methodology v1.0

Estimates you can
argue with.

We do not have private invoices. We build a visible model from public product behavior, documented architecture, scale signals and current vendor economics—then widen the range when evidence is weak.

01

Map the workload

We inventory the product's expensive behaviors: AI inference, video processing, realtime collaboration, event ingestion, search, email, storage, delivery and reliability. A marketing page is not treated as the product.

02

Collect public signals

Priority goes to official documentation, engineering posts, open-source repositories, public pricing and product evidence. Third-party technology detectors may support a clue but do not become proof of the complete stack.

03

Model a scale

Each profile receives an explicit operating class and a low/high range. The model covers recurring hosting, compute, databases, storage, bandwidth, monitoring and usage-based vendors. Payroll, marketing, rent and one-time model training are excluded unless stated.

04

Price uncertainty

Private enterprise contracts, reserved capacity, credits and internal infrastructure make false precision dangerous. Thin evidence produces a wide range and a lower confidence label. Open architecture plus multiple direct sources produces a narrower one.

05

Estimate build time honestly

Build time means a production-grade core product with accounts, billing, observability, security, data durability and the defining workflow. It does not mean reproducing brand polish or copying protected data, networks, models or content.