The schedule engine in wall-framing-takeoff takes a builder's own activity list (durations, predecessors, cure-time lags, crews, costs, lender draw terms) and returns the critical path, the finish date with and without crew limits, and a weekly and monthly cash flow. Hand-checked networks tested it; a hand-written network only covers the shapes its author thought of.
So it now has a seeded property test, schedule-fuzz.test.mjs. It builds 300 random networks of 2 to 15 activities (a directed acyclic graph by construction), with random lags, one or two crews of one or two heads, random costs, holidays and draw terms, and checks what must always hold:
- every activity starts after each predecessor finishes plus its lag, both in the unlimited-crew plan and after crew levelling;
- slack is never negative, and an activity is critical exactly when its slack is zero;
- levelling never finishes sooner than the unlimited-crew plan;
- no crew is over its head count on any working day;
- cash-flow weeks and months both add up to the typed cost total;
- a network closed into a loop is always refused with a reason.
To check the test has teeth, we broke the levelling pass to ignore lag. The test failed. With the engine restored, all 300 networks pass.
The limit: these are invariants over invented networks. No real builder's activity list or draw schedule has been run through it yet, and that is still the open item on the roadmap.