Change a task network and inspect the consequences

The earlier name was “WBS Graph Rewriter.” The implemented model is a finish-to-start precedence network. A WBS describes a deliverable hierarchy; the network here describes order and time.

1. Select a task in the graph or dropdown; edit its fields and apply the change.
2. Add tasks, gates or buffers; choose source and target to add a dependency. Select an edge to remove or rewrite it.
3. Compare earliest and resource-feasible timing, inspect total float, and read the diagnostic.
4. Save a named scenario before changing assumptions. Compare or restore it, undo/redo edits, or export a JSON bundle.

Tasks have integer durations and named resource demands. Gates and milestones can have zero duration. Buffers consume time with no resources. There are no calendars, deadlines, lags, costs or hidden technical constraints. The GAO Schedule Assessment Guide provides wider scheduling context.

• R1 · Collapse serial steps: A must have exactly one successor B, and B exactly one predecessor A. The new package retains both steps and their resource profiles, but assumes they run contiguously without a resource-waiting gap. This extra assumption can change the levelled finish. Expand it to edit internal steps.
• R2 · Add a finish buffer: choose one deterministic zero-float route, sum its supplied variances and use ceil(k × √sum). This is an illustrative independent-variance heuristic, not a critical-chain method or a confidence guarantee. A shared buffer follows every current terminal task.
• R3 · Gate note: append a hypothetical note to a zero-duration gate with one successor. This changes no approval, governance authority or dates.
• R4 · Remove a dependency assumption: allowed here only for disjoint resource profiles with no noOverlap flag. That check does not prove the link technically dispensable; explicitly acknowledge the hypothesis before applying it.

Changes validate the graph before acceptance. Structural validity does not establish real-world suitability.

A forward/backward pass finds earliest/latest dates and total float using precedence only. Zero float marks critical tasks, including ties. Resource leveling picks from tasks whose predecessors are already scheduled, then finds an available interval. Earliest-first and low-float-first are deterministic heuristics; neither is an optimality claim. Impossible demands produce an error, never a forced “feasible” placement.

For merged packages, resource demands vary across the preserved internal steps. The inspector's resource field shows peak demand for reference. The risk experiment separately samples each original step from an independent triangular distribution [0.7d, d, 1.3d]; it ignores resources and the editable variance field. That field is used only by R2. These are invented assumptions.

Drag a node to rearrange it; use zoom buttons or Ctrl/⌘ + mouse wheel to inspect the graph. Every task and edge is also available through labelled selectors. Use Undo/Redo for plan edits, or save and restore named scenarios. Export includes the plan, scenarios and log; nothing is automatically persisted in the browser.