Frequently asked questions
What the diagnostic does, what it does not do, and how Residual Forge keeps the work bounded.
Answers
01
AI-heavy founders, agencies, and small software teams with a repeatable workflow that already functions but costs more than expected, is hard to observe, or defaults to expensive models without a verified quality requirement.
02
No. The engagement measures and tests against approved inputs, then produces a decision package. Production implementation is separately scoped and approved.
03
No. Results depend on the workload, data quality, provider pricing, volume, and the agreed quality threshold. A cheaper candidate is rejected when it fails a mandatory gate.
04
A workflow walkthrough, provider and model configuration, sanitized request samples or an approved evaluation set, available usage and billing exports, expected output rules, and relevant security constraints. Never send secrets or data you are not authorized to share.
05
Yes, when local inference is relevant and the data-handling plan permits it. Local models are evaluated against the same quality and operational gates as cloud models.
06
You can implement the backlog internally, stop with the evidence, or request a separate fixed-price implementation scope. The diagnostic does not lock you into Residual Forge or any AI provider.
07
Residual Forge is in its founding phase. The sample exists to show the method honestly without inventing customer proof. Customer results will only be published with permission and appropriate redaction.
08
A future provider-neutral usage, cost, quality, latency, routing, and budget product. Its scope will be shaped by repeatable patterns found in real paid work.
Residual Forge / founding cohort
Tell us which AI workflow you run repeatedly and where cost, latency, or quality is creating friction.