Methodology

How estimates are produced

• • Workload classification is deterministic and based on visible inputs such as volume, latency, review requirements, and sensitive data.
• • Task decomposition uses a bounded set of execution methods: deterministic rules, retrieval, small-model classification, targeted generation, and human review.
• • Token estimates are approximate and use labelled assumptions; they are not a claim of actual production quality.
• • Cost totals are computed from the visible model catalog, the selected calls-per-execution, and the retry assumptions.
• • A model recommendation is rejected when it fails the budget, privacy, provider, or capability constraints.

Constraints that override a model recommendation

• • Privacy or residency requirements.
• • Missing price data or unsupported model capabilities.
• • Budget thresholds that cannot be satisfied without violating the user's minimum constraints.
• • Manual review requirements for regulated or consequential decisions.
• • Local-only or provider-approved restrictions.

Why this is a demo-first design

BuildWise is intentionally transparent about estimates and limitations. It shows realistic scenario economics without claiming live model quality or a measured production result. The live BYOK path is available for controlled tests, but the default product remains functional without an API key.