School Choice and Enrollment Lottery Software Problems: The 7 That Cost Real Money, and How to Avoid Them

Why do priority rules get hardcoded so often?

Because on paper they look like a settings screen. Sibling priority, geographic priority, a weighted lottery for educationally disadvantaged students, continuing student rights, staff children, programme specific eligibility. Six checkboxes and a sort order.

Then you write out what your rules actually say. Does a sibling in a feeder school count. What happens when the sibling's own placement is not yet decided in the same run. Is geographic priority by walk zone polygon, attendance boundary, or the neighbourhood definition the board adopted last year that matches neither. Federal Charter Schools Program rules permit a weighted lottery in defined circumstances, and it has to be implemented exactly as adopted rather than approximately.

Then the board amends one tier in February, six weeks before the run, because a council member heard from constituents. A hardcoded rule set means a code change on somebody else's release calendar. A configured product means a vendor ticket and a hope. That timing is the single most common reason cities move to a build: the policy owner and the implementation have to be in the same building.

Represent priorities as ordered, composable rules with an effective date, so this year's run uses this year's policy and last year's run still reproduces exactly as it happened. Store every applicant's tier assignment with the reason, which is what lets a staff member answer a parent's call in one minute rather than escalating it.

What goes wrong with address, boundary and applicant history data?

Every geographic priority depends on turning what a parent typed into a verified point inside a polygon. The city's geographic information system holds the parcel data, the district holds the attendance boundaries, and the two disagree in enough cases to matter. A general purpose geocoder will place a family on the wrong side of a line silently, and you find out when the appeal arrives with a lease attached.

Resolve to a parcel rather than a street interpolation, using the authoritative local layer, record a confidence value, and route ambiguous cases to a human queue before the match rather than after it. The parcel data usually lives with the city rather than the district, so start that conversation at kickoff, because it is the most common cause of a first release running late.

Two obligations sit on top and are routinely bolted on too late. Students experiencing homelessness have rights under McKinney-Vento that make an address requirement inappropriate as a hard gate, so there must be a path that places a student without forcing a family to produce documents they do not have. Foster and military family provisions in your state may add their own priority and timing rules.

Prior year applicant history is worth migrating for one specific purpose: replaying last year's applicants against this year's proposed policy so a board can see what an amendment actually does before adopting it.

Why do student information system integrations break after the match?

Because a match that ends at a results page has not finished the job. Every placement has to land in the receiving school's student information system, and if charters are in the run that is several different systems with several different identifier schemes and several different governance bodies deciding what they will accept.

The break is identity. Without a stable student identifier that works across sectors, the same child exists as three records, and September becomes a manual reconciliation of duplicates. That reconciliation is not just administrative pain, it undermines the credibility of the whole exercise, because a family whose placement did not transfer cleanly experiences it as the lottery failing.

The second break is timing. Late applicants keep arriving all summer, mid year transfers run against live capacity, and each of those has to reach the same downstream systems through the same path. An integration built only for the March results release will be operated by hand from June onwards.

Agree the identifier scheme before development, in writing, with every participating sector. Build the placement handoff as a continuous, idempotent process that can be rerun safely rather than a one time export. And instrument it, so an unacknowledged placement raises an exception instead of sitting in a file somebody assumed was collected.

What happens when the waitlist and appeals operation is not covered?

The match is one day. The waitlist runs from March through October, and it is where families actually experience your system. A seat opens because a family moved. Everyone below moves up. Someone accepts, which vacates a held seat somewhere else, which cascades.

Scope only the match and the operation reverts to what it was. Principals watch a roster and start calling families directly, which is how a parallel unofficial process begins and how equity guarantees quietly disappear. Nobody planned that. It happens because the official process is slower than a phone call.

Cover it properly. Automatic cascading with rules: offer windows with a defined expiry, automatic decline at expiry, position preserved or forfeited according to policy, and manual overrides that require an authorised user and a recorded reason. Every offer, response and movement in an append only log, because the question in September will be why a specific child at position four did not get a seat that a child at position nine received, and the answer has to be a record rather than a recollection.

Appeals need a stated basis, a reviewer, a deadline and a recorded outcome. Late applicants need a defined position rule so lateness is neither an advantage nor a random penalty. This tail is where staff time goes and it is the most common thing a city discovers it did not scope.

Should you build custom or configure what you already own?

If you are a single district running magnet or transfer lotteries with straightforward sibling and boundary priorities, no charter participation and no unusual seat types, buy. SchoolMint is strong on the family facing side, application capture, communication and registration, and PowerSchool Enrollment is genuinely good at registration forms and the handoff into the student information system. The match at that scale is not the hard part.

Look seriously at Avela before commissioning anything. It was built specifically for unified enrollment and matching, and if your rules fit its model you will save a year and considerable money. We would rather say that than sell a build you do not need.

Build when two or more of these are true. Charters and district schools must be matched together under a governance agreement. Your priority rules come from board policy or a court order and change on a political calendar rather than a release calendar. You need to publish a seed and let a third party reproduce your match. You have programme seat types that interact, for example dual language and special education capacity sitting inside a general capacity limit. Or your waitlist operation runs for six months and currently depends on principals making phone calls. The test is simple: if a reporter asked you to prove the draw was fair, could you today, without asking a vendor.

How do hidden costs get into the quote?

Enrollment quotes go wrong in five predictable places.

• Sector count is treated as a number of schools. Each participating sector brings its own governance body, its own student information system and its own approval cycle, and the meetings are the slow part.
• Translation is quoted as a plug in. A legally consequential application in several languages is professional translation work with review, and it recurs every time the application changes.
• Seat capacity is quoted as one number per school. It is per grade, per programme, per seat type, with special education and bilingual capacity interacting and class size limits that are state rules rather than preferences.
• Address validation is quoted as geocoding. Parcel level resolution against the authoritative city layer, with a confidence value and a human queue, is a different piece of work with an external dependency.
• The tail after the match is left out entirely. Appeals, late applicants, mid year transfers and the placement handoff are where staff time actually goes.

The honest bands from Digital Heroes delivery experience are $80,000 to $180,000 in 14 to 20 weeks for a first release covering the multilingual common application, priority rules with effective dates, the match with published verifiable seeds and results release, and $200,000 to $450,000 phased over 6 to 12 months for the full platform adding cascading waitlists, parcel level address validation, residency and document review, appeals, late applicants, mid year transfers and integration across sectors.

What separates a build that works from one that fails here?

Reproducibility, above everything. Commit to the random seed before the run, for example by publishing a hash of the seed and the input file, then release the seed and an anonymised input afterwards so anyone can rerun the match and get the same assignment. That converts a trust argument into arithmetic, and it protects your staff, because the accusation that somebody put a thumb on the scale becomes checkable rather than merely deniable.

Second, the team has to understand what they are implementing. Ask them to explain student proposing deferred acceptance in plain language, then ask what changes when you use a single tiebreak number per student instead of one per school. Both produce valid matches, they produce different placements and very different waitlist behaviour, and your board has to choose. A team that cannot explain the difference will build a lottery rather than a match, and you find out in March.

Third, the deadline does not move. Scope the first release backwards from the application window and keep it narrow: application, priorities, match, results. Waitlist automation and appeals follow in the same year.

Fourth, ask about McKinney-Vento and cross sector identifiers unprompted. A team that raises them first has done this before.

Fifth, ownership in the contract before kickoff: the repository, the cloud accounts and the right to hire anyone else. At Digital Heroes the client owns the code from the first commit. For a public body this is not commercial hygiene, it is what lets you answer a records request from your own systems rather than asking a vendor for permission.