At most of their public meetings, the OUSD School Board approves tens of millions of dollars in spending, in the form of dozens to hundreds of contracts, grants and agreements in one, sweeping vote called theGeneral Consent Report. It is up to both the board and the public to try and read through the entire consent report, often with only a few days notice before it is presented and voted on, in the form of a cluttered PDF or antiquated web platform.

This tool exists to make that process much easier, more organized, and historically trackable. It is built using official public data surfaced by OUSD's Legistar system, and is more or less a consolidated interface that organizes that data for higher transparency and understanding.

I am an involved parent of two OUSD students, and have struggled to understand the official language in basically every consent report. I have also been a web developer for over two decades, and wanted to find a way to leverage LLMs (AI) to systematically decode and demystify these reports as they come out. Massive expenditures of funds are approved by the board in the form of these blanket votes each year. While the district does provide all this information publicly, it clearly has neither the time or resources to present it in a user-friendly format. I do not believe this is done intentionally, but it is a critical gap in the transparency that we depend on to make informed decisions about the district's future.

This project was built to help our school board, district leadership, and the rest of the OUSD community find a set of shared facts about how money is spent here, in the hopes of fostering trust and collaboration between all of us.

Get notified

Sign up to get one email per meeting, sent once the plain-English summaries for that meeting's consent report are ready, usually a few days before the Board votes. You'll confirm your address first, and every email has an unsubscribe link. Your address is only used for these alerts.

Sign up for consent report alerts

How this project uses Artificial Intelligence

AI models are used in a few specific, limited steps. Everything else, including fetching the official records, checking amounts, adding up totals, and most flags, is done by ordinary code with no AI involved. The official text from Legistar is never changed, and it's always shown alongside anything AI wrote about it.

AI-written text is called out as AI-generated throughout this tool, marked with this little robot. Summaries, vendor descriptions and AI-confirmed discrepancies each carry their own label. Item headlines, which appear in long lists, are written by the same AI step as the summaries and are covered by the notice at the bottom of every page rather than labeled one by one.

Item headlines and summaries

For each consent item, an AI model reads only the official item text and writes a short plain-English headline and a 2–3 sentence summary. It's instructed to use neutral language, spell out jargon, avoid speculating beyond the text, and describe individual contractors by role rather than name. Every summary carries a label saying it was written by AI, along with its review status; headlines share that summary's label on the item's own page.

Reading the facts from the text

The same AI step pulls structured facts out of the official text: the vendor, any schools named, the contract dates, and the dollar amounts (what this vote adds, prior and new totals, and whether it's a yearly cap). These are then checked in code:

1. Every amount must appear, word for word, in the official text. If it doesn't, the summary isn't published and the item shows the official text alone.
2. A second, independent AI reading re-reads the money on items where a check raises a question, without seeing the first answer. If the two disagree in a way that changes totals, a third reading breaks the tie.
3. A human is alerted and an item flagged for manual review if disagreement is material (over $10,000 and over 1% of the item) or the official text leaves the money genuinely unclear.

Categories and action types

The AI model assigns each item one category (see Categories), a sub-category when the category is Special education, and one action type, such as new agreement, amendment or change order, using the written definitions on this page. These are this tool's classification, not fields from Legistar, and they're labeled as AI-assigned where they're summarized.

Flags

A few flags, such as “no competitive bid” or “emergency”, are judgments read from the text by AI. Most, such as “raises an existing contract” or “work began before approval”, are computed in code from the facts above. Each flag in the list below says which kind it is. When the AI notices a possible error or inconsistency in the official text itself, it's shown only after it's confirmed by arithmetic in code, a second independent AI reading, or a human, and the item says which.

Vendor descriptions

Some vendor pages have a short description with public contact details, marked with a checkmark in the vendor list and labeled as AI-researched. An AI model researches the vendor from public sources such as its own website and state, IRS and federal registries. The description is published only when:

• the model is highly confident it found the right organization, with at least two independent matches;
• a second, independent AI review checks the sources and agrees;
• every phone number, email, address and registry ID is found, word for word, on a page it cites.

Anything that doesn't pass is left out. A vendor can lack a description because it has little public presence online, its name matches other organizations, its sources couldn't be verified, or it hasn't been researched yet. Individual contractors are never researched. Descriptions don't come from OUSD or Legistar and may be out of date.

Where AI is not used

• Fetching meetings, items, votes, attachments and history from Legistar.
• Matching vendor records across meetings (done by rule, with hand-edited corrections).
• Totals, charts, and the “next consent report” preview.
• The official item text, which is always shown as published by OUSD.

How money is counted

Every item that involves money shows one main dollar figure: what this vote approves. Here's what that means in different situations, and how the meeting totals are added up.

• New contracts show the most the district has agreed it could pay. Most contracts set a ceiling (“not to exceed”), so the district may end up paying less, but never more without a new vote.
• Changes to existing contracts show only the new money being added, not the contract's new overall total. For example, if a $100,000 contract is raised to $150,000, this vote is $50,000. The earlier $100,000 was already approved at a past meeting, so counting it again would double-count it.
• Money coming in, such as grants the district receives, is counted separately from spending. Cuts to existing contracts aren't subtracted from spending either; they're noted on the item.
• Yearly limits: some agreements set a maximum per year instead of one overall amount. Adding a per-year figure to one-time amounts would mix two different things, so these are shown on their own, as a per-year total.
• Sales limits on auctions of surplus district property aren't spending, so they're left out of spending totals.

Every dollar figure comes straight from the official text. Nothing is estimated or filled in: if the text doesn't state an amount, none is shown. Meeting totals are simply these figures added together, and amounts are only rounded for display (for example, $1.2M), never in the underlying data.

Flags

Flags describe procedures, not motives. Most flagged items are routine and legal.

Categories

Each item gets exactly one category, based on what the money or agreement is for. When an item could fit more than one, it goes where its purpose is clearest. For example, busing and legal work for special education are under Special education, and software bought for a specific program goes with that program rather than under Technology.

A note on “Special Education”

“Special Education” is about a quarter of the consent-report spending this tracker counts, and it's the hardest category to read. The label covers very different things: full-time placements of students in private schools, therapists and nurses, busing, training, legal settlements, and everyday kinds of support, like staff who cover leaves and vacancies, that would be filed somewhere else if they didn't touch a student with an IEP.

Most of the money goes to nonpublic schools (NPS) and nonpublic agencies (NPA), which are very different. A nonpublic school is a state-certified private school where the district places a student full time, sometimes residentially. A nonpublic agency provides services like speech, occupational or behavioral therapy to students who stay in district schools. But the official text for most of these contracts is the same template wording, a not-to-exceed amount “to provide Special Education services,” so the public record rarely says which kind of provider it is, what's being bought, or for how many students.

The sub-categories below are my best attempt at sorting it out. Please read any “Special Education” total as a reason to ask questions, not as a conclusion. I wrote more about this in Addendum: First Revelations from the Consent Report Tracker.

Special education sub-categories

Each Special education item also gets one sub-category, assigned by AI the same way as its category, from the official text.

Action types

• New agreement: A new contract or agreement with a vendor or partner.
• Amendment: A change to an existing contract, such as adding money, time or work.
• Change order: A change to the work or price of a construction project already under contract.
• Bid award: Awarding a contract to the winner of a competitive bid.
• Cooperative purchase: Buying through a contract another public agency already bid out (sometimes called "piggybacking"), instead of running the district's own bid.
• MOU or data sharing: A memorandum of understanding (MOU) or data-sharing agreement, usually with a partner organization and often at no cost.
• Grant or funding in: Accepting a grant or other money coming in to the district.
• School plan: Approving a school's annual School Plan for Student Achievement (SPSA).
• Personnel: Staffing matters, such as personnel reports, job descriptions and employee agreements.
• Resolution or policy: A board resolution or a new or updated board policy.
• Termination: Ending an existing contract or agreement.
• Other: Anything that doesn't fit the types above.

Data

All of the data behind this tool, and the code that produces it, is public in the ousd-consent-data repository on GitHub, with a mirror on Codeberg. It includes the official record from Legistar for every meeting, every AI summary and second reading (each with the model and instructions that produced it), every correction, and the published files this site is built from, as JSON and as spreadsheet-friendly CSV. Each meeting page here also links to its own data as JSON.

The repository's history is a public record of every agenda revision, outcome and correction. The code is under the MIT license. This project's own work (headlines, summaries, classifications, flags, second readings and vendor profiles) is under CC BY 4.0; please credit “OUSD Consent Tracker, Oakland vs. the World” and link back. The official text is a public record of the Oakland Unified School District. Tagged releases are archived with a DOI for citing.

Corrections and reporting a problem

When a published claim changes (an amount, vendor, category, flag or the meaning of a summary), the item shows the date and a note. Corrections made by a person always take precedence over what the AI wrote.

If something looks wrong, first compare it with the item's official page on Legistar, which every item links to:

• If Legistar has the same problem, it's an error in OUSD's own records, which only the district can fix. Contact the Office of the Board of Education at boe@ousd.org or (510) 879-1940.
• If Legistar is right and this tool is wrong, please open an issue in the data repository using the “Something's wrong with an item” form. It asks for the meeting date, the item's file number and what the official text says, which is everything needed to check and fix it.

This project is maintained by one busy, human dad. Every report gets read, but it may take a little while to get to stuff.

Disclosure

This tool was created and is maintained as a personal project by Matt Glaser, an OUSD parent and volunteer serving as a lead delegate to OUSD's LCAP Parent and Student Advisory Committee (PSAC), as a founding member of the Multi-Stakeholder Engagement Group (MSEG), a community engagement body established under PSAC, and as the chair of the Peralta Elementary's School Site Council (SSC).

The General Consent Tracker is presented with best intentions but is not meant to speak for, and has not been reviewed by, PSAC, MSEG, any school site council, the Board of Education or the OUSD itself. Engineering, compute and hosting costs are donated by Aleph.