Experiment 05 · Feedback to plans
After the review,
what next?
Keep the mechanism explicit. Let feedback change the work. Generate a response that uses only what is known.
Some plans succeed if the outcome is kind. A policy must choose a next step for every outcome it promises to handle.
Explore the feedback loop
Runs in this browserChanging supply leaves the task rules intact. The example assumes the specialist package can deliver an accepted replacement.
Review or edit the mechanism and goal
Planned actions and uncontrolled outcomes are separate. Each step has explicit token inputs, outputs and duration. Editing these rules creates a different model; a review outcome follows the current rules.
The result will use this model.
Preparing the example…
Responses to the same goal
The horizon limits this question to at most eight events. It does not remove an adverse outcome from the mechanism. An exhausted computing budget means unknown.
Knowing each answer in advance is a different problem
These separate calculations fix the permit before planning. The actual policy above must decide at the point the mechanism makes that information available.
Current checkpoint
Observed event history
Current token counts and remaining events
This is a retrospective execution with its outcomes already specified. Selecting a successful path here does not let the online planner choose those outcomes.
Work structure
What this witness checks
Earliest schedule for this fixed execution
This can overlap independent work. It is separate from the policy’s sequential worst-case time. On a small screen, focus or scroll the chart sideways.
Follow the typed supplies and retained context
Expanded monoidal term and replay record
One choice at each observed state
A decision uses the current marking and remaining event allowance. Every enabled review outcome must be covered. No decision can depend on a later outcome.
Complete analysis counters and limits
Keep the model and observed checkpoint
Save in this browser, or download a portable copy. Reopening reruns the calculation and replays the observed events; imported claims are never accepted as proof.
No copy saved in this session.
Say which system and value you mean
A larger boundary can make a stimulus internal. It does not make that outcome controllable, change the token rules or prove uninterrupted service.
What the added sources changed
The lifecycle framework asks for a boundary, stimulus, response and value. The design-process sources ask what becomes stale and what must be repeated. Together they make a feedback experiment more useful here than starting with a reusable module library.
This construction uses ordinary token rules and finite observed-state planning. It adapts selected source methods; it does not implement the full ASM2.0 simulator or claim a general categorical theory of feedback.
Read the methods, source use and precise limits → · Build and independent challenge →