What the foray is seeking
The rotunda foray seeks states from which valid paths through a changing project can be constructed. The constructive-state essai retained every labelled continuation, and its open-interface continuation found no state merging for its small separate fronts. This trial changes the obligation. Can a dispatcher select one quickest next move from every reachable state, including a correct stop at completion? A dispatcher can succeed at this while being unable to predict the consequences of a different move. The collection’s curation record distinguishes this finite, resource-constrained construction from the separate studies of circular observation and geometry. The earlier working dossier posed a general project-state construction and asked how one might test it on a real workflow. The crew rule advances its discrete exemplar. The separate circle-and-slot studies and the proposed transfer to real workflows require their own evidence.
The trial
Use the existing four-brick, two-course model on each of two fronts. Placement takes two ticks. A completed course cures for two ticks at A and three at B. The pool supplies one or two hoists. From the detailed reachable graph compute minimum remaining ticks by reverse breadth-first search. Each state contributes its set of actions that reduce that distance by exactly one. A finished state contributes only STOP.
Group concrete states by what the dispatcher sees. A memoryless quickest rule exists for the whole group exactly when the intersection of those action sets is nonempty. Intersect all sets in the group. Pairwise overlap alone would be insufficient. Choose a common move and check it in every member. Since it reduces an integer distance, repeating it reaches the goal in exactly the minimum remaining time.
| Wiring | Reachable states | Classes retaining all labelled behaviour from the blueprint refinement | Distinct legal-action menus | Menus without a common quickest move | Ticks from empty |
|---|---|---|---|---|---|
| One hoist | 3,883 | 3,508 | 530 | 0 | 34 |
| Two hoists | 6,399 | 6,009 | 530 | 0 | 22 |
The legal menu gives the exact labelled choices currently available, including joint moves and wait. It does not expose the curing clocks or placed masks directly. The 530 menus differ between wirings. Pool both graphs and their union contains 1,014 menus. A single menu-indexed rule works everywhere in that union without a hoist-count label. Replays from empty reach 34 and 22 ticks. All 729 blueprint-only groups fail the quickest-move test in each wiring.
Break the result
Keep one hoist and swap the two curing times. In both worlds each front has placed ground-course bricks 1, 2 and 3, no front is busy and no curing is active. The menu is identical. With A curing in two ticks and B in three, only placing A's fourth brick now is quickest. Reverse the curing times and only placing B's fourth brick is quickest. Both detailed states have 22 ticks remaining on an optimal path.
Across the two calibrations, 32 of the 530 menu groups have no common quickest move. A change in capacity and a change in process law have sharply different consequences here. A dashboard designer could ask whether the legal menu is available reliably, and who tells the dispatcher when a governing process duration changes.
A dispatch instruction the crews can use
The 1,014-row lookup can be replaced by one instruction. If both fronts can start together, grant both requests. If just one can start, grant it. With no requests, wait until a crew is ready or the project reports completion. When both request one hoist, give it to the faster-curing front if its request is for the ground course. Otherwise give it to the slower-curing front. Each crew can choose any legal brick on the front it receives. The reader’s Follow crew rule button takes the first such labelled choice.
This delegates brick selection to the crews. The dispatcher needs a ready request from each front, whether both starts fit, whether the faster-curing front requests its ground course under contention, and a completion signal. Across the two capacities and both assignments of two- and three-tick curing, every one of the 30,240 permitted next moves at the 20,564 reachable state/configuration pairs preserves shortest completion. These counts include waits and exclude terminal stops. Terminal states were checked separately.
Can the course report be dropped? Hold the calibration fixed at A = 2, B = 3 with one hoist. At the original fork only A is quickest. Now suppose A has laid its first upper brick, while B has not started its ground course. Both request the hoist again but only B is quickest. Readiness alone cannot distinguish these decisions. The one course distinction suffices in this reporting arrangement. The calculation supplies the grant rule, while the crews or a separate system must still establish which local starts are legal.
There is also a limit to the instruction itself. Shorten placement from two ticks to one while retaining one hoist and the original curing times. The instruction then fails at 16 of 1,015 reachable states. For example, A has laid one ground brick and B has laid three. The faster-curing front A is requesting its ground course, yet granting B is the only quickest choice. This is a counterexample to extending this briefing by the relative order of curing times alone.
What was bought by weakening the question
This is an exact finite policy-synthesis result for a fixed objective, labelled actions, two capacities and stipulated process laws. Generating an accurate legal menu itself requires information or local judgement. The calculation supplies neither a sensing design nor a guarantee across arbitrary wiring, curing or priorities. The previously published behavioural classes answer a stronger preservation question and start with the placed-brick readout, so these counts do not establish a universal compression ratio.
Myers's readout and update distinction specifies what the dispatcher may observe and do. Shortest-path search and the common-action intersection supply the additional decision test. The exact model already exposes the required input fibres; fitting a continuous equation to the drawn rings would not answer this question.
Constructive state and policy · Earlier open-interface result · Myers on open dynamical systems
Reproduce
Run node research/dispatch-readout/run.cjs from the project root. Add --policy to emit the common 1,014-row dispatcher. The results include sensor comparisons, replays and a counterexample with both detailed states.