable.
Shape a project brief