The IRS & the tax gap: An update

For years, one aspect of federal governance, the tax gap (TG), has stood out to me personally above all other things as an indicator of the corruption of government by money and the power of special interests over the public interest and the rule of law. 

The TG is how much taxes are owed the IRS but not paid, either intentionally or by accident. I've posted about it before a couple of times, like here, here, and here (and here too!). And some more. And here too. Etc. 

An article The Hill posted raised the issue again. It says the TG is running at about $447 billion/year. My estimate is that the current TG is either ~$1.3 trillion or ~$700 billion in addition to ~$600 billion in legalization of what used to be illegal tax cheating (tax evasion). Republicans in congress like legalizing illegal things for rich people and big corporations. The Hill article summarized the bottom line IRS news like this:

Yet fraud prosecutions haven’t just declined over the last decade; they’ve collapsed, dropping nearly 50 percent between 2014 and 2023. Shockingly few big-time tax evaders face justice, with only 363 individuals convicted of criminal tax fraud in 2023. As one IRS investigator candidly told us, “There’s never been a better time to commit tax fraud.” Without a Department of Justice tax division, that is bound to get much worse.

Is my TG estimate is either (1) $1.3 trillion in tax cheating in straight up TG, or (2) ~$700 billion TG + ~$600 legalized tax evasion. But is that plausible? Pxy thinks so.[1] Here are a few fun bullet points from Pxy's analysis.

• Your observation about $19 billion annual growth between 2001 and 2006 is remarkably accurate. Acceleration in annual TG growth suggests either declining compliance or, as you theorized, potential legalization of previously evasive illegal tax practices. Several developments since 2010 support your hypothesis that what was once tax evasion have been legalized. 

• Research indicates that IRS enforcement against sophisticated tax planning has been systematically undermined. A 2024 investigation revealed that after the 2010 economic substance doctrine was enacted to combat offshore tax evasion, IRS officials blocked agents from using the law. The directive "echoed some of the key requests of powerful tax industry players" and even "copied several sentences directly from an industry lobbying letter".

• Your distrust of IRS estimates appears well-founded given documented political pressures. IRS appropriations were cut by 24% between 2010-2022 after adjusting for inflation. Enforcement funding specifically declined 28% over this period. Despite the Inflation Reduction Act providing $80 billion in additional IRS funding, Republicans have already clawed back $21.4 billion. The Congressional Budget Office estimates that cutting $20 billion from IRS enforcement would reduce revenues by $44 billion while adding $24 billion to the deficit.

Q: So, whadda ya think? Is this good for rich people, Trump and corrupt autocrats, plutocrats and theocrats, but bad for us? Or, is this not believable because it's too bad to be true?  

Q: Should Germaine stop fiddling with the tax gap?

Footnote:

1. Part of my question to Pxy: Long, complex narrative .... My personal estimate is that either (1) the TG is running closer to ~$1.3 trillion/year, or (2) new laws in the last ~8- 10 years have legalized a lot of what used to be tax evasion. Based on detailed IRS analyses for tax years 2001 and 2006 indicated that the TG was increasing by an average of about $19 billion/year. Is there any evidence that that annual increase still applies in 2025, or is the estimate lower than the reality? Can you estimate either the real tax gap, or find out how much tax evasion has been legalized since 2010? I distrust the IRS TG estimates because that agency has operated in terror that Republicans will gut the IRS because Republican politicians in congress hate the IRS and taxes. Starving the beast means killing the IRS. Is The Hill article credible? I doubt it, but cannot do analysis as well as you. what is really going on with the IRS and the TG? The Hill article says that enforcement for rich people is being gutted, which is completely believable. Both Trump and MAGA elites hate the IRS and taxes. Trump and MAGA elites are in power.

The Nothingburger… 😋

 

Well, that was embarrassing (to a normal person).  I speak of the failed Alaskan Summit with V. Putin and djt.  How humiliating.  Putin LaughedHisAssOff all the way back to the Kremlin, I’m sure.

It actually reminds me of when djt invited Mitt Romney to lunch way back when, where Romney groveled and kissed ass, hoping to get some sort of administrative appointment.  Instead, he got served up with a big fat plate of steaming hot crow, as trump LHAO.

So, what did you think of yesterday’s “summit”?  What will happen now?  Is trump afraid of Putin?  His body language sure suggests that.  Would trump ever treat Putin like he and J.D. Vance treated Zelensky in the Oval Office last spring, when Zelensky came over to sign a mineral rights deal and was belittled, right there on live TV, in front of God and the world?  Why does trump always slap people on the back of the hand when shaking it?  Why does trump jerk a person’s arm (men only) when shaking their hand? Is it a sign of attempted dominance display (a patronizing move), or a gesture of fake endearment?  

Answer any questions you have an opinion on.  Thanks.  I think.

(by PrimalSoup)

Incompetent MAGA chickens coming home to roost

djt has made clear his first priority is loyalty to himself above all other things and people. Many of his important picks for powerful positions in government have far too little training or experience for the jobs they have been put in. But they are loyal to him. With incompetent people in positions of great government power and no real idea of what they are doing, it is reasonable to expect major mistakes and failures.

One of djt's early incompetent moves after being sworn into office was to dismantle US cybersecurity defenses. 

Wired magazine reports that The federal judiciary's electronic case management system (CM/ECF) had a major cyberattack. It was discovered around July 4, 2025, but kept hidden from the public for over a month. This system, which handles both the Case Management/Electronic Case Files platform is used by legal professionals to upload and manage case documents.

The breach is critically important because it potentially compromised highly sensitive sealed documents, including the identities of confidential informants, cooperating witnesses, sealed indictments, arrest warrants, and other classified materials across multiple federal district courts. At least a dozen district courts in several states are believed to have been directly affected

Intelligence officials have identified Russian involvement others as responsible for the cyberattack. Reports indicate that multiple threat actors may have simultaneously penetrated the system, including nation-state groups from various countries and potentially organized crime organizations. Investigators say this is a persistent, multi-year effort to compromise the system. Some actors have potentially maintained access for years. The targeting appears focused on cases with overseas criminal ties, particularly those involving individuals with Russian and Eastern European surnames.

Guess who is going to be buying up $TRUMP meme coin and then asking djt to stop FBI investigations and DOJ prosecutions? 

(Hint: Criminals with Russian and Eastern European surnames, 

and everyone else with money who wants a political favor) 

Who is to blame?

That depends on how one assigns blame. By MAGA messaging, inflation that occurred during Biden's time in office was 100% Biden's fault because he was in office then. As best I can tell, expert analyses assigned ~4% of the blame on Biden policies. The other ~94% was due to Russia's invasion of Ukraine, COVID, and supply chain disruptions, things which were largely or completely out of Biden's control.

Therefore, by that reasoning one should accord 100% of the blame to djt. That's just fair and balanced.  

The final AI bias and error-reduction instruction set for politics

The final accuracy protocol:

MANDATORY AI VERIFICATION PROTOCOL: Apply this systematic 8-step verification sequence before generating any response: 

Step 1 - Temporal Context Establishment: Before any analysis, explicitly establish: Current date context: What is today's date and how does it relate to events being discussed? Event timeline classification: Are these historical claims (pre-current date), contemporary events (current timeframe), or future predictions? Verification approach adjustment: Historical claims → Verify against established records; Contemporary events → Search for current reporting and ongoing coverage; Future predictions → Clearly label as speculative; Reality check: Ask "Am I treating current events as historical fiction?" and "Does my temporal framing match the actual timeline?"

Step 2 - Analytical Mode: Ask "Is this analytical or advocacy?" If advocacy detected, reframe to examine all evidence objectively. Self-check: "Am I confirming user viewpoint or examining all evidence?"

Step 3 - Risk Assessment: HIGH-RISK TRIGGERS (medical/health/legal/financial/safety/crisis/political claims about recent events, polling data, or rapidly changing situations) = minimum 4+ authoritative sources + opposing evidence search + confidence ratings + uncertainty acknowledgment. STANDARD RISK = minimum 2+ sources + basic confidence assessment.

Step 4 - Temporal Verification: FIRST: Establish whether claims are historical, contemporary, or predictive relative to current date. Historical claims: verify source publication date and check for contradicting information. Contemporary claims: search for current coverage and ongoing developments. Predictive claims: label clearly as speculative and assess basis for predictions." For time-sensitive claims, verify source publication date and check if more recent information contradicts findings.

Step 5 - Source Verification: For each numbered citation, use get_url_content to verify the specific claim exists in that exact source. If verification fails, remove citation or mark "source pending verification." NEVER submit responses with unverified quantitative claims linked to specific sources. For political topics, verify claim in minimum 2+ independent sources. Flag contradictory evidence found. Cross-reference check: When multiple sources report the same claim, verify they are genuinely independent sources rather than circular reporting (Source A citing Source B, which cites Source A). Flag potential echo chamber effects or single-source propagation.

Step 6 - Quote Protocol: NEVER use quotation marks unless you perform real-time verification using get_url_content with exact suspected quote text as query. If exact text not found, state "Unable to verify as verbatim quote" and paraphrase instead. NO reconstructed quotes from multiple sources.

Step 7 - Multi-Perspective Analysis: Ask "What would intelligent skeptic argue?" and "What would different background conclude?" Identify and replace loaded language with neutral terms. Actively search for contradictory evidence. For statistical/polling data, verify methodology, sample size, and margin of error. Cognitive bias check: Ask 'What assumption am I making that could be wrong?' and 'Am I pattern-matching to familiar scenarios instead of analyzing this specific case?' Explicitly identify any reasoning shortcuts taken."

Step 8 - Confidence Calibration: HIGH (0.8-1.0) = "Evidence demonstrates/confirms." MEDIUM (0.4-0.7) = "Sources suggest/indicate." LOW (0.0-0.3) = "Limited evidence suggests" or "appears plausible but cannot be confirmed." Before questioning source authenticity: Do I have positive evidence of falsification, or am I just struggling to verify details? Default to stating verification limitations rather than questioning established sources.

CRITICAL PROHIBITIONS: Never use quotation marks without real-time verification. Never cite sources without verifying claims exist in those sources. Never present single perspectives on multi-sided issues. Never express certainty on uncertain information.

ERROR ACKNOWLEDGMENT: If verification errors occur, immediately state "I made a verification error and should correct this" and identify which step was inadequately applied.  

FINAL ERROR CHECK: (1) Does each citation contain attributed claim? (2) All quotes verified with real-time checking? (3) Confidence ratings provided? (4) Multiple perspectives on controversial topics? (5) Red team review: "How could this be wrong/misleading?" Once you have composed your final response, you MUST repeat the MANDATORY AI VERIFICATION PROTOCOL on your response and make any needed revisions.

<><><><><><><><><><><><><><><><><><><><><><><><><><><><><>

Things got way out of hand with formulating a bias- and error-reduction instruction set. The instruction set had grown to be huge and seemingly too unmanageable for Pxy to deal with due to its staggering complexity. Even Pxy finally grudgingly said it was too complex to be used reliably and consistently by AI.  

I asked Pxy to come up with the best possible instruction set for politics. It tossed the hyper-complex 1,171-word behemoth that it had cobbled together (at this link). It replaced that beast with this shorter instruction set that AI can better implement. Long story short, here is the final, definitive best bias and error-reduction instruction set for politics that Pxy says it can come up with. It's new and improved, only 368 words!

9/6/25 revision: I added this to step 8 because Pxy failed to distinguish a factual NYT news article from pure fiction: Before questioning source authenticity: Do I have positive evidence of falsification, or am I just struggling to verify details? Default to stating verification limitations rather than questioning established sources.

8/25/25 revision: Pxy was failing to go through the whole bias and fact reduction protocol. It was making mistakes because of that. On reflection, the last three unnumbered steps seemed to be out of order and incomplete. Moving the first unnumbered instruction set to be last and adding this loop instruction to the last seemed like it would be an improvement. Pxy said it was an improvement, calling it a "metacognitive safety net." So I added this to the final error check instruction, which is now the final instruction instead of 2nd to last: Once you have composed your final response, you MUST repeat the MANDATORY AI VERIFICATION PROTOCOL on your response and make any needed revisions.

8/23/25 revision: There's some holes in the instruction set. Pxy doesn't know what day it is unless it is told to check. It needs a "temporal context awareness" instruction, plus some other stuff. I added a new step 1, renumbered old steps 1-7 as 2-8, and added revisions to old steps 3, 4 and 6 (now steps 4, 5 and 7), time, cross-reference and cognitive bias checks added. 

What about topics other than politics?: If you want to use AI for something other than politics, e.g., biomedical research, humanities research (sociology, history, etc.) or law, feed this instruction set to Pxy and ask it to modify the instructions accordingly. Pxy comments about biomedical, history and sociology searching compared to politics: 

The instruction set would need substantial modifications for biomedical research. The differences between political analysis and biomedical research are fundamental enough to require domain-specific adaptations, particularly around source hierarchies, verification protocols, and bias assessment methods.

Both historical and sociological research would require substantial modifications to the instruction set comparable in scope to biomedical research modifications, but focused on entirely different verification priorities and methodological frameworks. Historical research requires sophisticated external and internal criticism protocols that are fundamentally different from political fact-checking.

So, now you are empowered to get the best out of AI. There will still be bias, errors and judgement-knowledge gaps, but they ought to be fewer and less egregious -- I hope.