WHAT THE TRIAL ESTABLISHES
One recovery project.
Different reasoning contracts.
Each method reveals, derives or rules out something specific. A common scenario makes their differences easier to inspect.
Read the results at the right scale
The fictional recovery is inspired by the collapsed footbridge at Gimårasten on the Gimån, near Stavre. Its two-section construction, temporary supports, methods, quantities and costs are invented. Nothing here diagnoses the actual collapse, establishes its cause or proposes engineering work. Per Sonnvik’s walk account, dated 11 May 2024, provides background about the crossing; no construction assumptions below are inferred from it.
The new illustration is an imagined autumn scene generated with OpenAI, using the locality’s broad river, low wooded banks, long broken crossing and adjacent railway as cues. It is neither a photograph nor a reconstruction of the actual structure. Sonnvik’s photographs remain his copyright and are linked rather than reproduced. For another view of the river, Andreaze’s 2008 Gimån photograph is released into the public domain; its precise reach is not established here.
Initial boundary: the crossing is closed, damaged material has been made safe and removed, and the supplied condition evidence supports the declared abutment and installation assumptions. Placement permission is a separate prerequisite. Low-water access and lifting capacity are shared controls. The optional decision survey concerns extra logistical readiness; it cannot satisfy missing structural-condition evidence or permission.
The examples do not form one integrated engineering model. Logical moves, abstract schedule time, calendar windows, full-path budget and expected decision costs have different meanings. Curing is explicit in the state model, aggregated into placement duration in the scheduler, and omitted from the small semantic action catalogue. No figure is a real estimate or verified construction limit.
Model correctness, research contribution, reader understanding and usefulness in practice are separate questions. Automated checks do not establish human learning or bridge safety. “Best for” judgements on the comparison page refer to evidenced uses, not a universal numerical ranking.
110 · Process to plan
Seven explicit facts feed a projection. Five constraints are retained; two narrative facts are omitted with recorded reasons. Two process descriptions yield the same four-part summary, but only one contains bank-based placement when the low-water work window disappears. Permission and condition evidence remain separate supported or missing claims.
The counterexample demonstrates information loss under a projection. It is not proof of functoriality or an inverse reconstruction algorithm. Having a plan does not grant permission.
120 · Dynamic project states
The state records temporary supports, two deck-section stages (absent, curing, ready, accepted), clearance and reopening. Explicit update rules generate valid moves; breadth-first exploration finds a shortest completing continuation. At the default start, one lifting resource needs nine logical moves; two allow a joint placement and need eight. These are not durations. Reopening follows acceptance of both sections and removal of supports.
Two states can both show “two sections installed” while only one allows inspection. State sufficiency belongs to the declared update law. Missing access blocks the selected low-water method; changing to bank-based placement changes the mechanism. This is finite dynamics, not a sheaf-gluing result.
140 · Local to global
Three accounts constrain month and method. The access review also records independent permission and condition-evidence domains. The solver intersects every relevant domain and enumerates joint assignments. The pairwise trap uses {July, August}, {August, September} and {July, September}: every pair intersects, but all three do not.
These finite constraint sets are not declared sheaf sections with a gluing theorem. Missing evidence is not agreement; fixing a date does not repair missing permission. A real review must also distinguish units, versions, scope and authority.
150 · Mechanisms to possible plans
Nine once-only transitions model supports, preparation of A/B, placement of A/B, inspection of A/B, support removal and reopening. All 52 complete atomic execution orders are enumerated. Placement reads permission, condition evidence and method access; a lifting resource is reusable. Acceptance records are retained through support removal and reopening: the signatures carry them forward as evidence, and the scheduler reads completed prerequisites.
A separate scheduler reserves lifting capacity throughout placement and credits completion at its end. Its placement duration includes the toy settling/curing allowance. With one resource there are two earliest-start schedules finishing in nine abstract units; with two resources there is one finishing in seven. These priority schedules do not enumerate deliberate waits. Independent replay checks durations, precedence, coverage and aggregate capacity.
Typed signatures do not implement a general Petri-to-SMC functor. WBS grouping is separate from causal order and does not recover every dependency.
160 · Process wiring
The diagram checks four nominal interfaces. A mounted section cannot satisfy an accepted-section input. A separate simultaneous-use contract adds lifting demand and checks capacity, both evidence prerequisites and low-water access. Repairing a type mismatch need not repair a resource failure.
This is a small composition check, not a full wiring-diagram operad. Matching a port does not certify structural or hydraulic performance. The handover output means the two declared acceptance interfaces match; the full clearance and reopening process appears in the other examples.
Composing work packages (the broader native investigation) · Comparison implementation
170 · Shared constraints and review
The model enumerates two placements in August/September under accepted access, evidence and capacity. Each named month represents one indivisible placement slot; sequential placements within a month are not modelled. Two placements in the same slot therefore demand two lifting resources. With one resource and both months, two opposite orders remain. Restricting access to one month removes them unless capacity changes.
A reviewer proposes an access change; accepting it updates only this local demonstration and its reproducible URL. Proposed and accepted values remain distinct. This is neither live collaboration nor approval in a project system. Calendar assignments are not physical differential dynamics.
Composing open systems · Finite shared-constraint adaptation. The native investigation has a broader open-system scope than this two-window example.
180 · Feasible decks and their construction paths
Three fictional deck configurations combine with two placement methods. Feasibility checks a supplied load-rating floor of 10, lifting-load compatibility, access, both evidence prerequisites, capacity, temporary supports and full-path budget. Rating and mass units are illustrative attributes; no structure is calculated.
The one-piece deck with low-water placement costs 68 units before temporary works and 78 with them, exceeding the default budget of 70. Two standard sections remain feasible at 64 units with low-water placement or 60 with bank-based placement. The first has shorter duration and site occupation. Light sections fail the rating floor even when cheap. The nondominated set therefore expresses trade-offs without inventing a universal weighting.
These are six finite witnesses. They do not search the full design space or embed the separate co-design feedback solver. One-piece versus two-section configurations are local design alternatives; the other examples hold two standard sections fixed.
210 · Meaning, goals and required work
The desired end is accepted sections, clear temporary works and an open crossing. A local typed vocabulary compiles explicit method rules into planning actions. Setup destroys “clear”; removal restores it after acceptance; reopening requires clearance and acceptance. Empty recovery scope generates no work. A two-section method requires seven actions; one section requires five.
Necessity is tested by removing each function and searching again. It is necessity within this catalogue, not all possible repair techniques. Changing product/process packaging preserves action identities. The semantic route searches fact sets; a separately declared parameter-free STRIPS domain is parsed into an independent bit-mask search. Both receive the same initial facts and goal, and parity is computed.
This is not full RDF/OWL reasoning or evidence of superior planning. The supplied-WBS validator checks a declared single-parent acyclic hierarchy and an output for every leaf, with summaries exempt.
220 · Evidence before commitment
Three decision opportunities, a fixed hidden favourable/adverse site-readiness regime and one optional survey define a finite belief model. A controlled low-water placement attempt succeeds with conditional probability 0.9/0.2 and costs 10 units. Non-completion safely stops the attempt, consumes an opportunity and updates belief. Bank-based placement completes delivery for 27 units. Continuing closure through the horizon costs 50. These are invented logistics outcomes, not structural-failure probabilities or the whole recovery budget.
Required structural-condition evidence, permission and lifting capacity gate either method. The optional survey concerns additional logistical readiness; it cannot supply missing prerequisites. Dynamic programming compares low-water attempts, bank placement, waiting, continued closure and a survey. Waiting provides no new signal. The hidden regime does not evolve; river levels are not simulated.
Survey signals and observed non-completion update the posterior. Informative early evidence can change the chosen action; noise can cost money without changing belief; late evidence cannot help before the horizon. A 0.6 success-probability threshold policy uses the same costs and observation rules. Independent tests enumerate deterministic policy costs and compare their minimum with the calculated optimum. Optimality holds only inside this finite contract.
How to interpret the comparison
The shared disturbance removes a low-water work window. Some methods include a bank-based fallback; other rows explicitly select low-water placement. Their different outcomes expose declared scope and do not rank techniques by a common score. A stronger comparison asks the same question with controlled assumptions and a suitable simpler baseline.
The research programmes remain independent. Update a trial when its source method, assumptions or evidence changes; preserve useful counterexamples and interactions. Model tests, source-backed explanations and human usefulness remain separate evidence claims.
Return to the nine working examples → · Compare the wider collection →