Glossary

Document status: current facts. This page explains the recurring domain terms used across the public Ingot documentation. Specific numbers and maturity follow current status, and product positioning follows the brand guidelines.

Runs and recipes

• Run: The complete record of one real production execution that links actual conditions, process trajectory, production context, and quality outcomes through a single run identity.
• Recipe: A set of process settings for a specific product and equipment; the object the "next recipe" recommendation proposes.
• Recipe version: Versioned configuration that defines variables, units, quality rules, and safety boundaries.
• Operating region: The validated parameter range and safety boundaries that optimization recommendations must not exceed.
• Process trajectory: The stage-by-stage characteristics of a run over time, used to locate where a deviation occurs.

Evidence and admission

• Run evidence: A set of facts linked through one run identity, traceable to their source and independently reviewable.
• Admission: The completeness, comparability, and coverage checks that decide whether a real run may enter analysis; when they fail, recommendations stop and the reason is recorded.
• Optimization observation: A real recipe run plus its quality outcomes that passed quality and coverage admission, forming a reviewable optimization sample.
• Comparability: Whether different runs are sufficiently alike in material, equipment, tooling, time, and product to be compared.
• Evidence gap: A missing piece of information recorded explicitly when current facts cannot yet support a conclusion.

Analysis and recommendations

• Process diagnosis: Comparing runs that satisfy comparability conditions to form candidate causes, counter-evidence, and evidence gaps, rather than asserting a root cause from history alone.
• Next recipe: One recipe recommended inside objectives, safety boundaries, and observed coverage, with candidate process settings, prediction intervals, risks, and rationale.
• Candidate-coverage design (DOE): A statistical design method for understanding variable coverage and interaction; Ingot's current product loop admits only real runs and their quality outcomes.
• Response surface: A surrogate model that approximates the relationship between targets and parameters with an interpretable function, suited to limited samples.
• Bayesian optimization: A sequential optimization method for costly, stepwise decisions; it requires few variables, a quantifiable objective, and clear safety boundaries.
• Surrogate model: A statistical model, such as a Gaussian process, that approximates the real process response from limited samples.
• Mechanism knowledge: Reviewable knowledge from physical, chemical, or engineering principles used to constrain or explain recommendations without replacing real-run evidence.

Deployment and responsibilities

• Site: A business scope with isolation boundaries; data, permissions, and tokens are assigned by site.
• Edge ingestion token: A separate, rotatable credential that a field connector uses to send events, configured independently per field.
• Engineering decision: Recommendations are never dispatched automatically; engineers define objectives and boundaries and review, adopt, or reject each recommendation.