Ingot
Menu

Project introduction

Ingot documentationGetting startedCurrent status

Integration and operations

Recipe-Optimization Pilot GuideData integrationDeploymentFrequently asked questions

System and algorithms

System designAnalysis and OptimizationMechanism Knowledge Design

Validation and production

Scenario Evaluation BoundaryProduction architecture

Project governance

RoadmapBrand guideOpen-source dependenciesGlossary

Recipe-Optimization Pilot Guide

Document status: current operating guide. This guide validates the implemented production-evidence recommendation workflow. Next-recipe recommendations require admitted production-run evidence; each run is an experiment, with actual parameters and quality outcomes retained in the run-evidence chain. See Current status for the full capability boundary.

Pilot Scope

Limit the pilot to one product or process scope and one quality objective. Provide real recipe runs, actual settings, process context, and quality outcomes for the current production-evidence workflow; data may come from manual operations or connectors configured by the deployer. First verify that identity, units, provenance, and quality review are reliable.

Operating Sequence

  1. In the recipe-recommendation brief, select the site and published recipe, then define product scope, quality objectives, adjustable parameters, and safety boundaries.
  2. Let completed real recipe runs in scope enter admission checks automatically.
  3. Generate one next-recipe recommendation after at least three valid runs cover two distinct actual recipes.
  4. Record engineer adoption, modification, or rejection, with the reason and final actual recipe.
  5. Link the later real production run to that decision; a rejection requires no invented execution.
  6. Freeze the quality outcome from source data after parameter readback and inspection facts are complete.
  7. Let new real runs and outcomes return independently as the next observations and recommendations.

Check Every Recommendation

  • Are input runs, quality outcomes, and context traceable?
  • Does the recommendation stay within declared safety boundaries and observed coverage?
  • Does the engineer decision include adoption, modification, or rejection and a reason?
  • Does an adoption or modification link to an actual execution?
  • Is the outcome frozen only after parameter readback and inspection facts are complete?

Insufficient Samples or Bad Data

When samples are insufficient, continue normal production and wait for new real runs, or repair the data chain. Missing data, incomparable runs, conflicting constraints, or unmet model conditions must not be bypassed to generate a recommendation.

Completion Criterion

A pilot completes at least one auditable “recommendation → decision → actual run → quality outcome” chain. It demonstrates the system flow and the scenario's data chain; it does not establish a general benefit or causal conclusion.