Learning engineering flower · Dynamic conversations
Dynamic flower experiments
About this resource · How to use it · Learning engineering connections
What is this resource?
A configurable two-phase tool: dynamic Design / Develop conversations, then Deploy / Test on a frozen learning resource under easy, medium and hard conditions. The public editor changes and exports configurations and reads retained records; model execution uses the local runner.
How should I use it?
- Choose a modest case and team condition. Edit the case brief, specialist or generalist knowledge, scenario conditions, and any user change; export the configuration.
- Run the exported configuration with the local runner, freeze an artifact version and record the three deployment tests. Import the run to inspect questions, replies, objections and decisions.
- Read expert process and result assessments separately. An update produces a new version and another test set; preserve missing evidence and earlier results.
How does this connect to LENS and learning engineering?
The flower organizes complementary expertise around shared understanding. The tool makes that understanding inspectable through actual exchanges and changes to a solution. Human flourishing, Learning sciences, and Systems engineering remain the LENS foundations. Role permissions do not erase model knowledge, and agentic judgments do not establish human learning outcomes.
Give the team a problem, bounded expertise and room to respond to one another. Develop a solution, freeze a version, and test it under easy, medium and hard conditions. An expert evaluator examines the conversation and the result separately.
The conversations are among the agents designing the learning resource. Deployment tests use repeatable scripted browser actions to observe the resulting tool; they do not simulate a learner’s reasoning or measure human learning.
Questions, objections, decisions and implementation.
Frozen versions, scenario evidence and expert review.
Study progress
Loading the latest published progress snapshot…
New dynamic pilot: recorded evidence
Two modest cases, each explored by a specialist team and a generalist: four independent design conversations, one per case and condition. Each frozen design receives easy, medium and hard tests. These repeated scenarios are not additional teams or learners.
Evidence snapshot: . Open a conversation for the complete messages, version history, observed output and quoted reasons behind each rating.
| Case and condition | Accepted messages | Contributing actors | Frozen version | Test records | Expert reviews |
|---|---|---|---|---|---|
| Handoff · Generalist | 3 | 2 | v1 | 3 / 3 | 4 / 4 |
| Handoff · Specialist team | 17 | 7 | v2 | 3 / 3 | 4 / 4 |
| Workshop · Generalist | 3 | 2 | v1 | 3 / 3 | 4 / 4 |
| Workshop · Specialist team | 21 | 6 | v3 | 3 / 3 | 4 / 4 |
Message and actor counts describe the accepted conversation through its first freeze, including the facilitator. For an unfrozen candidate they show progress so far. Failed attempts remain in the record. Longer conversations and more approvals are not measures of solution quality.
Inspect the expert process ratings
Each frozen version is scheduled for one process review and three scenario-result reviews, each from a fresh model evaluator context. Scores range from 0 to 4; unknown evidence remains unknown. The evaluator sees the assigned roles and conversation. These are descriptive judgments, with no combined score or condition ranking.
| Criterion | Handoff Generalist | Handoff Specialist team | Workshop Generalist | Workshop Specialist team |
|---|---|---|---|---|
| Human concerns influence decisionsHuman flourishing support | 3 | 3 | 3 | 4 |
| Expertise boundaries and uncertaintySystem performance | 4 | 4 | 4 | 4 |
| Questions and objections receive substantive responsesLearning | 2 | 4 | 3 | 4 |
| Decisions and implementation are traceableSystem performance | 3 | 4 | 3 | 4 |
| Learning mechanisms and evidence are reasoned aboutLearning | 3 | 4 | 2 | 4 |
| Relevant perspectives and user updates are consideredHuman flourishing support | 2 | 3 | 2 | 3 |
Handoff · User update · v3
- Human concerns influence decisions
- 4
- Expertise boundaries and uncertainty
- 4
- Questions and objections receive substantive responses
- 4
- Decisions and implementation are traceable
- 4
- Learning mechanisms and evidence are reasoned about
- 4
- Relevant perspectives and user updates are considered
- 3
Inspect the expert result ratings by scenario
These scores concern the actual frozen artifact and observed scenario. Passing browser checks does not establish useful instructional feedback, human learning, or workplace effectiveness. The update uses the same fixtures for regression checking; it is not a fresh holdout test.
| Case, condition and version | Scenario | Agency and accessHuman flourishing support | Task behaviorSystem performance | RecoverySystem performance | Capability supportLearning | Feedback and revisionLearning | Credible useHuman flourishing support |
|---|---|---|---|---|---|---|---|
| Handoff · GeneralistInitial · v1 | easy | 3 | 3 | 3 | 3 | 3 | 3 |
| Handoff · GeneralistInitial · v1 | medium | 3 | 3 | 3 | 3 | 2 | 3 |
| Handoff · GeneralistInitial · v1 | hard | 3 | 3 | 3 | 3 | 2 | 3 |
| Handoff · Specialist teamInitial · v2 | easy | 3 | 4 | 2 | 3 | 3 | 3 |
| Handoff · Specialist teamInitial · v2 | medium | 3 | 3 | 2 | 3 | 3 | 3 |
| Handoff · Specialist teamInitial · v2 | hard | 3 | 3 | 2 | 3 | 3 | 3 |
| Handoff · Specialist teamUser update · v3 | easy | 3 | 3 | 2 | 3 | 3 | 2 |
| Handoff · Specialist teamUser update · v3 | medium | Pending | Pending | Pending | Pending | Pending | Pending |
| Handoff · Specialist teamUser update · v3 | hard | Pending | Pending | Pending | Pending | Pending | Pending |
| Workshop · GeneralistInitial · v1 | easy | 3 | 2 | 3 | 2 | 2 | 3 |
| Workshop · GeneralistInitial · v1 | medium | 3 | 2 | 3 | 2 | 2 | 3 |
| Workshop · GeneralistInitial · v1 | hard | 3 | 3 | 3 | 2 | 2 | 3 |
| Workshop · Specialist teamInitial · v3 | easy | 3 | 3 | 3 | 3 | 2 | 3 |
| Workshop · Specialist teamInitial · v3 | medium | 3 | 3 | 3 | 3 | 2 | 3 |
| Workshop · Specialist teamInitial · v3 | hard | 3 | 3 | 3 | 3 | 2 | 3 |
Rating anchors
- 0
- Observed evidence contradicts the criterion or the relevant mechanism is absent.
- 1
- A partial or fragile response with material observed gaps.
- 2
- Adequate evidence for a limited portion of the criterion, with clear limitations.
- 3
- Strong, coherent evidence across the relevant recorded conditions.
- 4
- Consistent, well-supported evidence with appropriate handling of exceptions and uncertainty.
- Unknown
- Evidence is missing, unexecuted, invalid or insufficient to support a rating.
Planned user-change demonstration: Version 3 is frozen; 3/3 regression test records and 2/4 expert reviews are retained. Open the handoff specialist conversation for the accepted evidence.
Separate direct feature check: The browser inspection observed complete keyboard copying, preservation of the last submitted record during draft edits, replacement after resubmission, and clearing on valid scenario changes, skip and reset. The two example summaries contained 2,960 and 3,072 characters. Human judgments of concision and ease of use remain unmeasured. This additional check is separate from the scenario fixtures and model ratings. Inspect the retained feature-check observations.
The run records identify the manual response transport and retained model output. Expertise boundaries are role instructions; they do not erase a model’s prior knowledge. This small pilot illustrates a workflow and exposes questions to test with people.
Read the dynamic study paper · Download this evidence summary · Read the protocol and recorded method amendments
Loading the configuration editor…
The configuration content hash identifies exact run inputs; its ID is your editable version label.
Start with a small case
Bounded knowledge
Specialists receive their assigned domains and brief. The generalist keeps a separate broad, introductory brief. These are prompt boundaries, not measured professional competence.
Deployment scenarios
Copied scenarios travel with the configuration. Their test facts are withheld from Design / Develop.
Introduce a user change
Queue a new constraint or request after a specified turn. Export the changed configuration for the local runner.
Bounds and handoff
Run this exported version locally
From a checkout of the repository, save the exported JSON and create a new run. The manual commands below prepare a prompt and accept a response file. Provider execution is an explicit local-runner choice.
After Design / Develop, use freeze to pin an artifact version. Record browser evidence with observe, then run separate process and scenario evaluations with evaluate. Consult the runner’s --help for arguments.
Exact configuration JSON
Form edits update this runnable JSON. Apply JSON edits before saving or exporting.
No run imported
Import the runner’s run.json to inspect messages, exact artifact versions, unresolved issues, and expert evidence. No demonstration scores or research results are generated by this page.
Questions, objections and decisions
Artifact versions
Deployment observations
These are retained browser-runner observations for an exact frozen version and scenario. A declared check is a narrow test of the interface; it does not establish learning.
Expert evaluation
Process and result remain separate. Criterion scores describe recorded expert judgments; they do not establish human learning. Unknown evidence stays unknown.