Ezriel Schwartz · 40 schools · written 9 Nov 2026

What has to be built, step by step, and how hard each step is

Built from his original message and a line-by-line read of today's call, then checked against what already exists in the market.

The short answer

Yes, this can be built without Akiva, and built so it does not break. Nothing in it needs skills we lack. The scheduling brain runs on a free, proven solver from Google. Everything around it is ordinary business software, or something we buy.

Nobody sells the piece Ezriel cares about most, which is choosing which children share a room so a school needs fewer teachers. We tested that piece today on a made-up school. It found the best one to three child moves in about a second.

What going without Akiva costs is speed and cover. One person takes 7 to 12 weeks to reach a pilot at one school; two take about 4 to 7. And someone has to answer when a director calls and you are away.

Update, 10 Nov: Han's package arrived

Han built and tested a working schedule engine on four real schools. It changes this plan in five ways.

The hardest piece is mostly done. Her engine already builds shifts, lunches, prep time, floater routes and start-and-end-of-day merges, and checks an existing schedule against the rules. We adapt and harden it instead of inventing it.
The rule book is seeded. She sent rules for 29 states plus New York City, each with its legal source. Her own note says they still need a legal check before anything is enforced.
The saving is now measured on real schools. Her four test schools came out 9 to 44 paid hours a day lower than today's schedules, about 19 on average.
One real disagreement. Han's written rule says the software never changes which children are in which room. Ezriel's headline example on the call was exactly that. They need to settle it between them.
New work she found. In New York City a qualified teacher must be on site for each licence at all times. So each staff member's qualifications have to be stored and respected. That was not in our plan.

Her notes also mention about 60 schools and a written requirements document we have not seen. Ezriel said 40 schools on the call.

How to read the difficulty marks

simple real work, well understood hard, needs rounds with real people or real data do not attempt yet

"Build" means we write it. "Buy" means an existing product does it. "Copy" means a proven public method exists and we follow it.

Groundwork (needed before any phase)

Ordinary software

Step	What has to be done	How hard
Get the children's data	Pull each child's name, birth date, room, start and end dates, and every check-in and check-out from Illumine. Keep an untouched copy of everything received. Illumine sells this access only on its top plan and publishes no instructions, so we have not seen it yet. Until we do, phase 1 can run on spreadsheet exports.	Build · unknown until seen
The rule book	Every ratio, group size, age cut-off, mixed-age rule and nap rule goes in as a setting with a start date and the law it comes from. Careful work, not clever work. Two surprises from the research: New York City has different rules from New York State, and there are at least four different ways states handle mixed ages.	Build · copy the method
Schools, rooms, staff, logins	A list of schools, rooms and people. Who is a lead (salaried, fixed to a room) and who is an assistant (hourly). Each director sees only their school; head office sees all.	Build · buy the sign-in
Hosting and security	One database, one website, daily backups, a security check before go-live.	Copy

Phase 1: the weekly schedule

75–80% of his saving · about 40 users

Each director receives a finished schedule two weeks ahead. They only put names into slots.

Step	What has to be done	How hard
1. Count the children	From real check-in history, work out how many children are actually in each room every 15 minutes, for each weekday. Look two weeks ahead for birthdays that force a room change, new children and leavers.	Build
2. Set up the rooms	His "6 and 7 kids" example. Decide which children belong in which room so no room sits one child over a ratio step, and children with the same pickup time sit together. Output is a short list: "move these 3 children, saves this much." Nobody sells this. Tested today on made-up data.	Build
3. Teachers needed	Children divided by the ratio, rounded up, for every room and every 15 minutes.	Build
4. Planned merges	Combine half-empty rooms early and late in the day when it saves a teacher. A child moves at most once. Tested on made-up data on 6 Nov. Ezriel still has to say whether this is phase 1 or phase 2.	Build
5. Shifts and breaks	Turn "teachers needed" into assistant shifts: start and end times, lunch, who covers each break. Keep it the same as last week unless a change saves enough to be worth it. This was the hardest piece of code. Han's engine already does it for one standard day and was tested on four real schools. Left for us: a weekly version, each person's qualifications, covering call-outs, and making it faster.	Adapt Han's engine
6. Director's screen	Shows the finished plan. The director drops names into slots and can swap people. They cannot change shift lengths or room set-up. Anything that costs overtime goes up to head office.	Build · copy the layout
7. Send it out	Email the schedule to every staff member; printable copy for the school.	Buy the email service
8. Head-office view	One person sets the rules, sees every school, and sees hours scheduled before and after. That last number is his proof of saving.	Build
9. Getting a director to accept it	The first real schedule will have things a director says cannot work. Expect several rounds at the pilot school. This is people work, and it decides whether the saving is real.	Not code

Size: 200–350 working hours to a pilot at one school, then 120–205 hours to roll out to the other 39 over about three months. These are our own earlier estimates and have not changed.

Phase 2: the live day

He definitely wants it

As children arrive early, stay late or do not come, the director is told "this room is about to be short" or "these two rooms can merge and an assistant can go."

Step	What has to be done	How hard
1. Live check-ins	Receive each check-in and check-out from Illumine within a minute or so.	Build · depends on Illumine
2. Live room picture	Children really present against the plan. It knows which teachers are scheduled, not which are standing in the room. The main danger is wrong data: a parent forgets to sign out and the system sees a child who is not there.	Build
3. Decide what to do	Re-run the phase 1 brain for the rest of the day, following a fixed order agreed in advance: who covers a gap, who is offered to leave first. Before anyone is sent home, the director taps to confirm a head count.	Build · copy the playbook idea
4. Tell people	Alerts on the director's screen and the room iPads (teachers' phones are put away), with texts for urgent ones. Only interrupt when someone must act now, or people stop looking.	Buy texting · copy alert rules
5. Stay up	If Illumine or the internet goes down, every screen says "not watching, follow today's plan" and we get a text. The weekly plan still stands.	Build

Existing childcare products only alert after a room is already out of ratio. None of them predicts it or suggests a merge. Size: not estimated yet; it needs one look at Illumine's live feed first.

Phase 3: clock to payroll

At most ~$200k a year

Buy almost all of it from whichever payroll company he picks. Their clock, timesheets, corrections and overtime rules already exist.

Step	What has to be done	How hard
Clock, timesheets, corrections, overtime	Use the payroll company's own.	Buy
Connect our schedule to it	Send our schedule in so their system flags missed punches. Pull punches back for the live screen and for a report of scheduled, punched and paid hours.	Build
Paying to the real leaving time	This is a policy, not technology. When someone leaves without punching out, the director enters the real time the same day and the employee can dispute it. A lawyer in each state has to approve the wording.	Policy · lawyer signs off

Time-sensitive. He is choosing payroll within two weeks. Payroll companies that let a customer send hours in directly: Rippling, UKG Ready, ADP Workforce Now, Paylocity. Gusto does not. He should get that access confirmed in writing before signing.

Phase 4: badges and room tracking

Recommend not building

Step	What has to be done	How hard
Staff badges and room sensors	Hospital-grade equipment, roughly $8k–$16k per school per year. For 40 schools that is more than the $200k it could save. Never used in childcare. If operations insists on knowing who is in a room, a "here / not here" tap on the room iPad does most of it for almost nothing.	Skip

Why it will not break with 40 directors and 800 staff

The load is small

Even with every room iPad refreshing, the system handles fewer than 20 requests a second at its busiest. One modest server carries that. This is arithmetic, not yet a live test.

Phase 1 cannot fail in the morning

The schedule is made once a week, in the background. If our system is down on a Tuesday, the schedule already emailed still stands.

Rules are settings, checked on every change

A new state or a changed ratio is typed into the rule book, not programmed. Each state's own worked examples are stored as tests. A wrong entry is refused before it can reach a schedule.

Every schedule is checked twice

The brain makes the schedule. A separate, much simpler checker confirms every room is in ratio at every 15 minutes before a director sees it.

The live phase fails safely

When data stops arriving, screens say so and people follow the plan. Nobody is sent home on the system's word alone.

We buy what others do well

Payroll clock, texting, email and sign-in come from established companies. We build only what nobody sells.

What is still unproven

Nothing has run on a real school's data. Every test so far used made-up children.
We have read Han's engine and her results but have not run it ourselves or seen the real school files behind them.
We have not seen Illumine's data access or the written requirements document Han refers to.
Han's state rules were copied from a research document, not from the regulations themselves. Legal has to confirm them.
Counting each child by date of birth, the way each state requires for mixed ages, is not in Han's engine. States use more than a dozen different methods.
Akiva's "4 to 8 weeks for phases 1 and 2 together" is not supported by our estimates.

What we need from Ezriel

The written requirements document Han refers to, and the real attendance and schedule files for her test schools. She will send the files once a data agreement is in place.
A ruling between Ezriel and Han: may the software propose moving children between rooms, or not?
Illumine access: the licence, the instructions and a test account.
One pilot school and its director's time.
Is the planned end-of-day room merge part of phase 1 or phase 2?
Who approves moving a child to another room, and who tells the parent?
Who at head office owns the rules and can hold directors to the plan?

First thing to build

One real school, start to finish: its Illumine data in, room set-up and shifts out, the director's screen, the email. That single run on real data tells us more than any further planning, and it is the pilot's first week.