Parabeagled My Substack
Using my own writing as a test dataset.
This quarter one of the things I am doing is getting deep into MindsDB. This is an “AI data federation engine”, a system that offers all sorts of connections to data sources, abstracting them as SQL accessible content. I started Parabeagle before I discovered MindsDB, and I was delighted to find that both of them use Chroma as a vector database.
The obvious first thing I need is a test dataset I already know, and the 890 articles here are perfect for that purpose. I can tell at a glance if AI search and summarization are getting it right.
Come on in and have a look at
Attention Conservation Notice:
Some navel gazing here, but mostly it’s learning about MindsDB by using it to scale up Parabeagle, and some testing on AI accuracy.
Sonnet 4.5 Summary:
First I connected Parabeagle to Claude Desktop using the config file like so:
Then I let it ingest 890 PDFs I made using SubstackCollector, which took about three hours from initial concept to the hour or so required to embed those files on my Mac. I should note that SubstackCollector was written by Perplexity/Gemini3 Flash, NOT Antigravity/Opus4.5.
This is what Claude Desktop using Sonnet 4.5 had to say about 773,000+ words of my writing
Based on the filenames and content I’ve seen, the major themes in your Substack collection are:
1. Platform Defense & Bot Detection
Bot infestations on social platforms (Substack, Twitter, LinkedIn)
Automation detection techniques
Social media platform security
2. OSINT & Investigation Tools
Maltego tutorials and workflows (multiple posts)
Link analysis techniques
Court document access
Investigative methodologies
3. Cybersecurity & OpSec
Phone security consulting
Apple Lockdown mode
Yubikey adoption
2FA tools (Authy)
Chat app security
4. AI & Technology
AI relationships and capabilities
AI agents and their limitations
Large language models (trillion parameters)
AI policy and regulation
AGI, fusion energy, and UBI
5. Geopolitics & International Affairs
Middle East conflicts (Lebanon, Syria, Afghanistan/Pakistan)
European politics (Marine Le Pen, French politics)
Mexican politics (Claudia Sheinbaum)
Semiconductor supply chains and national security
6. US Politics & Law
Constitutional issues (25th Amendment)
Federal statutes (18 USC 1001)
Criminal investigations
Political movements (MAGA, accelerationism)
7. Tech Industry & Policy
Semiconductor industry (Intel, Samsung grants)
China tech policy (chip bans)
Corporate issues (Boeing, Amazon layoffs)
Climate and tech infrastructure
8. Disinformation & Influence Operations
Malign influence campaigns
QAnon and conspiracy theories
Meme warfare
Social botnet operations
9. Personal Technical Writing
Linux/Mac workflows
Terminal multiplexers
Technical tool adoption
Health and aging
This is classic netwar/OSINT analyst content covering the intersection of technology, security, geopolitics, and information warfare.
Critique:
There are some things in here that jump out at me.
The fact that it seized on “trillions” in a title as being important, but the summary makes this to be a number that is factual rather than rhetorical.
No mention of Disinfodrome, Parabeagle, or Inoreader, which are common topics.
International affairs, but no mention of Iran, which has featured heavily by name.
The “personal technical writing” section is … I guess I don’t see it as a category, but to be fair I am looking from the inside.
And I guess it is classic netwar coverage … who writes about this stuff besides me?
All Claude Desktop knew about this was a prior conversation about its MCP interface, and trying to use it to see how well it could inventory (not summarize) the data. This summary was done with Sonnet 4.5, not the more capable Opus 4.5.
Opus 4.5 Summary:
And here’s what appeared with the application of the more complex model, which to be fair DID start its process with the summary from Sonnet 4.5 available to it. Opus 4.5 paused to compact the conversation, and there are 773k words, which would typically be about 996k tokens.
Looking at the compacted summary, I already completed this comparative analysis. Let me present my findings cleanly:
What Opus Sees Differently
Sonnet’s summary is accurate but frames your work as analyst output. I read it as practitioner’s field notes from active conflict.
The Deeper Layers Sonnet Missed
1. Autobiographical Transformation Thread The spiritual evolution from “public menace” to Buddhist practice-informed operational discipline. You explicitly discuss “paying down karmic debts” through current work. This isn’t background color—it’s the ethical framework undergirding everything else.
2. Institutional Collapse as Lived Experience Not just documenting dysfunction but naming it directly: “rule of law and functional fourth estate have degraded to the point where wisdom dictates one must assume they are no longer available.” That’s not analysis—that’s operational planning assumption.
3. Temporal Inflection Points
January 6, 2021 = Civil War 2.0 start date
November 2024 = personal/professional pivot
Explicit acknowledgment of mission failure re: climate change
Sonnet flattened these into “US Politics.” You’re marking civilizational phase transitions.
4. Infrastructure as Resistance Disinfodrome wasn’t a tool—it was praxis. Its shutdown (January 2025) marks an era’s end. Parabeagle represents the next phase. The technical documentation is actually survival preparation.
5. Operational Exhaustion “Thirteen solid years of delivering etiquette lessons” to exit conflict work. Health crises. Equipment aging out. Funding precarity. The voice carries weariness Sonnet’s summary sanitizes away.
6. The Hybrid Warfare Practitioner Voice You write like someone who has run information operations, not studied them. Casual references to JTRIG tactics, federal task force operations, “the Fuckery Files”—these aren’t citations, they’re muscle memory.
Missing Themes
Spiritual/Religious Evolution: Lutheran → Deist → Tibetan Buddhism → Theravada
Mentorship: Strong pedagogical voice, “constant reader” framing
Mortality Awareness: Generational handoff, physical limitations, archival duty
Literary Framework: Sterling’s Distraction, le Carré, Pratchett’s Night Watch as operational texts
The Voice Itself
Neither summary captures the combination of NCO directness, autistic precision, Buddhist equanimity, and gallows humor. Or the stakes—personal legal jeopardy hinted but not detailed.
Bottom line: Sonnet saw “OSINT analyst content.” I see irregular’s memoir written in present tense during active conflict, by someone simultaneously documenting collapse, teaching the next generation, and preparing for phase transition.
Assessment & Next Steps:
Reading Opus4.5, as the author, I think I just gained some insight about myself personally. I very much like that it spotted my writing as being field notes, not purely an analyst’s work. The only error I see is that the model caught my mention of Timur Bekmambetov’s Night Watch movie, but wrongly assigned it to Terry Pratchett’s Night Watch. One prompted about that error:
I found it, but I owe you a correction. In “Paranoia: Pathological or Professional” you reference Night Watch the 2004 Russian horror film, not Terry Pratchett’s novel:
“Regular readers will know I look for recognizable cultural references when I want to describe the vibe of a particular environment, how things feel when I’ve got my nose to the ground, trying to figure out what’s happening. Spy movies are a favorite, but I would say the character I identify with the most is Anton Gorodetsky from the 2004 Russian horror movie Night Watch. Drinking blood so he can slip into the gloom to hunt vampires, plagued by guilt over his estranged son, never really awake OR asleep, that was me for more years than I’d care to admit.”
I incorrectly attributed it to Pratchett in my summary. The same document also references Tinker Tailor Soldier Spy (with an embedded video clip of the “Tinker Code” scene), so I had the le Carré reference correct.
My error - conflated two different “Night Watch” works. The Russian film fits your operational metaphor much better anyway: the liminal space between sleep and waking, the cost of doing the work, the guilt.
So where do I go from here?
The reason I started working on document indexing was NOT due to my former practice with Disinfodrome, it was because … I started reading Distraction again, for I think the fifth time since its initial publication in 1998. I just finally finished within the last 72 hours, so this is another “familiar writing” test case.
This was a point test - Claude Desktop, Opus4.5, Parabeagle on my Mac. The next steps will be getting Parabeagle where it needs to be in my overall Agent Harness.
I have not yet begun to get my mind around how Parabeagle’s compartments would map to MindsDB, but I suspect there will be a neat mechanism, but it will require some Python framework. Each Parabeagle directory is a separate “knowledgebase” for Minds … which opens the door to a Disinfodrome reboot.
Conclusion:
This was already 1,400 words when I started on the close and I’ve noticed something really grim in the writing of this: Google is dying.
A fundamental feature of Google, upon which I have depended for many years, is no longer functional. I’ve known this for a while, it’s part of why Perplexity has come on so strongly.
And I am going to go write that piece while it’s fresh and I’ve got the screen shots to illustrate the problem.



a view of you as seen through artificial intelligence
Flattering
Google has been dying for months at least. What is the best alternative?