PDDL → Explainable WBS (Interactive) — Emergency Shelter “halfway up” Gimmer Crag, Langdale

The model is a small STRIPS/PDDL-style planning problem: a target end-state (goal literals) + a catalogue of actions (preconditions/effects). Clicking Generate plan & WBS runs a lightweight best‑first planner, then compiles an explainable Work Breakdown Structure by goal decomposition + causal links. This stays high-level (no technical rigging/installation instructions); treat it as a planning/specification aid and consult qualified professionals for real-world execution.
About this workbench
Choose a button to view details.
Scenario
Think of literals as “auditably true statements.” Toggle assumptions; the planner can find a plan, exhaust the finite model with no plan, or stop at its search limit with reachability still unknown.
For cliff/rock contexts: professional oversight, rescue readiness, and legal/landowner constraints are non‑negotiable in practice; the toggles here are epistemic switches for modelling.
Initial state & goals
Init (facts already true) —
Goal (end-state) —
Status: idle
Planning model (PDDL-style)
The UI edits the problem (:init and :goal); the domain is fixed but readable/exportable.
Domain (actions)
Problem (init & goal)
How the WBS is derived (explainability rule)
  • End-state literals are clustered into deliverables (work packages) like “Access & Rescue Readiness.”
  • As a grouping heuristic, each plan action is assigned to the first deliverable whose target literals it contributes to (direct effects), and is displayed with its causal justification. Reordering deliverables can change the grouping; the plan does not determine a unique WBS.
  • Within each deliverable, actions are ordered as in the plan, giving a traceable “because → therefore” chain.
This is deliberately “explainable by construction”: every work package is grounded in explicit predicates, not informal phase labels.
Outputs
WBS (tree)
Plan (actions)
Explain (causal links)
PNML → Tasks → Gantt
Tip: click the squares to collapse/expand. Each node has a short “why” grounded in literals the planner used.
Safety & realism notes (non-optional)
This planner treats “actions” as abstract commitments; it cannot capture real cliffside uncertainty (rock quality, anchor integrity, wind loading, human factors). In practice you’d add (at least) probabilistic risk models, verification gates, professional sign-off, and explicit constraints (e.g., “no single-point failures”).
If you use the model, use it for documentation and explainability—not for operational instructions. Treat every “access/anchor/rig” step as requiring qualified specialists and local authority guidance.