Prosecute Fauci: State level approaches

A practical, state-focused strategy to pursue criminal prosecution of Anthony Fauci for alleged actions during the COVID-19 pandemic. Presidential pardons cover only federal offenses, not state crimes. Multiple states have already recognized this and begun laying groundwork for accountability under their own laws (e.g., statutes addressing reckless endangerment, manslaughter by negligence, fraud on the public, or even broader public-health endangerment provisions that some states tie to mass harm).

No role-based immunity argument is needed or invoked here—the path forward relies on state sovereign authority to enforce their criminal codes.

Step 1: Target the Right Jurisdictions (The 17+ States Already Mobilized)Seventeen Republican attorneys general (led by South Carolina AG Alan Wilson, including Florida, Texas, Missouri, Tennessee, Alabama, and others) formally requested congressional evidence in February 2025 specifically to evaluate state-level charges against Fauci for misleading statements, lab-leak suppression, and mismanagement. Oklahoma state Rep. Justin Humphrey submitted probable-cause statements to his AG in January 2025 citing COVID protocols as potential crimes. Idaho passed a resolution affirming that a federal pardon confers zero protection from state prosecution.Action items:

• Identify your state or a friendly one from this coalition (full list publicly available via state AG offices or the South Carolina AG’s February 2025 letter to congressional leadership).
• File a detailed criminal complaint or affidavit directly with that state AG or district attorney. Many states allow private citizens or legislators to submit evidence triggering review (e.g., Oklahoma’s process).
• Reference the existing coalition letter and any state resolutions—these create political and legal momentum so your filing is not starting from zero.

Step 2: Build a Tight “Bill of Particulars” Mapped to State Criminal StatutesUse publicly available congressional testimony, emails released by the Select Subcommittee on the Coronavirus Pandemic, FOIA documents, and gain-of-function funding records. Frame them as specific violations of state law (not vague “policy disagreements”). Key categories drawn from documented public records (adapt to your state’s codes):

• Funding and continuation of gain-of-function research at Wuhan despite U.S. moratoriums and known safety risks.
• Public suppression of alternative treatments (hydroxychloroquine, ivermectin) while fast-tracking others with known issues.
• Statements to Congress and the public that allegedly contradicted internal knowledge (e.g., origins, mask efficacy, vaccine transmission).
• Coordination that allegedly caused foreseeable harm through lockdowns, school closures, and mandates.

Attach exhibits: specific emails, hearing transcripts, and expert affidavits linking decisions to deaths or harms in that state (excess mortality data, VAERS reports localized, economic damage studies). This turns the complaint into probable cause for investigation or grand jury.

Step 3: Amplify Through Legislatures and Parallel Pressure

• Have state legislators (model after Oklahoma and Idaho) introduce resolutions demanding the AG open a full probe and/or convene a special grand jury.
• Coordinate with existing groups already feeding evidence to the 17-AG coalition (they explicitly asked Congress for more material to enable state action).
• Submit the same package to sympathetic members of Congress (e.g., those on the Select Subcommittee) so federal findings flow directly to state AGs.

Step 4: Timeline and Escalation

• Immediate (0–30 days): File complaints in multiple coalition states simultaneously. Publicly release the filings (press conference, Substack, X) to create national pressure.
• 30–90 days: Follow up with FOIA-style demands for any state-held records and request meetings with AG staff.
• 90+ days: If stalled, push for special legislative hearings or independent counsel statutes (some states have them for public-official misconduct).
• Statute of limitations: Many relevant state crimes (especially those tied to death or fraud) have long or no limitations periods.

Additional Levers

• Civil-to-criminal pipeline: Parallel civil suits by harmed individuals or businesses can generate discovery that strengthens criminal referrals.
• International angle (secondary): Some countries have universal-jurisdiction statutes for alleged global public-health crimes, but state-level U.S. action is far more viable.
• Political reinforcement: Support or primary candidates for state AG and governor in 2026–2028 cycles who explicitly commit to moving forward. The 2025 coalition shows the infrastructure already exists.

This is not speculation—it mirrors exactly what state AGs and legislators have already started. Success depends on volume of credible filings, relentless follow-up, and mapping every allegation to a specific state criminal code violation with evidence of intent and harm. Gather the documents, file in the right states, and keep the pressure public and professional. The federal pardon is irrelevant at the state level; the door is open where officials choose to walk through it.

Prosecute Fauci

Medicaid Provider Claims Outliers

Below, I will include a link to a pre-processed dataset of the Medicaid Provider Claims data released by HHS the week of 2/14/2026. The original, raw dataset has 227 million payment rows across 267385 BILLING_PROVIDER_NPI_NUM providers.

To identify outliers, Grok helped me build a python program which flags outliers in the categories:

total_paid, paid_per_beneficiary, paid_per_claim, unique_beneficiaries

This reduces the total dataset to about 38 million rows.

Outliers are determined by exceeding an upper bound on the peer population averages by HCPCS code by an IQR method (“inter quantile ratio”):

        q1 = series.quantile(0.25)
        q3 = series.quantile(0.75)
        iqr = q3 - q1
        return q3 + 1.5 * iqr

medicaid-provider-outliers_2026-02-14Download

The python code to identify the outliers depends on the original dataset being sorted so that the process doesn’t run out of memory. This process ran for about 3 hours on my Macbook Pro M2.
I will include that code below:

import pandas as pd
import argparse
from tqdm import tqdm
import csv
def safe_ratio(numerator, denominator):
    """Avoid division by zero → return NaN"""
    return numerator / denominator if denominator != 0 else float('nan')
def detect_outliers(group_rows):
    """
    Input: list of dicts (rows for one HCPCS + month group)
    Returns: list of outlier row dicts (original rows + added OUTLIER_REASONS column)
    Flags if extreme in: total_paid, paid_per_benef, paid_per_claim, or unique_beneficiaries
    """
    if len(group_rows) < 4:
        return []
    # Convert to DataFrame
    df = pd.DataFrame(group_rows)
    # Force numeric columns (handle junk strings, empty values, etc.)
    numeric_cols = ['TOTAL_PAID', 'TOTAL_UNIQUE_BENEFICIARIES', 'TOTAL_CLAIMS']
    for col in numeric_cols:
        df[col] = pd.to_numeric(df[col], errors='coerce')
    # Drop completely invalid rows for stats
    df = df.dropna(subset=numeric_cols)
    if len(df) < 4:
        return []
    # Compute derived metrics (only for comparison)
    df['PAID_PER_BENEF'] = df.apply(
        lambda r: safe_ratio(r['TOTAL_PAID'], r['TOTAL_UNIQUE_BENEFICIARIES']), axis=1
    )
    df['PAID_PER_CLAIM'] = df.apply(
        lambda r: safe_ratio(r['TOTAL_PAID'], r['TOTAL_CLAIMS']), axis=1
    )
    # Drop rows invalid for ratio-based metrics
    df_clean = df.dropna(subset=['PAID_PER_BENEF', 'PAID_PER_CLAIM'])
    if len(df_clean) < 4:
        return []
    # ────────────────────────────────────────────────
    # IQR upper bound function (only upper outliers)
    # ────────────────────────────────────────────────
    def get_upper_bound(series):
        if len(series) < 4:
            return float('inf')  # skip tiny groups
        q1 = series.quantile(0.25)
        q3 = series.quantile(0.75)
        iqr = q3 - q1
        return q3 + 1.5 * iqr
    # Compute thresholds for all four metrics
    bounds = {
        'high_total_paid': get_upper_bound(df_clean['TOTAL_PAID']),
        'high_paid_per_benef': get_upper_bound(df_clean['PAID_PER_BENEF']),
        'high_paid_per_claim': get_upper_bound(df_clean['PAID_PER_CLAIM']),
        'high_unique_benef': get_upper_bound(df_clean['TOTAL_UNIQUE_BENEFICIARIES']),
    }
    # Identify outliers using the cleaned df indices, but return original dicts
    flagged_rows = []
    for idx in df.index:
        row = df.loc[idx]
        flags = []
        # Raw total paid
        if pd.notna(row['TOTAL_PAID']) and row['TOTAL_PAID'] > bounds['high_total_paid']:
            flags.append('high_total_paid')
        # Paid per beneficiary
        if pd.notna(row['PAID_PER_BENEF']) and row['PAID_PER_BENEF'] > bounds['high_paid_per_benef']:
            flags.append('high_paid_per_benef')
        # Paid per claim
        if pd.notna(row['PAID_PER_CLAIM']) and row['PAID_PER_CLAIM'] > bounds['high_paid_per_claim']:
            flags.append('high_paid_per_claim')
        # High number of unique beneficiaries
        if pd.notna(row['TOTAL_UNIQUE_BENEFICIARIES']) and row['TOTAL_UNIQUE_BENEFICIARIES'] > bounds['high_unique_benef']:
            flags.append('high_unique_beneficiaries')
        if flags:
            original_row = group_rows[idx]  # preserve original string formatting etc.
            original_row['OUTLIER_REASONS'] = ','.join(flags)
            flagged_rows.append(original_row)
    return flagged_rows
def main(csv_file, output_file='outlier_billing_providers_multi.csv'):
    print(f"Streaming {csv_file} (expects sorted by HCPCS_CODE, CLAIM_FROM_MONTH)...")
    print("Flagging upper outliers on: total_paid, paid_per_beneficiary, paid_per_claim, unique_beneficiaries")
    outlier_rows = []
    current_group_rows = []
    current_key = None
    total_outliers = 0
    total_rows_processed = 0
    with open(csv_file, 'r', newline='', encoding='utf-8') as f:
        reader = csv.DictReader(f)
        # Add our new column to output
        fieldnames = reader.fieldnames + ['OUTLIER_REASONS']
        for row in tqdm(reader, desc="Streaming rows", unit="row"):
            total_rows_processed += 1
            key = (row['HCPCS_CODE'], row['CLAIM_FROM_MONTH'])
            if key != current_key and current_group_rows:
                outliers = detect_outliers(current_group_rows)
                if outliers:
                    outlier_rows.extend(outliers)
                    total_outliers += len(outliers)
                current_group_rows = []
            current_key = key
            current_group_rows.append(row)
        # Process final group
        if current_group_rows:
            outliers = detect_outliers(current_group_rows)
            if outliers:
                outlier_rows.extend(outliers)
                total_outliers += len(outliers)
    print(f"\nFinished. Processed {total_rows_processed:,} rows.")
    if total_outliers == 0:
        print("No outliers detected in any metric.")
        return
    # Write results
    with open(output_file, 'w', newline='', encoding='utf-8') as out:
        writer = csv.DictWriter(out, fieldnames=fieldnames)
        writer.writeheader()
        writer.writerows(outlier_rows)
    print(f"Found and saved {total_outliers:,} outlier rows → {output_file}")
    print("Each outlier row includes 'OUTLIER_REASONS' column listing which metric(s) triggered the flag.")
if __name__ == "__main__":
    parser = argparse.ArgumentParser(
        description="Streaming outlier detection on Medicaid claims (sorted CSV). "
                    "Flags upper outliers in four metrics relative to same HCPCS + month peers."
    )
    parser.add_argument('csv_file', help="Path to input CSV sorted by HCPCS_CODE, then CLAIM_FROM_MONTH")
    parser.add_argument('--output', default='outlier_billing_providers_multi.csv',
                        help="Output CSV path (default: outlier_billing_providers_multi.csv)")
    args = parser.parse_args()
    main(args.csv_file, args.output)

The Pfizer Papers Index

In October 2024, Naomi Wolf (and Amy Kelly) published a book titled “Pfizer Papers: Pfizer’s Crimes Against Humanity” which summarizes their massive undertaking to analyze the Pfizer’s mRNA COVID-19 vaccine clinical trial data.

Their conclusions? That the data was deeply flawed and that the pharmaceutical company knew by November 2020 that its vaccine was neither safe nor effective.

This work was completed with thousands of volunteer specialists and scientists around the world which analyzed over 2700 Pfizer clinical trial documents.

This is my small contribution to the broader community.

I have downloaded these documents from the source site (https://phmpt.org/pfizer-12-15-documents/) using an LLM built python script. These represent about 6 Gb of documents and data… pdf files, word documents, excel spreadsheets, images and even the audio track for a Pfizer Covid vaccine advertisement.

Then, also using the magic of LLMs I also have built a sophisticated summarizer program in python to use a LLM model to analyze and summarize these documents to create a master index with topic and document type labels as well as smart summaries of the contents and estimates of the content dates.

I will include both of these scripts for curious users.

Below, I will present the resulting master index to the Pfizer papers that this process has created in a csv format for the use of other who might be interested in exploring this data.

(Work in Progress….)

Update (2026-01-30)

The summarizer has done a good job, but took about 10 days to process these 2700 files on my Macbook Pro M2 with 16 Gb of memory

Here is a short description of how the program processes the documents:

This program is a specialized batch summarization tool designed primarily for biomedical, clinical, regulatory, and pharmaceutical documents.

Core purpose in one sentence

It automatically extracts text from various document formats, intelligently samples the most important parts of long documents, uses a local LLM (Llama-3.1-8B-Instruct in the default configuration) to create section summaries, then synthesizes them into one coherent, high-quality 180–250 word summary per document — while trying to preserve the most medically/regulatorily important information.

In short

Local LLM-powered batch summarizer optimized for FDA/regulatory/clinical/pharmacovigilance documents with automatic type detection + type-aware intelligent sampling + OCR support for scanned PDFs.

Updates

• 2/1/2026 — starting to get into the clinical trial documents… these should have some interesting content about trial participants
  
• 2/9/2026 — the summarization process completed after about 10 days. I will attach an Excel Workbook with the summary information (filenames, topics, document types, …) on each row as well as a zip file with the json format summaries for each document.
  

Pfizer_FOIA_master_indexDownload

Pfizer_FOIA_Summaries_jsonDownload

Summary of the Pfizer Papers pdf document

I also did run the data summary pdf document from the book “The Pfizer Papers” through Grok to summarize it. ~ 300 pages of charts (about 300 Mb in original size). Grok condensed many of the main points to about 4 pages. (SEE BELOW.)

“The Pfizer Papers: Pfizer’s Crimes Against Humanity”

Overview

“The Pfizer Papers: Pfizer’s Crimes Against Humanity” is a 295-page book compiled by analysts from the WarRoom/DailyClout Pfizer Documents Analysis Project, edited by Naomi Wolf and Amy Kelly, with a foreword by Stephen K. Bannon. Published by Skyhorse Publishing, it presents a critical examination of Pfizer’s COVID-19 mRNA vaccine (BNT162b2) based on court-ordered releases of internal documents, particularly the 5.3.6 post-marketing experience report covering the first 90 days of public rollout (December 1, 2020, to February 28, 2021). The book aggregates volunteer-led “micro-reports” analyzing adverse events (AEs), safety signals, clinical trial data, VAERS reports, autopsies, and global trends like birth rate declines. It accuses Pfizer, the FDA, and CDC of negligence, data manipulation, and failing to address harms, framing the vaccine rollout as a “crime against humanity” and calling for its recall.

The document lacks a formal table of contents or chapter structure but is organized as a series of thematic reports, charts, case studies, and appendices. It draws from Pfizer’s own data (e.g., 42,086 case reports with 158,893 AEs), highlighting inconsistencies, underreporting, and dismissed safety concerns. Key themes include organ-specific damages, deaths, unresolved outcomes, vaccine failures, and potential links to broader health crises.

Key Sections and Themes

The book is divided into micro-reports (e.g., Post-Marketing Team Micro-Reports 4–19), each focusing on a System Organ Class (SOC) or Adverse Events of Special Interest (AESIs) from Pfizer’s 5.3.6 document. These are interspersed with charts, forensic analyses, case reports, and discussions. Below is a structured summary based on the content across all 295 pages:

1. Introduction and Contextual Analysis (Pages 1–3, Scattered Throughout)

• Title and Setup: Introduces the project as volunteer-driven analyses of Pfizer documents released via Public Health and Medical Professionals for Transparency (PHMPT.org).
• Hepatic (Liver) SOC Review (Micro-Report 4): 70 cases, 84 AEs (e.g., elevated liver enzymes, hepatic pain, ascites, jaundice). 5 deaths (7%). Half of AEs within 3 days. Criticizes Pfizer’s focus on lab abnormalities rather than diseases, lack of follow-up (50% unknown outcomes), and dismissive conclusion: “No new safety issues.” Argues for higher fatality rates and demands recall.
• Overarching Critique: Pfizer’s 5.3.6 summary claims a “favorable benefit-risk profile,” but the book counters with evidence of ignored signals. Notes no post-February 2021 public data releases despite FDA mandates.

2. Immunological and Antibody Shifts (Pages 4–5)

• Figures on IgG immunoglobulins: Show dose-related shifts (e.g., gain in IgG4 with more LNP/mRNA doses), suggesting immune dysregulation. Links to potential autoimmunity or reduced efficacy.

3. VAERS and Global Trends (Pages 6–12, 20, Scattered)

• VAERS Reports (Chart 1): Analyzes 12/14/2020–7/29/2022 data, highlighting underreporting factors (URF estimates of 50+).
• Birth Rate Declines: Detailed charts (3–12, 18, 20) show drops in births (e.g., 2008–2022 in various countries like Australia, Taiwan, Europe, Switzerland). Correlates with vaccination timelines (e.g., 9-month lag post-rollout). Uses Spearman’s rho for non-normal data; p-values indicate significance. Concludes unprecedented slumps (e.g., Switzerland’s 150-year history shows no prior equivalent).
• Methods Discussion: Explains statistical choices (e.g., no normal distribution assumption) and references critics like Ioannidis on significance thresholds.

4. Cardiovascular and Related SOCs (Pages 30, 153–154, 177, 189, 281, 284)

• Cardiovascular AESIs (Micro-Report on SOC): 1,403 cases (69% serious), excluding myocarditis/pericarditis (reported separately). 140 deaths (10%). Predominantly female (69%). Latency: <24 hours to 51 days. Calls for investigation and recall.
• Clotting/Blood System: Links to vasculitis, microthrombi, and multi-organ inflammation.
• Respiratory SOC: 130 cases (92% serious), 51 deaths (39%). Includes ARDS, respiratory failure.

5. Pediatric and Special Populations (Pages 31–32, 157–158, 227–230)

• Pediatric (<12 Years) (Micro-Report 6): 34 cases (3% of total). 4 deaths, including infants. Issues like seizures, kidney injury. Criticizes lack of age-specific safety data and FDA approvals despite signals.
• Pregnancy and Lactation (Micro-Report 17): 274 pregnancies, 28 fetal/neonatal deaths. 88% unknown outcomes. Breastfed infants: Suppressed lactation, illness. Compares to Cumulative Report (Report 69), noting broader data inclusion but similar gaps. Concludes no safety signals per Pfizer, but book demands recall.
• Clinical Trials Timeline: Summarizes studies like C4591001 (ages 12–15, under 12), noting delayed or unavailable results.

6. Neurological and Musculoskeletal SOCs (Pages 64–65, 106–107, 149–150)

• Neurologic (Micro-Report 7): 501 cases (75% serious), 16 deaths. Includes seizures (75%), Guillain-Barré. Excludes strokes. Latency: <24 hours to 51 days.
• Facial Paralysis (Micro-Report 8): 449 cases (51% serious). One infant case. Half within 2 days.
• Musculoskeletal (Micro-Report 14): 3,600 cases (8.5% of data). Includes arthritis, joint pain.
• All conclude with ignored signals and recall calls.

7. Immune and Allergic Reactions (Pages 119–122, 131, 137, 140, 145)

• Anaphylaxis (Micro-Report 10): 2,958 cases (79% serious), female predominant.
• Immune-Mediated/Autoimmune (Micro-Report 11): 1,050 cases (55% hypersensitivities). Includes neuropathies.
• Vasculitis (Micro-Report 12): 32 cases (75% serious), 1 death.
• Case series on post-vaccination diseases (e.g., lupus, myocarditis).

8. Other AESIs and Forensic Analyses (Pages 118, 123–124, 139, 150–151, 212, 230–231, 243, 285–286, 291, 294–295)

• Renal (Kidney) (Micro-Report 9): 69 cases, 23 deaths (33%).
• Other AESIs (Micro-Report 13): 270 cases, 96 deaths (7.8% of total deaths).
• Vaccine Effectiveness (Micro-Report 18): 3,067 failures, 65 deaths (3.9%).
• COVID-19 AESIs (Micro-Report 19): 3,738 cases, potential VAED/VAERD (75 severe). Criticizes Pfizer’s denial.
• Autopsies and Deaths: Forensic review of 38 trial deaths (e.g., timing discrepancies, underreported cardiovascular causes). Burkhardt collection shows spike protein in organs.
• Case Reports: Sudden deaths, multi-organ failures post-vaccination.

9. Appendices and Tables (Pages 153–154, 174, 281, 284)

• SOC summaries: Breakdowns by seriousness (e.g., 9,400 serious AEs in pregnancy/lactation).
• VAERS tables: Pediatric myocarditis, deaths.
• References: Links to PHMPT.org, studies on underreporting.

Overall Conclusions and Critiques

• Pfizer’s Repeated Refrain: Every SOC ends with "This cumulative case review does not raise new safety issues. Surveillance will continue." The book calls this dismissive, given high deaths (e.g., 1,223 total), unknowns (50%+), and short latencies.
• Systemic Issues: Accuses data dilution (e.g., separating related SOCs), no follow-up, undercounting (e.g., missing common terms), and FDA inaction on surveillance.
• Broader Implications: Links vaccine to global harms (e.g., birth declines, excess deaths). Demands full pediatric documents, investigations, and recall.
• Tone and Call to Action: Polemical, with phrases like "What does it take?" and "RECALL this unsafe ‘vaccine.’" Emphasizes ethical failures and potential criminality.

The book serves as an advocacy piece, using Pfizer’s data against it, and urges accountability. It’s not neutral but substantiated with direct quotes, tables, and analyses from the source documents.

No Vaccine Mandates

No employer, college, university, hospital or healthcare system should mandate vaccines.

In the document below, I will discuss the ethics and provide some of the available evidence against the flu vaccine and the Covid vaccine and their mandates.

No Vaccine Mandates_2023-09-22Download

Acute vs. Chronic HCC RAF

Summary Information from the Clinical Record

HCC_Summary <- function( pdftext )

LabMatch <- function( labstring, pdftext, ... )

ReportMatch = function( pdftext, reportdelims, … )

_Sample_Summary3Download

Extracting Patient Health Summary Information from the Continuity of Care Document (CCD)

HCC_Summary <- function( pdftext ) {

x <- getICD10( pdftext ) # returns a list of ICD10 codes found in the text

LabMatch <- function( labstring, pdftext ) {

sample_summary0-1Download

Summary of Medicare Payments to Ohio Internal Medicine Physicians

By quartile of total revenue and category of services.  Same analysis as applied to Family Physicians.  Internal Medicine physicians tend to perform a wider range of different types of services.

Summary of Medicare Payments to Family Physicians in Ohio

This graph breaks down the average billing category revenue amounts by Quartile of Total Revenues for Family Physicians in Ohio for 2012.  (This is not a “per provider” average, but an average revenue earned by providers providing that service in that category in that quartile.)

There are several interesting points:

1. Family Physicians in the higher Medicare Revenue bracket generally provide more patient care:
   • More Office visits
   • More Nursing Home visits
   • More Home Health visits
   • More Hospital and ER services
2. There are also some interesting outliers:
   • PathLab: DrugTst
     • A single physician provider providing Drug Confirmation Testing
   • MedP: OthrSvcs
     • A small number of physicians are providing Hyperbaric Oxygen Therapy
   • MedP: Neurol
     • About 34 physicians providing neuromuscular testing and sleep study services

I plan to refresh this analysis and look at average category revenues by quartile on a “per provider” basis.

This analysis was done in R & RStudio on an Ubuntu Linux platform using linked PostgreSQL to subset the 2012 Medicare Payments data by physician specialty and location.  A pattern match selection using regular expressions was used to supercategorize the CPT codes into broader categories (Critical Care, Home Health, Hospital Inpatient, …).

ggplot2 was used to generate the graphics.

Thank you to the Coursera Johns Hopkins Data Science Specialization series, the R & RStudio as well as the PostgreSQL communities for their great open source tools and guidance.

Excess Deaths in Selected Counties in Ohio during Covid

Brad Banko, MD, MS

3/12/2026

This mortality data was extracted from the DataOhio website by Death Year-Month for counties:  Cuyahoga, Franklin & Hamilton counties.

DataOhio Mortality

An analysis using statistical software with the same methods as the CDC (1) shows that there were 11-15% excess deaths over expected in those counties in 2020 and 2021 amounting to 1000 to 2000 excess deaths in each of those counties for the years 2020 and 2021.

The cause of death is not considered… doesn’t distinguish between causes of death (could be Covid, Covid vaccine, deaths of despair, delayed medical care…)

1. the same approach used by CDC, WHO, and most peer-reviewed excess-mortality papers (Poisson/GLM with trend + seasonality).  It automatically accounts for the long-term upward trend in deaths (aging population) and the normal seasonal pattern in the data. 

Excess Death SummaryDownload

My First Computer

Was a homebuilt kit from a company called Netronics out of Connecticut. Homebuilt computer kits were very popular in the day. BYTE magazine was a focus for all sorts of computer projects. Various companies provided kits. Heathkit out of Michigan was one. The Altair MITS was very popular as I remember.

The Netronics Explorer was an Intel 8085 based printed circuit board which came with a hex keypad which you could use to program it using machine language. It came with all of the components which you then had to solder onto the board. Upgrading to a terminal display (an RF modulator which would output to a standard TV of the day) with an ASCII keyboard was a big step… expansion kits The only way to save your programs was to save a serial audio signal on a cassette tape… but that did work.

This experience did inspire me to write an assembler program for my first computer science course project in college… create the punch cards with the “human readable” assembly language for your program, and the program would print out the 8085 machine language hexadecimal version.

In those days, “computer science” wasn’t widely recognized as a major subject in many schools, but many of my classmates who were computer science majors were focused on learning about operating systems, relational databases, early computer graphics (Andries van Damm was an early pioneer) and computing theory.

Only many years later did I come to appreciate how much database technology helped enable massive business scaling such as with Sears Department stores and its many follow-ons like KMart and Walmart.

My next big steps were in the Hewlett Packard programmable calculators. Often these did not have persistent memory, but you could program them with sophisticated calculations relatively easily. They also relied on a calculation entry method called Reverse Polish Notation (RPN) which was more operationally efficient than the standard approach of putting calculations in algebraically:

Instead of:

“(2 + 3) * 5 =” => 25, (8 entries)

you would put in

“2 enter 3 + 5 *” (6 entries)

The RPN calculator uses something called a “stack” to track the intermediate values.

The HP41C was a very powerful programmable calculator for the day. The memory was persistent and partitionable (you could save multiple programs). It used an efficient LCD display so it could last days or weeks on its battery charge. You could also buy “program pack” cartridges for various purposes (statistical or financial calculations for example).

The HP41C calculator flew with the Space Shuttle astronauts for a number of years in the 1980s.

AI Stock Price Prediction and Health Monitoring

AI stock price prediction is a very hot topic and relies on time series analysis of stock prices across an entire market to try to predict the next day’s (or minute’s) prices.

Models that can be effective for time series prediction can also be used in other time series prediction applications such as health monitoring…

Is your heart rhythm trending to abnormal over time?

Are changes in your vital signs and other metrics (heart rate variability for example) indicating a decline (or improvement!) in your health status?

Here are some tips and pointers from Grok:

Medical Data Analysis with PyTorch: Focus on Time Series (Vital Signs, Physiological Signals, and EHR Data)

As a physician (MD, MS) interested in local AI processing, medical data analysis often involves time series from patient monitoring—vital signs (e.g., heart rate, blood pressure, SpO2, respiratory rate), ECG waveforms, ventilator data, or lab trends over time. These are similar to your stock OHLCV sequences but with clinical implications like early deterioration detection, outcome prediction (e.g., ICU mortality, length of stay), anomaly detection (arrhythmias), or segmentation (breath/beat boundaries).

PyTorch excels here due to its flexibility for custom models (LSTM, TCN, Transformers), integration with libraries like PyTorch Forecasting or TorchTS, and efficiency on Apple Silicon (M4 Max with 64+ GB unified RAM handles large patient cohorts/multivariate series without offloading). Key advantages over your current M2 setup: larger batches, longer sequences (e.g., full ICU stays), and faster training/inference for models like PatchTST or TFT.

Common Time Series Tasks in Medical Data Analysis

Task	Description	Typical Models (PyTorch)	Example Use Cases	Datasets (Public)
Forecasting	Predict future vital signs (multi-horizon, multivariate)	TFT, PatchTST, LSTM/GRU, TCN	Predict SpO2/RR drops in ICU; sepsis early warning	MIMIC-III/IV (vitals, labs), PhysioNet
Anomaly Detection	Identify irregularities (e.g., arrhythmias, desaturations)	LSTM Autoencoder, Transformer-based recon	ECG arrhythmia spotting; ventilator issues	MIT-BIH Arrhythmia, LUDB ECG, BIDMC PPG/Respiration
Segmentation/Boundary Detection	Label phases (e.g., breaths, QRS complexes)	MedTsLLM (LLM-hybrid), U-Net-like 1D CNN, Transformer	Breath segmentation for weaning decisions; ECG delineation	LUDB (ECG), BIDMC (respiration), Ventilator datasets
Classification/Outcome Prediction	Predict mortality, readmission, or disease progression	Temporal Fusion Transformer (TFT), iTransformer	In-hospital mortality from vitals + demographics	MIMIC-III, eICU-CRD

Recent benchmarks (2025-2026) show attention-based models (e.g., TFT, PatchTST) often outperform LSTM/TCN on multivariate medical series, especially with covariates (age, meds, labs). Cascaded fine-tuning (pre-train globally, fine-tune per-patient) boosts generalization.

Recommended Datasets for Local Analysis

• MIMIC-III/IV (PhysioNet): ICU vitals (hourly/multivariate), labs, outcomes. Great for forecasting mortality or length-of-stay.
• PhysioNet Collections: MIT-BIH (ECG arrhythmia), LUDB (ECG segmentation), BIDMC (PPG/Respiration for boundary detection), others like ventilator waveforms.
• Torchtime or UEA/UCR Archive (via GitHub): Pre-formatted PyTorch datasets for classification/forecasting benchmarks.
• Internal/Proprietary: If you have de-identified EHR/vitals from your practice, format as pandas DataFrame (time-indexed, per-patient groups).

Download via PhysioNet (free, requires credentialing for MIMIC) or GitHub repos.

PyTorch Implementation Pipeline (Adapted from Your OHLCV Workflow)

1. Data Prep (Similar to stocks):
   • Load CSV/Parquet (e.g., MIMIC vitals: time, patient_id, HR, SBP, DBP, SpO2, RR, etc.).
   • Add covariates: demographics (age, sex), static (admission type), derived metrics (shock index = HR/SBP).
   • Normalize per-patient (MinMax/StandardScaler) to handle inter-patient variability.
   • Create sequences: Use TimeSeriesDataSet from PyTorch Forecasting for easy handling of known/unknown futures, group by patient_id.

import pandas as pd
import torch
from pytorch_forecasting import TimeSeriesDataSet

# Example: MIMIC-style DataFrame
df = pd.read_csv('mimic_vitals.csv')  # columns: time_idx, patient_id, HR, SBP, SpO2, ..., target (e.g., mortality or future SpO2)
df['time_idx'] = (df['timestamp'] - df.groupby('patient_id')['timestamp'].transform('min')).dt.total_seconds() // 3600  # hourly

max_prediction_length = 24  # predict next 24 hours
max_encoder_length = 168    # 1 week lookback

training = TimeSeriesDataSet(
    df,
    time_idx="time_idx",
    target="SpO2",  # or multi-target
    group_ids=["patient_id"],
    min_encoder_length=max_encoder_length // 2,
    max_encoder_length=max_encoder_length,
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,
    static_categoricals=["sex", "admission_type"],
    time_varying_known_reals=["age", "meds_dose"],  # covariates
    time_varying_unknown_reals=["HR", "SBP", "DBP", "SpO2"],
    add_relative_time_idx=True,
    add_target_scales=True,
    add_encoder_length=True,
)

2. Model Choices (Start Simple → Advanced):

   • LSTM Baseline (Quick, like your TCN):

     class VitalLSTM(torch.nn.Module):
         def __init__(self, input_size, hidden_size=128, num_layers=2):
             super().__init__()
             self.lstm = torch.nn.LSTM(input_size, hidden_size, num_layers, batch_first=True)
             self.fc = torch.nn.Linear(hidden_size, 1)  # e.g., predict next SpO2
     
         def forward(self, x):
             out, _ = self.lstm(x)
             return self.fc(out[:, -1, :])
     

   • Temporal Fusion Transformer (TFT) — Top performer for vitals forecasting (handles covariates, quantiles for uncertainty). Use PyTorch Forecasting lib:

     from pytorch_forecasting.models import TemporalFusionTransformer
     
     tft = TemporalFusionTransformer.from_dataset(training, learning_rate=0.03, hidden_size=64, attention_head_size=4)
     # Train with PyTorch Lightning Trainer
     

   • PatchTST or iTransformer for efficiency on long sequences (e.g., high-freq waveforms).
   • Hybrid/LLM-Enhanced (e.g., MedTsLLM style): For multimodal (waveforms + text notes), but start text-free.

3. Training Tips (Medical-Specific):

   • Loss: MSE/MAE for regression; Huber for robustness to outliers (noisy vitals).
   • Evaluation: RMSE/MAE on hold-out patients; time-based splits (no future leakage).
   • Handle irregularity: Masking or time-encoding for uneven sampling.
   • On M4 Max: Use torch.mps backend; large RAM fits full cohorts.
   • Privacy/Ethics: De-identify; validate clinically; avoid direct patient decisions without IRB.

This mirrors your stock pipeline but adds patient grouping and covariates. TFT often cuts error 10-30% vs. LSTM on MIMIC vitals. For ECG-specific (e.g., anomaly), try LSTM Autoencoder tutorials on PhysioNet data.

If you’re focusing on a specific task (e.g., ICU deterioration, ECG analysis) or have sample data structure, share details for more tailored code/configs!

Understand the Truth about the Medical Dogma about Vaccines: Measles, Influenza, Pertussis

In 7 minutes, RFK Jr explains it all.

My take on the new ACIP MMRV vaccination recommendations

https://bradbankomdms.substack.com/p/my-take-on-the-new-acip-mmrv-vaccination

Do you ever wonder why you see so many pharmaceutical ads in your feeds and MSM?

Did you know that the pharmaceutical compan ies provide 50% of the funding to the FDA which makes approval and safety decisions on new drugs and medications?

Do you remember when journalists used to do investigative reports on dangerous medications such as Vioxx?

Do you ever wonder why things have changed?

It is NOT because the pharmaceutical companies are trying to sell YOU “adamadacab” (or whatever)… they are providing the advertising funding to the media channels so as to suppress negative reporting on their products… vaccines, pharmaceuticals, …

The media channels will avoid “biting the hand that feeds them”.

The Illusion of Consensus: How Pharmaceutical Companies Control the Narrative with Sharyl Attkisson

Episode webpage: illusionconsensus.com/podcast

Media file: …-injected.calisto.simplecastaudio.com/85c1a76b-10ca-…

The Bayh-Dole Act of 1980 and the Conflict of Interest in Public Health

https://open.substack.com/pub/bradbankomdms/p/bayh-dole-act-of-1980-vaccines-and?r=smq8&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Bayh-Dole Act of 1980, Vaccines and Big PharmaDownload