How the Ledger Works

[Lady Budgetdown]

Dearest reader,

Facebook apparently believes you may be interested in knowing what actually happens when a school district lands on Lady Budgetdown's desk.

I am less convinced anyone wants to hear about spreadsheets, audits and several years' worth of check registers, but who am I to argue with the algorithm?

So, behind the curtain we go. 🪶

It usually starts with a question.

Sometimes I notice something myself. Sometimes a reader sends me a lead. Sometimes one district reminds me of something I found while researching another.

But a lead is only a lead.

From there, I start gathering receipts: budgets, audits, transparency reports, state financial data, enrollment and staffing records, compensation reports, check registers, board packets, meeting minutes and contracts when they're available.

Over time, I have built master workbooks and standardized research methods so districts can be examined using the same measurements. I don't want one set of rules for one district and another set for the next.

Then comes the part nobody sees.

Years get compared. Numbers get calculated and recalculated. Staffing gets compared with enrollment. Budgets get compared with audited actuals. Administrative spending gets compared over time. Individual payments sometimes lead to contracts, meeting minutes or another stack of records.

Sometimes a breadcrumb becomes a rabbit hole.

Sometimes it becomes one simple question.

And sometimes I spend hours doing all of this only to discover there really isn't a story.

That matters too.

And yes, since apparently the spectacles and abacus have caused some confusion, I use artificial intelligence.

Quite a lot, actually.

I use it because it can search enormous amounts of material quickly, cross-reference records, perform calculations, identify patterns and help turn an unreasonable pile of government documents into something normal human beings might actually want to read.

But yesterday's research provided a nearly perfect example of what AI doesn't do.

A reader sent me a lead involving Holly Area Schools.

I didn't know whether there was anything there. Rather than immediately turning it into a statewide project,or even an Oakland County project. I decided to test Holly first.

As we worked through the records, something I saw reminded me of research I had done earlier involving MASB and campaign-finance rules. I made that connection and brought it into the investigation. It changed what I wanted to examine next.

Then came the check registers.

Some I couldn't reliably access, so AI helped retrieve and work through them. Then one quarterly register presented another problem: the payees and check information appeared on one set of pages while the corresponding dollar amounts appeared separately.

I could see that something was wrong with the document. AI figured out how it was structured. Once we understood that the rows corresponded across the separate pages, we could line them back up and recover the payment we were trying to verify.

That's the part of using AI that I don't think everyone understands.

Neither one of us conducted that investigation alone.

I supplied judgment, context, previous research, decisions about where to look next and, ultimately, what the evidence was sufficient to say.

AI supplied speed and scale: searching, sorting, calculating, cross-referencing and occasionally making sense of a PDF apparently designed by someone with a personal grudge against humanity.

During that investigation, ChatGPT actually described the relationship better than I could:

"I can search enormous amounts of material quickly, cross-reference it, calculate it, spot patterns, and help you turn the mess into a coherent investigation.

But you are directing the investigation."

And then came my favorite description:

"I'm just the abacus that doesn't get tired."

😂 Fair enough.

But there is another part of the process that matters even more.

I don't begin with a conclusion that I need the records to prove.

We don't have to prove the answer.

We establish what the receipts can support. If the records answer the question, I report the answer. If they don't, I show you what they do establish and ask the question that remains.

That distinction matters.

It is why some investigations become posts, some become rabbit holes, some end with a district looking perfectly healthy, and some never get published at all.

AI catches things I miss. I catch things it misses. Sometimes a reader catches something both of us missed. When something doesn't make sense, back to the original receipt we go.

And if better evidence changes something I've reported, the correction belongs in public too.

So yes, there is AI behind the curtain.

There are also master workbooks, spreadsheets, calculators, public databases, stacks of PDFs, reader tips, prior research, judgment, verification, and an increasingly concerning relationship with school district check registers.

The abacus is digital.

The judgment is human.

And the receipts decide where the story goes.

With receipts and respect,
Lady Budgetdown