The DOGE of it All

The Department of Government Efficiency (DOGE) is BS. I think it’s fair to start this with that blunt statement. I’m not going to bother evaluating the legality of the endeavor, that’s for lawyers and judges and history to decide. I want you to understand that DOGE being BS has very little to do with some of the stated goals of DOGE, which you may agree or disagree with, and has very little to do with any debate about the scope and power of the executive branch. I do think it is worth remembering, as a citizen, that the entire point of our system was to make it hard for a President to do things without some affirmative action from the other two branches of government, but that has nothing to do with why DOGE is BS. 

DOGE is fundamentally unserious because their work is flawed and executed poorly. They fail technically, they are sloppy, which should be enough for you to distrust their work, and ultimately they are failing to address the most straightforward solution available to close the gap on the deficit. There is no reason to take anything they are doing seriously, because it is not serious work. I am confident that most programmers and data professionals would agree with me on this, if they took a cursory glance at the work they are producing. The NYTimes has done excellent work documenting the evidence for DOGE being BS, but I will go a bit further and explain why these discrepancies are sloppy, lazy, and frankly avoidable errors, from the perspective of a guy who works with data and data systems every single day. 

Let’s start with an accounting of the types of errors that were found on the official DOGE website. Politico writes: 

“But among the 1,100-plus contracts purportedly canceled, POLITICO found:

• Contracts that had not yet been awarded
• Instances where a single pot of money is listed multiple times — tripling or quadrupling the amount of savings claimed
• Purchase agreements that have no record of being canceled, but were instead stripped of language related to diversity, equity and inclusion
• Contract savings identified by DOGE that do not match with records they refer to in the Federal Procurement Data System
• Contracts where the underlying document is for an entirely different contract”

As a data scientist that has worked on government projects, and more broadly as a consultant, I’m confident in saying that none of these errors were nefarious reporting done to overstate the impact of DOGE, they were just sloppy work built on bad code and misunderstandings of the data. I’ll take each error one by one and tell you how they probably went wrong.

Technical Errors Made by DOGE

1. Identifying contracts that had not been awarded as “savings”: This is almost certainly due to the fact that they used a few wrong fields in the database. They likely erroneously took the date that a Request for Proposals (RFP) was issued or responses were due as the start date of the contract. They also likely used a field that estimates the maximum contract award value to calculate a real contract value, or it’s possible that the field typically used for total contract value gets imputed with estimated values for contracts yet-to-be awarded. 
2. Single pot of money listed multiple times: This is what is known as a “bad join”. Modern databases organize storage in multiple tables that help reduce the size of the overall database while enabling “relationships” between the tables that can help apps or analysts access the many dimensions of data associated with different records in the database. Below is an example of how a DOGE staffer could erroneously create a table that double or triple counts contracts. Imagine a contracts table that records high level contract data that looks like this:

And an amendments table that records amendments to contracts that looks like this. 

One could imagine that an analyst might want to combine these tables so that they had records for contracts and amendments all in one table they could analyze. In this case Gamma Inc has no amendments, Acme Corp has two amendments, and Beta LLC has one amendment. How would we combine these datasets so we have contract value, amendment value, and amendment date all in one dataset? One might use the contract ID to help create this table:

Here we have at least one record per contract, but for any contract with multiple amendments we have multiple rows. The problem here is that Acme Corp has two records in the dataset. If Acme Corp’s contract is cancelled, and the analyst reports the sum of the cancelled contract value by contract ID, they would erroneously report that they had cancelled $200,000 worth of contracts from Acme Corp when it was really $100,000. Additionally, if they count the records they cancelled, they would erroneously say that two contracts to Acme Corp were cancelled instead of one. This is an easy mistake to make, but a good analyst will check for these kinds of errors as they code up their analysis. 

3. Purchase agreements that were stripped of DEI language: I think this error is self-evident. They counted any DOGE-related modifications to contracts as cancellations and savings, even though there was no monetary saving associated with the modification.
4. Contract savings ID’d by DOGE that do not match the Federal Procurement Data System: This seems like a similar error to #2. DOGE analysts almost certainly created a bad join that tied incorrect contract identifiers to the wrong contracts.
5. Contract doc for an entirely different contract: This is likely a problem with the DOGE database itself, where link outs to contracts were wrongly assigned. This is also likely related to the bad joins I discussed in 2 and 4. 

The bottom line is this is sloppy work, and while the errors are common, they are not commonly found in final, published analysis. I have made all of these mistakes as I was learning to process data, but it certainly never made it into a final presentation. Publishing analysis containing these fundamental errors on a platform viewed by millions would be professionally unacceptable.

Other Sources of Sloppiness 

There were many other errors that DOGE made related to contract knowledge. Many noted that the contract cancellation values were surprisingly round numbers (numbers like $10,000,000). Most of the time, in Government contracting, a large round number like that is associated with a Blanket Purchase Agreement (BPA). BPA contract values are like a credit card limit, and often get used for things like IT contracts. Agencies know they want IT services, but are not sure when problems will come up. So agencies issue a BPA for the services over five years, and pay as they go. These contracts rarely ever realize a value anywhere near the “credit card limit” of the BPA, so the best way to estimate the value of the cancelled contract is to project forward the current amount spent. So if a contract spent $1.2 million in year one of a five year contract, you can probably estimate that cancelling would save $1.2*4 (years remaining) or $4.8 million. DOGE often not only failed to do this kind of projection for BPAs, but they also did not account for money already spent. Again, sloppy work. 

Given the nature of the errors and their frequency, I infer this reflects a culture of hubris, where DOGE staff may not have adequately respected or engaged federal employees responsible for the underlying data. If I were new to an organization's database structures, the first thing I would do would be to request a data dictionary if it’s available, so I can be clear on what each data field represents, and if possible I would try to talk to people involved in data entry, data management, and vendor management so I would have a holistic view of the data. These are basic professional steps to take in order to conduct professional data analysis. This is not fully a prerequisite, I would start doing my analysis in parallel to that process, but I would not publish work until I had the full understanding of the data, so I could be sure my analysis was correct. These people might have sought out documentation, but I think their evident lack of respect for federal employees guarantees that they did not bother to talk to the people responsible for data entry and management. They thought they could plug into the system and figure it out, because they are overconfident and lack respect for the hard work that went into developing that system. Precisely what I’d expect from a bunch of tech whizzes in their early 20’s who lack strong leadership. 

A One-Sided Solution to a Two-Sided Problem, and They are Missing the Biggest Side

But beyond the technical errors and overall sloppiness, the approach to the deficit is suboptimal. I know it’s easy to get the electorate on board with cutting costs, and I understand why that’s an easy sell, but the reality is it doesn’t get us very far. There is a GAO report, developed under Biden, that Elon cites as evidence that fraudulent payments are rampant. The reality is that the report identifies around $162 billion in improper payments in 2024, down from $236 billion in 2023 and falling rapidly because a huge chunk of those fraudulent payments were due to COVID era programs like the Paycheck Protection Program. And while these are staggering numbers in any other context, in the context of government payments that is a drop in the bucket. 

I do not think it makes sense to analogize government to a business, but since the crowd that loves DOGE insists on doing so, let’s play their game. In this case our business has three problems, it has a wide array of services driving costs but the vast majority of them are mandatory, quite literally required by law. It has a small amount of fraud (as a percentage of the entire enterprise). Finally, it has uncollected revenue. The uncollected revenue is estimated to be $606 billion, almost four times as large as 2024’s estimated fraudulent payments. Any smart business would put much greater emphasis on collecting all revenues owed rather than nibbling at costs. Any “businesslike” deficit reduction efforts that fail to address the revenue side of the equation should not be taken seriously. 

In other words, focusing only on cutting costs, while entirely ignoring revenue, will never produce meaningful deficit reduction.

DOGE - Not to be Trusted

Taken together—poor technical execution, repeated analytical sloppiness, disregard for available expertise, and a fundamental misunderstanding of how to meaningfully reduce deficits—make DOGE’s analysis unworthy of serious consideration. If a team in your professional life consistently demonstrated this level of incompetence, you'd rightly refuse to trust their work going forward.