Engineering Concept Registry

This registry does not assume prior knowledge of payroll systems, GIS, MBSE, or agent architecture. Start with a case: understand the task, failure, and delivered result, then open a concept page.

Case-coined records language formed in a specific implementation. ADPS restatement relates the mechanism to established software-engineering ideas. Candidate concept still needs evidence from independent systems.

Dongfang Yiteng · Long-Running Execution Agent

A payroll configuration task crosses several dependent steps. The runtime must preserve the original goal, progress, parameter sources, and approval state. Read the case report

• Control/Narrative Planes — Separate discrete program control from the language context used for reasoning.
• Execution vs Content — Classify the delivery and its failure cost before choosing the runtime.
• Three State Planes — Keep narrative, mechanical values, and task scheduling in separate state planes.
• Mechanical State Plane — Bind business parameters to auditable sources instead of regenerating them through a model.
• Anchor / Ledger / Collection — Keep the original goal fixed, append milestones, and project only the facts needed now.
• Scratchpad Board — Store the current goal, step, observation, blocker, evidence, and next action in a small working surface.
• Memory Envelope — Assemble only the goal, progress, intent, evidence, and recalled experience needed by the current step.
• Layered Memory L1/L2/L3 — Separate current input, task facts, and cross-task experience, with references back to source records.
• Orchestrator and Facade — Keep control flow in the orchestrator and input/output presentation in the facade.
• Intent as Compilation — Compile open language into a bounded control signal with an explicit fallback.
• Task DAG and State Machine — Encode dependencies in a graph and lifecycle transitions in a state machine.
• Activity Events and Timeline — Record model, tool, state, latency, and cost events on one execution timeline.
• HITL Block and Resume — Pause a sensitive node, preserve its state, revalidate, and resume after approval.

Xuanxu Technology · GIS Publishing Agent

A map dataset passes through processing, publication, and user-side verification. An internal success log does not prove that a user can see the map. Read the case report (Chinese)

• Reasoning Assetization — Move repeated, reviewable decisions into versioned capability assets before runtime.
• Disk Fact Plane — Keep stage facts, progress, and acceptance evidence in recoverable external contracts.
• External Acceptance Probe — Test the result from the user side and let that evidence decide success.
• Failure-to-Rule Loop — Require every repeated failure to change a rule, test, template, or capability asset.
• Capability Certification — Grant automatic execution rights only to a tested capability version and scope.

AI4MBSE · Modeling Agent Project

A natural-language request will change a shared engineering model. Before write-back, the system must resolve the objects, relations, scope, and acceptance conditions. Read the research report

• Typed IR Chain — Narrow natural language through typed, verifiable structures before write-back.
• One-Diagram Unit of Work — Commit one diagram, one resolved scope, one write set, and one receipt at a time.
• Structural Write-Back Gate — Check scope, types, references, completeness, and real change before side effects.
• Resumable Clarification Job — Pause the same durable job when one required field is missing, then resume it.
• Vocabulary Equivalence Layer — Align planning terms, domain types, and the host metamodel through one tested mapping.