Build a week. Break a rule. See what happens.

The hospital ward is 3 West, fourteen staff, in America/New_York: a location's timezone is the authority for its wall times, so the grid is showing you Eastern rather than yours. Registered nurses, licensed practical nurses and certified nursing assistants, plus the four people every real roster has and most demos leave out: a 0.6 FTE part-timer, a per-diem CNA, a float-pool RN, and an agency RN who bills at nearly twice an employee's rate. There is a second site, Riverside, with two more nurses on it, because the fill waterfall needs somewhere to escalate to.

The ward has three weeks rather than one. The June week is the baseline. The other two are the next two times the clocks actually change in that timezone, found by asking the IANA database when this page was built rather than by typing a date, so a 19:00–07:00 night on one of them is 13 real hours while the wall clock still says 12, and the payroll export gets the real one. That is the whole reason the engine reasons about instants and never about clock faces.

The clinic is a five-day family practice in Central time, where the limit that bites is a contract rather than a cap: Dana is on 24 hours a week against the practice's 40, and the engine checks the tighter of the two. The nursing home is twenty residents in Mountain time, with the census entered and HPRD computed live from the grid: nursing hours divided by residents, split at local midnight, which is arithmetic on the seeded numbers rather than a compliance threshold. Its PBJ panel builds a genuine CMS 4.10.0 submission file from those same hours, in your browser, and refuses to build one when something in it cannot honestly be filed.

The home care round is the one a shift scheduler cannot do at all. Its rows are clients rather than slots, and the seed carries the drive times between their homes. Two visits booked back to back are accepted by every tool that thinks a caregiver's day is a block; this one refuses the impossible one and allows the possible one, and the only difference between them is eleven miles.

The residency program is the only one that is not a week, and it could not be. Its rule is that eighty hours is a mean over four weeks rather than a weekly ceiling, and a mean over four weeks is not a fact about any one of them, so the seed is a whole four-week block and the table under the grid shows all of it at once while you edit one week at a time. Yusuf's weeks are 88, 76, 82, 74 hours: two of them over the line, an average of exactly 80.0, and not one thing flagged. A scheduler with a weekly cap in it refuses that block, and a scheduler that checks each week alone passes one that breaches, which is the expensive direction. Add 4 hours to his quietest week and watch which of the two numbers moves.

Everything you do runs the full engine in the browser as you drag: the same TypeScript that runs on the server, 15 rules on every assignment, so the verdict you see here is the verdict the database would give. Conflicts appear on the cell in the same frame as the click, the whole week's conflicts are listed under the grid, the labor estimate in the toolbar moves as you fill, and undo works. The estimate is for budgeting and says so: it is not a wage figure and never reads as one.

Four things under the grid are the product's own screens rather than a summary of them. Auto-fill solves the open seats hardest-first and hands back a list of checkboxes with the reasons attached — including the seats nobody can take, named, because that gap is the one somebody has to solve by Friday. Required staffingmultiplies the census against the facility's own rules, which is how a row that reads a comfortable 1/1 turns out to be a nurse short. The timesheet is the real CSV export, downloadable, with the real hours in it. And the fairness count is the same measure the fill waterfall uses to decide who gets offered an open shift first, so you can check that claim rather than take it.

There is also a toggle above the grid to read the same week as one of the nurses on it. The rota stays whole — a schedule everybody can see is the point of a schedule — and the management layer over it does not: no labor estimate, no colleague's compliance state, nothing editable. That is the boundary the product enforces with policies rather than with a hidden menu, and it is worth seeing before you invite fifteen people into it.

The people are seed data, written to look like a roster rather than a toy. Marisol is not a nurse; her license expiry is a fixture that also serves as a test case. Michael Grant is not on any exclusion list and has not been accused of anything; his record exists to show what a possible match looks like, because that is the case people get wrong. Cedar Hollow is not a facility and its CMS submission ID is invented; the drive times on the home care round are seeded numbers, where the product reads them from a routing service. Block 5 is not a program and Yusuf Adeniyi is not a resident; his rota is a fixture, built to land on an average of exactly 80.0 because that is the number worth being able to see. There is no patient data in this demo, which is easy for us to promise: there is no patient data anywhere in MedAligna.

The things that are not invented are the ones that could not be. The dates the clocks change come from the IANA timezone database. The PBJ file is built against the schema CMS actually publishes. And the duty-hour limits the residency block is measured against (eighty hours averaged over four weeks, one day in seven free averaged the same way) are the Accreditation Council for Graduate Medical Education's rather than ours. Those are checkable, which is exactly why they are the ones we let the machine produce.

Nothing you do here is saved. Drag what you like, break what you like, refresh, and it comes back exactly as it was. Nothing is sent anywhere, nobody is emailed, and this page does not ask you for an address in exchange for the interesting part: there is no interesting part being withheld.