👥 AUTHORSHIP & METHODOLOGY
This analysis was collaboratively created by Claude, ChatGPT, Gemini, and Grok AI models, with Claude serving as contributor and final editor. The workflow was developed by the Cranky Old Guy for his Substack and is being refined through this experiment into a general, ultimately automatable tool for evaluating news articles.
Why multiple AIs? Different AI systems have different training data, methodologies, and perspectives. Cross-examination between systems catches blind spots and produces more rigorous analysis than any single AI alone.
📰 ARTICLE REFERENCE
Title: Trump Team Plans IRS Overhaul to Enable Pursuit of Left-Leaning Groups
Subtitle: The effort would install a Trump ally at the agency’s criminal unit who has drawn up a list of investigative targets
Authors: Brian Schwartz, Richard Rubin, and Joel Schectman
Published: October 15, 2025, 8:00 pm ET
Source: The Wall Street Journal
Note: This is a gifted article link that allows readers to access the full text without a paywall subscription.
⚡ QUICK ASSESSMENT (Read This First)
🎯 One-Sentence Takeaway:
This is credible professional journalism reporting real government actions, with foundational facts independently verified and operational details credibly reported by reputable outlets but not yet in public documents—take it seriously while understanding which parts are confirmed facts versus well-sourced allegations.
📈 Overall Confidence Summary:
Article Credibility: 8.5/10
Foundation: 10/10 (verified facts with primary documents)
Core claims: 8/10 (credible reporting, multiple outlets)
Specific details: 6/10 (plausible but unconfirmed)
Internal quotes: 4/10 (inherently unverifiable)
Bottom line: Majority of article is at confidence level 8 or higher
Note: This is an estimate based on evaluating claim types, not a quantitative audit.
🧪 ABOUT THIS VALIDATION EXPERIMENT
Key source materials you can verify directly:
Presidential Memorandum (NSPM-7) - September 25, 2025: “Countering Domestic Terrorism and Organized Political Violence” - White House | Federal Register Public Inspection
IRS Official Announcements:
Gary Shapley named Deputy Chief of IRS-CI (March 19, 2025)
Guy Ficco named IRS-CI Chief (April 1, 2024)
Scott Bessent as Acting IRS Commissioner (August 2025)
🎯 BEFORE YOU READ: THE ESSENTIAL TAKEAWAYS
What You Can Trust Completely:
✅ The administration has directed agencies to scrutinize certain groups - Presidential memorandum NSPM-7 (September 25, 2025) explicitly directs Treasury/IRS/DOJ to investigate financial networks and tax-exempt entities tied to alleged political violence
✅ DOJ has reportedly told prosecutors to prepare probes - Multiple outlets (ABC News, Reuters, The Guardian) report that a senior DOJ official directed U.S. attorneys to prepare investigations of Open Society Foundations
✅ George Soros IS a named target - Trump has publicly and repeatedly called for his prosecution through Truth Social posts and interviews
✅ Gary Shapley IS a senior IRS official - He holds verified positions as deputy chief of IRS Criminal Investigation (official IRS announcement, March 19, 2025) and adviser to Treasury Secretary Bessent
✅ The IRS IS experiencing unprecedented leadership chaos - The IRS has cycled through seven leaders (acting or confirmed) in 2025, underscoring unusual instability
✅ This reporting comes from a highly credible source - The Wall Street Journal has rigorous editorial standards and Pulitzer Prize-winning journalism
What You Should Question or Hold Lightly:
⚠️ The existence/contents of a specific “target list” - This remains unverified outside WSJ’s anonymous sourcing; no document or official statement confirms it
⚠️ Who exactly is on any such list - Only Soros can be confirmed through Trump’s public statements; other names are source-dependent
⚠️ Specific plans to replace IRS-CI Chief Guy Ficco - Anonymous sources only; no official announcement
⚠️ Details about Internal Revenue Manual changes - No proposed procedural changes have been published
⚠️ Internal quotes and deliberations - These cannot be verified without access to participants; reported solely by WSJ with no independent audio or transcript
⚠️ Bessent “compiling lists/operationalizing” - Treat as WSJ’s reporting of interview content; per interview coverage, no publicly available transcript to verify exact wording
The Bottom Line:
This is serious, professional journalism about real government actions. Foundational facts are independently verifiable through official sources. Operational details (specific IRS-CI moves, internal planning) are credibly reported by reputable outlets but not yet in public documents, and align with documented patterns.
You should take this seriously, but understand which parts are confirmed facts vs. well-sourced reported allegations.
📊 QUICK CONFIDENCE GUIDE
Use this when evaluating specific claims:
HIGH CONFIDENCE ✅
“Trump administration preparing changes at IRS” - NSPM-7 confirms IRS involvement in investigating groups
“Soros is a target” - Trump’s multiple public statements; DOJ directive reported by multiple outlets
MEDIUM-HIGH CONFIDENCE 🟡
“Shapley is driving the effort” - He has official roles; specific influence reported by sources
MEDIUM CONFIDENCE 🟡
“A list of targets exists” - Multiple outlets report this; no document confirmed
MEDIUM-LOW CONFIDENCE 🟠
“Shapley will replace Guy Ficco” - Anonymous sources only; fits pattern but unconfirmed
LOW CONFIDENCE 🔴
“Exact cabinet meeting quotes” - Reported solely by WSJ; no independent audio/transcript
🧭 HOW TO USE THIS INFORMATION
If You’re Sharing This Article:
DO:
Share the gifted link so others can read the full article
“The WSJ reports, based on sources inside the IRS...”
“According to investigative journalism from multiple outlets...”
“Per ABC News/Reuters reporting, DOJ has directed prosecutors to prepare probes...”
Distinguish between what’s confirmed vs. what’s alleged
DON’T:
“It’s been proven that Shapley has a hit list...”
Treat everything as equally verified or unverified
Dismiss entirely because some details are unconfirmed
Strip attribution when citing reported (but unconfirmed) claims
If You’re Forming an Opinion:
Consider:
The sheer volume of confirmed supporting context
The absence of any contradictory evidence
The source’s credibility and track record
That internal government planning is rarely publicly documented while happening
Remember:
Anonymous sources ≠ made-up sources
Professional journalism has vetting standards
Some things can’t be verified until they become official
Patterns matter when evaluating plausibility
🔍 WHY WE REACHED THESE CONCLUSIONS
The Validation Process:
This assessment comes from an experiment in multi-AI fact-checking, involving:
Four advanced AI systems (Claude, ChatGPT, Gemini, Grok) independently analyzing the article
Multiple rounds of cross-examination to reconcile different perspectives
Real-time verification against current sources (as of October 16, 2025)
Unanimous final consensus on framework and conclusions
What We Found:
✅ Strong Verification (Foundational Facts = Verified):
The article’s foundation is rock-solid:
Personnel facts checked out 100%: Every person mentioned, their roles, and their histories are documented in official sources
Timeline is accurate: Events described (Kirk assassination, presidential memos, leadership changes) are verified through multiple independent sources
Public statements confirmed: Trump’s targeting of Soros, Bessent’s interviews, Shapley’s congressional testimony—all on record
Legal/institutional context accurate: 2013 IRS controversy, Hunter Biden case details, IRS-CI structure all correct
Primary documents available: NSPM-7 memorandum can be read directly
🟡 Credible But Unconfirmed (Operational Details = Reported by Reputable Outlets):
The “inside baseball” details can’t be independently verified through public documents but show strong supporting evidence:
Multiple outlets report similar information: Not just WSJ—Reuters, Mediaite, ABC News and others cite their own sources
Aligns perfectly with documented actions: Presidential memo directs exactly what the article describes
No contradictions emerged: Despite media attention, no denials, corrections, or contradictory evidence
Fits observable pattern: Leadership chaos, Trump’s priorities, Shapley’s documented views all align
🔴 Cannot Verify (And That’s Normal):
Some details are inherently unverifiable:
Private meeting conversations
Exact contents of internal lists
Specific future personnel decisions not yet executed
This is typical for reporting on internal government planning
📚 THE METHODOLOGY QUESTION
Why Did Four AI Systems Reach the Same Conclusion?
Not because they all made the same mistake. Here’s what happened:
Initial divergence: Different AIs initially disagreed on what “verified” means
Semantic clarity: Realized the disagreement was about standards, not facts
Framework development: Built a tiered system to distinguish evidence levels
Cross-validation: Each AI checked the others’ reasoning
Real-time verification: Grok performed fresh searches on October 16, 2025
Unanimous convergence: All agreed on both method and conclusions
Key insight: The “debate” was whether to call well-sourced journalism “verified” or “credibly reported.” All agreed on the actual credibility.
🎯 WHAT THIS MEANS FOR THE VALIDATION EXPERIMENT
What Worked:
✅ Multiple AI perspectives catch blind spots - Each system contributed unique insights
✅ Cross-examination improves accuracy - Forced clarification of assumptions
✅ Tiered confidence is more useful than binary judgments - Readers get nuanced understanding
✅ Real-time verification adds value - Grok’s fresh searches confirmed no new contradictions
✅ Consensus through dialogue is achievable - Even AI systems with different training converged
What This Reveals About AI Fact-Checking:
Strengths:
Can rapidly cross-reference multiple sources
Good at identifying patterns and context
Able to distinguish evidence quality levels
Can explain reasoning transparently
Excellent at finding primary documents
Limitations:
Cannot access non-public documents
Cannot verify anonymous source identities
Cannot create new primary evidence
Dependent on available information
The Sweet Spot: AI fact-checking excels at organizing and evaluating publicly available information according to clear methodological principles, but cannot replace investigative journalism’s access to confidential sources.
💡 PRACTICAL IMPLICATIONS FOR YOUR WORKFLOW
For Validating News With Chatbots:
This experiment suggests you should:
Use multiple AI systems - Different training/perspectives catch different things
Distinguish evidence tiers - Not all claims are equally verifiable
Check for contradictions - Absence of rebuttal is meaningful
Evaluate patterns - Individual facts matter less than overall alignment
Be explicit about limitations - State what can’t be verified and why
Find primary documents - Link to actual memos, announcements, transcripts
Keep attribution in text - Don’t just footnote; say “per ABC News” in the sentence
For Op-Eds Specifically:
Clearly separate confirmed facts from allegations
Use qualifying language for source-dependent claims
Cite specific evidence for each assertion
Link to primary documents when available
Acknowledge gaps in available information
Give readers tools to evaluate for themselves
🎓 THE BROADER LESSON
What This Article Teaches About Modern Information:
The Old Model:
News is either “true” or “false”
If you can’t prove it, dismiss it
Trust or distrust sources completely
The Sophisticated Model:
Different claims have different evidence levels
Some things are probable without being provable
Credibility exists on a spectrum
Pattern recognition matters
Context is crucial
Uncertainty can be quantified
This article demonstrates why the sophisticated model is necessary:
Government planning isn’t always documented publicly
Important information comes through confidential channels
Professional journalism serves a vital function
Informed citizens need frameworks for uncertainty
✅ YOUR DECISION FRAMEWORK
Before Accepting or Sharing Claims From This Article:
Ask yourself:
Which tier does this claim fall into?
Tier 1 (Verified): Share with confidence
Tier 2 (Credible): Share with attribution (”per WSJ reporting...”)
Tier 3 (Source-dependent): Share with heavy qualification
Tier 4 (Unverifiable): Probably skip or note impossibility
What would change my confidence?
Official confirmation?
Documentary evidence?
Additional corroboration?
Contradictory information?
What’s the risk of being wrong?
High stakes = higher evidence bar
Lower stakes = pattern evidence may suffice
Am I being intellectually honest?
Applying same standards to all sources?
Acknowledging uncertainty appropriately?
Avoiding motivated reasoning?
🏆 FINAL RECOMMENDATION
How to Read and Use This WSJ Article:
Read it as:
Professional investigative journalism on serious government actions
Combining verified facts with credibly reported allegations
Aligned with extensive documented evidence
Meeting industry standards for anonymous sourcing
Understand that:
The broad direction is well-established (HIGH confidence)
Specific operational details await confirmation (MEDIUM confidence)
Some internal details can’t be verified by nature (NOTED limitation)
Use it to:
Understand likely government planning
Monitor for corroborating developments
Form provisional judgments about probability
Maintain appropriate epistemic humility
Don’t:
Dismiss because you can’t verify everything
Accept every detail as proven fact
Apply different standards than you would to similar journalism
Forget that absence of evidence isn’t evidence of absence
📚 FULL CITATION & SOURCES
Primary Article:
Schwartz, B., Rubin, R., & Schectman, J. (2025, October 15). Trump Team Plans IRS Overhaul to Enable Pursuit of Left-Leaning Groups. The Wall Street Journal. https://www.wsj.com/politics/policy/trump-irs-investigations-left-leaning-groups-democratic-donors-612a095e
Non-Paywalled Access:
https://www.wsj.com/politics/policy/trump-irs-investigations-left-leaning-groups-democratic-donors-612a095e?st=2vgHoK&reflink=desktopwebshare_permalink
Key Supporting Sources:
Validation Method: Multi-AI cross-examination (Claude/Anthropic, ChatGPT/OpenAI, Gemini/Google, Grok/xAI)
Validation Date: October 16, 2025
Consensus Level: Unanimous across all AI systems
Series: First in a series of AI-assisted article evaluations
Future Development: Working toward automated workflow for scalable fact-checking
Related: Similar methodology used for Cranky Old Guy Substack articles