Domain 1
Learning Sciences
Learning mechanisms, prior knowledge, practice, retention, and transfer.
Agentic learning engineering · Team framework
A guide to the learning engineering flower and its eight starting expertise domains, shared learner and stakeholder responsibilities, and controlled team comparisons. Earlier role contracts remain available for interpreting historical runs.
Shared understanding connects domain contributions into a coherent design. Team composition may change with the challenge and stage. Flower domains organize expertise; they are not validated individual competency scores or a replacement for involving learners and stakeholders. LENS foundations remain Human flourishing, Learning sciences, and Systems engineering.
Learning engineering brings complementary expertise together around a shared challenge. The flower places shared understanding at the center and connects the whole team with learners and other stakeholders.
The default simulation assigns one member to each domain and a coordinator. A particular challenge may need a different mix. One person can contribute several domains, and several people can share a domain. These editable instructions organize contributions; they do not establish anyone’s professional competence.
Domain 1
Learning mechanisms, prior knowledge, practice, retention, and transfer.
Domain 2
Learner journeys, activity structure, feedback, and revision.
Domain 3
Data quality, analysis, uncertainty, and model assumptions.
Domain 4
Constructs, measures, comparisons, and limits of inference.
Domain 5
Operating environment, dependencies, access, and recovery.
Domain 6
Executable behavior, state management, testing, and maintenance.
Domain 7
Content accuracy, authentic tasks, and consequential errors.
Domain 8
Facilitation, learner support, adoption, and delivery conditions.
Source: Ali and colleagues, Learning engineering is a team sport (2025). The flower domains are starting points and are not conclusive. Composition may shift with the challenge and stage.
Distinguish a missing responsibility from losing a member’s time and contribution opportunities. Fix the case facts, challenge schedule, execution settings, and coordination policy, then inspect what each contrast actually changes.
| Condition | Change | Capacity and interpretation |
|---|---|---|
| Full team | Retain the configured specialist assignments and available members. | Route questions and objections dynamically within the common conversation budget. |
| Domain omitted everywhere | Withdraw the selected domain’s explicit instructions and approval authority from every member. | Retain identities and make the expertise gap visible. |
| Domain withdrawn from one member | Withdraw that domain from one selected member; other assignments remain. | Keep this member available to ask questions; assigned overlap may compensate. |
| Member removed | Remove the selected member and their available contributions. | Retain any declared overlap held by other members; report actual participation. |
| Broad but shallow generalist | One generalist uses an introductory cross-domain brief and can make routine proposals and implementations. | No specialist playbooks are injected into later turns; no specialist signoff is implied. |
The pilot exercised the full specialist team and broad generalist configurations across two cases. The other three configurations remain available for future experiments.
The revised tool uses dynamic Design / Develop conversations followed by Deploy / Test scenarios. Questions, replies, objections, user updates and unresolved issues determine the next contribution. A bounded conversation preserves its full accepted transcript. The earlier fixed-workflow study used forty scheduled calls, or thirty-six after member removal; those counts do not describe the revised protocol.
Overlap may allow another member to identify a concern. Omission does not erase a model’s latent knowledge, and no condition is programmed to fail. Matching calls also does not match token volume, independent perspectives, or human effort.
The current protocol freezes an implemented design before testing. A separate evaluator receives the assigned roles, complete accepted design conversation, frozen source and relevant scenario observations. The review is not blinded. A proposed repair becomes new evidence only after implementation, a new freeze and new tests.
The revised expert evaluation reads the complete design conversation, scenario conditions and observed browser evidence. It scores the process once per artifact version and the result separately for easy, medium and hard conditions. The preserved earlier study used a different fixed workflow and cannot validate the revised design. Neither agent judgments nor scripted browser observations establish human learning.
Inspect the dynamic conversations, tests and prototypes · Read Exploring the learning engineering flower with agentic simulation · Read the execution and evaluation documentation
These profiles document the earlier replay collection. Their original duties and requested LENS targets remain attached to those saved runs. They are not a crosswalk from the flower domains to a validated competency framework.
Earlier roster
Integrate learning sciences, human-centered design, systems requirements and evidence. Preserve dissent and choose the earliest stage to revisit.
Earlier roster
Observable capability, descriptive feedback, retention, deliberate practice and transfer. Completion is not competence.
Earlier roster
Human-system coupling, constraints, requirements, recovery and non-training levers.
Earlier roster
Affected people’s voice, accessibility, workload, participation and agency.
Earlier roster
Task evidence, comparison, transfer, causal threats, subgroup coverage and unknowns.
Earlier roster
Delegation, privacy, surveillance, correction and accountable human decisions.
Earlier roster
Small runnable artifact, technical limits, instrumentation and fallback.
Earlier roster
Use supplied domain context; state limits of expertise and required practitioner validation.
Earlier roster
Introduce a plausible hypothetical objection or unexpected event that challenges a named assumption and requests a requirement change.
Registry 1.2.0; framework 1.1.0. Requested targets have no demonstrated proficiency evidence by themselves.
Inspect the original resolved role contract · Open a historical replay