After several delays, @DOGE has finally posted its purported savings. Why did it take so long to create a simple webpage with a 1000-row table? Who knows! Let's dig in.
Headline number: $55B saved. They list the savings per nixed contract. This should be easy to verify then.
The first thing I did is add up the "saved" column for all canceled contracts and real estate. The numbers are $16.5B and $0.14B, respectively. Odd...
Since almost all of the purported savings come from contracts, we'll focus on that.
2/
The single biggest ticket item is a DHS contract listed as saving $8 billion. Wow, that's a huge contract!
Actually no, it's $8 million. They must have tried to automate scraping the FPDS form and failed.
That means we're down to $8.5B in savings.
3/
The next 3 biggest ticket items are all USAID contracts listed as $655M each, so $2B total. Wow, pretty big.
Wait, these are IDVs, not contracts. $655M is the entire set-aside, being triple counted. In the first 5 years, only $73M was awarded, and only 2 years remain.
4/
So we're down to $6.5B in savings, and an alarming trend emerges: @DOGE does not seem to understand how the government contracts they are canceling work. The savings they are claiming are not annual savings, but rather hypothetical savings if we spent every unobligated penny.
5/
And more importantly, they are just getting it *wrong*, with alarming consistency. These numbers are erroneous. This "select group of geniuses" has not double checked even the LARGEST items accounting for the bulk of their claimed savings. This is a sad, pathetic farce
@elonmusk
@elonmusk Here's the next biggest item: an IT services contract for the Social Security Administration worth $1B. That's a lot of savings!
Well, again, this contract spanned 6 years. 80% has already been spent. Ah well, more like $240M in savings spread over the next 3 years. $80M/year.
@elonmusk In 2023, this contract funded 1000 FTEs working $100/hr. Did we need 1000 SWEs working on SSA infrastructure? Probably not - these could be valid savings (disclaimer: no idea what they actually did). But worth noting that these cuts will impact many private sector jobs as well.
@elonmusk And if anyone is curious, there are currently 17 lines that say "SEE FPDS" rather than the savings amount, I guess because their automated scraping failed. I did it manually and it took roughly ~10 minutes. But that's too much to ask of super geniuses working 120 hours/week!
@elonmusk An important update for those just now reading this. This is not what transparency looks like!
@philotheis49253 @marilynmaupin @JuddLegum @DataRepublican @Oilfield_Rando I ran the same query and got completely different results. Of contracts > $1M ending in 2024:
- 30.5% had obligations < potential value
- 69.3% were equal
- 0.2% went over
Furthermore, if we project "savings" using DOGE math (potential - obligations) on 1/1/2024...
@philotheis49253 @marilynmaupin @JuddLegum @DataRepublican @Oilfield_Rando ...45% end up falling short of the projected "savings" (comparing final numbers to beginning of year snapshot), with actual savings of those contracts averaging 40% of the "DOGE projection".
I'm not sure what she did, but it's wrong. I do have a guess though:
@philotheis49253 @marilynmaupin @JuddLegum @DataRepublican @Oilfield_Rando If she downloaded data using USAspending's bulk_download endpoint, data is returned as a list of transactions. Each contract has a transaction any time there was a funding action, in which case there are fields denoting the change, as well as fields denoting the totals to date.
1) "savings" divided by years remaining 2) mean annual spend to date
2) is only considered if there is at least 1 full year of data. This number is sometimes much lower, meaning the contract was not on pace to realize the full award.
Now I'll list 10 contracts accounting for >40% of the newly listed "savings" along with an explanation of each.
#1-#4 will be repeats of my last thread, but including annual spend so you can see how DOGE tends to exaggerate "savings" obtained by cutting worthwhile programs.
The EPA press release echoes NY Post, which on Wednesday published one of the most pathetically under-researched articles I've ever seen. It reads like a @DataRepublican post, alleging that the EPA gave BILLIONS to unaccountable groups that had only been formed recently.
The first example is Climate United Fund, which the author complains formed only months before being awarded $7B.
In reality, it's a joint venture between Calvert Impact, Self-Help Ventures Fund, and Community Preservation Corp.
They are 30, 44, and 50 years old, respectively.
Calvert is an investment firm focused on social and environmental impact. The latter two are CDFIs - credit unions that invest in revitalizing underserved communities across the country.
JVs that draw on expertise of multiple orgs to meet the needs of a contract is not unusual.
DOGE currently lists $30B of cuts consisting of $15.2B to contracts, $14.5B to grants, & $470M to leases.
After fixing accounting errors, this amounts to $4B/year of contracts and $170M/year of lease cancellations. If we assume similar term for grants, total is roughly $8B/year?
(Grants can't be independently audited as DOGE currently provides no metadata)
Of note, 80% of cuts came from dismantling USAID. Today, Supreme Court ruled that Trump admin must release $2B to contractors for work rendered, so this # should come down.
My commentary: why did we spend $40M on a slipshod flock of twits to turn off USAID? By all accounts, they made no effort to understand the importance of the programs they killed, and stranded US workers in dangerous situations abroad in the process.
a) This is not what Elon said at all
b) Appeal to false authority (he makes rockets!)
c) Weak analogy - calls Social Security a "complex system" likened to ML because it has... 3 steps?
d) Applies irrelevant concepts, none of which are described correctly
"These errors compound, not linearly, but exponentially"
No, these errors would compound sublinearly. If we have an error rate e for a 2 step process, the probability of correctness is (1-e)^2, or 1 - 2e + e^2, which is always greater than 1 - 2e (simple addition of errors).
"A small data flaw at the base level snowballs into massive discrepancies"
You're describing chaos. Complex estimators with millions of parameters can overfit to noise. Long iterative processes can diverge. This is all irrelevant; nothing about SSA is "complex" in this sense.
The first $16M of @KristiNoem's wasteful $200M advertising campaign "warning illegal aliens to leave our country" has been awarded to Safe America Media LLC after a "competitive procurement process"
Safe America Media LLC is a shell company that was incorporated in Delaware 2 weeks (!) ago. That's only 11 days before Noem made the announcement. No wonder they don't even exist in SAM's entity search.
So what did this newborn vendor do to win this contract?...
Well, the contract is tied to a residential address that I was able to link via property records to... the former deputy political director of the National Republican Congressional Committee.
I can see why they hired the guy! What's DMM, where he works as a partner?