Veriphix

Always-on Behavioral Science Insights (group belief). NATO Countering Cognitive Warfare Challenge winner. MissionLink '23 cohort.

[John R Fuisz @jrfuisz]

FIMI and GDP from a B3 perspective - The signature (epsilon) of an economic cognitive attack.

Across four straight quarters, professional economists under-forecast Singapore's GDP — every time. The Belief3 model did not.

This short walkthrough puts Belief3, a longitudinal population-belief panel, against quarterly GDP and the Reuters consensus. The claim is narrow on purpose: not that belief predicts GDP — nothing forecasts GDP perfectly, since events move it — but that once trained, the belief model was the best available forecast, and it didn't share the consensus's systematic downside bias.

The video closes on why that matters beyond forecasting: if measured belief carries economic weight, and belief can be moved, the macro picture has an attack surface few people are watching. I walk through what that looks like. Small sample, fully caveated — a demonstration of method, not a proven track record.

Video below (5 minutes).

See the free Veriphix Substack for the full GDP analysis — model and every caveat.

[John R Fuisz @jrfuisz]

Your LLM doesn't know what populations actually believe. Obviating adversary poisoning of your LLM training data is possible (now). You just need to want to.

Training data poisoning is an emerging attack vector. Adversaries inject biased or manipulated content into the public web, knowing it will eventually get scraped into LLM training corpora. The model learns false associations. Your analysis inherits the bias.

For population sentiment analysis, there's a structural fix: fine-tune on controlled-provenance data.

If you've been running the @Veriphix Belief3 tool (longitudinal belief measurement panels for 6+ months), you have something valuable—empirical ground truth about what a population actually thinks, collected through a methodology adversaries can't easily compromise. Anonymous panelists. Quality filters. Longitudinal consistency checks. Physical separation from the open web.

When you fine-tune an LLM on this data (or inject it via RAG), you're creating an authoritative anchor that competes with whatever the base model absorbed from the internet. Ask about population sentiment, and the model draws from measurement rather than inference from potentially-manipulated discourse.

This isn't a complete solution to LLM poisoning—nothing is. But for the specific domain of understanding what populations believe, it's a layer of defense that also happens to dramatically improve accuracy.

Proprietary data isn't just a business moat. It's an integrity moat.

[John R Fuisz @jrfuisz]

https://t.co/cmnUMcFwTy

[John R Fuisz @jrfuisz]

https://t.co/9R2zY5Iuti

[John R Fuisz @jrfuisz]

https://t.co/BK5TVDhpnf

[John R Fuisz @jrfuisz]

https://t.co/i6tkKjavC8

[John R Fuisz @jrfuisz]

https://t.co/mUHXzKR9Bi

[John R Fuisz @jrfuisz]

https://t.co/mlI8lk4zdP

[John R Fuisz @jrfuisz]

https://t.co/0LqbeGPN6m

[John R Fuisz @jrfuisz]

https://t.co/GzPYa0AOCY

[John R Fuisz @jrfuisz]

What about these quotes:

Early-warning radar for policy hot spots: Surfaced inflection points months before traditional lagging indicators.

Targeted intervention design: Flag perception gaps that ministries can address with communications or service adjustments, conserving resources compared with blanket campaigns.

Measuring soft-power pay-offs: Tracking sentiment that lets foreign-affairs teams evaluate whether trade pacts, defence dialogues, or high-level visits shift public attitudes, and recalibrate diplomacy in near-real time.

Political-risk management: Alerting well before elections or protests, giving governments space to tweak fiscal or social policies proactively.

Cross-linking with hard data: Overlaying outputs with statistical releases validates where belief drives behaviour versus where narrative correction is needed—turning sentiment into a quantifiable policy variable.

(That is not the PLTR product but what replaced it.)

$PLTR #PLTR @PalantirTech

[John R Fuisz @jrfuisz]

https://t.co/T4js0nHvEy

[Veriphix @veriphix]

“US-Japan Cooperation in a World of Information Uncertainty.”

https://t.co/HN9vkbRtDT

[John R Fuisz @jrfuisz]

https://t.co/c56Ey13A3c

[Veriphix @veriphix]

Want to see Belief3 in action? The tool the US would not let NATO use. That outs even the best cognitive campaign. Check out what is happening in Panama, the impact of the @POTUS Panama Canal tweets and a live FIMI against the Panamanian Gov.

https://t.co/oIG3Of8axM

[John R Fuisz @jrfuisz]

Can the @veriphix Belief3 Panama Data give us real-time insights into what’s happening 2,000 miles to the north, on the U.S. Southern Border?

Secretary of State Rubio is headed to Panama, so let’s analyze some fascinating Panama data. Can the Veriphix Belief3 Panama Data give us real-time insights into what’s happening 2,000 miles to the north, on the U.S. Southern Border? I wasn’t sure at first, but after diving into the numbers, the answer is a surprising YES.

It’s worth noting that the people in Panama are more than 2,000 miles away from the U.S. border. That’s roughly equivalent to asking someone in Chicago about the situation in Nogales, Mexico. Yet, Panama is located on a major immigration route, making its population’s perspectives particularly relevant.

The Connection: Veriphix Belief3 Data and U.S. Immigration Numbers

The Veriphix Belief3 data and actual immigration numbers reveal a meaningful connection.* While it isn’t a perfect correlation, we observe a moderate relationship: when consensus predictions about U.S. immigration trends increase, actual immigration numbers tend to rise as well. This finding highlights the potential for belief data to reflect real-world patterns, even from a distance.

Digging Deeper: Correlations Across 16 Topics

The dataset includes forecasts for 16 different issues. Here are the three topics most strongly correlated with actual immigration numbers on the U.S. Southern Border:
Drug Trade Increasing: Correlation = 0.71
Immigration to the United States Increasing:
Correlation = 0.63
Relations with the U.S. Increasing: Correlation = 0.60

Ensembling for Greater Accuracy

Next, we combined the data from these three topics into an ensemble model to see if we could better capture the complexity of immigration trends. The result? The correlation improved to 0.87, indicating a strong positive relationship with actual immigration numbers.

A correlation this high means the ensemble is a highly reliable predictor of immigration trends, capturing most of the variability in the real-world data. While not perfect, this result demonstrates the power of combining multiple perspectives to gain a clearer understanding of complex issues like immigration.

What This Means

Imagine that! We don’t have to be physically present at the U.S. Southern Border to understand what’s happening there. By leveraging Veriphix Belief3 data from Panama, we can identify key factors influencing migration patterns and predict trends with remarkable accuracy. The addition of more variables or more granular data could push the correlation even closer to perfection.

This analysis not only validates the significance of belief data but also underscores the potential for innovative approaches to real-time problem-solving in global affairs.

*Immigration numbers sourced from U.S. Customs and Border Protection (CBP) reports for December 2024 and January 2025, as cited in U.S. News and The New York Post articles.

[John R Fuisz @jrfuisz]

Can You Operate at the Speed of Trump 2.0?

At any moment, "Trump 2.0" may post on social media, potentially impacting your area of responsibility. When your boss or Commander asks about the implications of that post, how do you respond? You will be held accountable and are expected to provide a clear and timely answer.

Currently, DOD Information Environment Analysts (IEAs) might monitor social media and HUMINT to estimate the impact. They hope adversaries aren’t manipulating the narrative on social media and that the platform reflects broader population sentiment. Often, they rely on outdated quarterly survey data, crossing their fingers that it’s still relevant. Let’s face it: this isn’t fair to the IEAs.

Company employees (brands) have fewer resources. They really stand no chance.

The Problem: Speed

Trump 2.0 is effectively an incredibly sophisticated influence platform that leverages speed as a weapon, manipulating the pace of dialogue to achieve objectives. With direct access to Truth Social and X (Twitter), the speed will only increase. The tools IEAs currently rely on are simply not built for this rapid pace.

Speed introduces new technical challenges. You need near-continuous data from the target country, even if social media is banned. Veriphix Belief3 offers weekly or bi-weekly data. If that isn't fast enough, we can do every 3-4 days. Our solution provides real-time, bespoke data from nearly any country, even during conflicts—Russia during the war, for example. Once collected, you don’t have weeks to wait for analysis. With Belief3, you receive raw and processed data, along with a full Veriphix analysis, the next morning. Although we promise a 24-hour turnaround, we often deliver within 2–3 hours.

Scale and budget constraints shouldn’t be barriers. We scale in days and at costs that will make you question how other vendors allocate their resources.

The Choice is Yours

Spend more time and money for subpar answers, or equip yourself with tools for success. Trump 2.0 isn’t slowing down for anyone.

Round 3 of the Veriphix Panama feed launches on January 9, 2025. Between the Baseline (December 12, 2024) and Round 2 (December 25, 2024), we’ve already analyzed the impact of POTUS Trump’s Panama Canal posts. Our findings include key drivers of U.S.-bound immigration, the Chinese attack vector, and insights into the target audience. And that’s just the tip of the iceberg—we can’t responsibly disclose everything here.

(Better Than) White House-Level Briefings

Available to virtually anyone. Awardable through the Tradewinds AI Marketplace.

You have no excuse. Equip yourself for success.

[John R Fuisz @jrfuisz]

Do you want to know the impact of Trump's Panama Canal social media posts had on the Panamanian public? Did they go pro-US? Pro-China? Pro-Russia? What about immigration did it go up or down?

Round 2 of the Veriphix Panama Feed is out. Data from 12/25 is compared to the 12/12 baseline data. We even included some sample nudges so you can see how messaging can be improved and/or hunt for adversary influence campaigns.

This is NOT based on social media scraping. All data is sourced from ethical (consenting, compensated and anonymous) sources inside Panama. And we do it again in 2 weeks.

Ping if you need access. Some groups already have it. If you don't, then you definitely want to ping.

[Veriphix @veriphix]

Congratulations to the all of the winners of the Small Wars Journal and Motive International's Horizon 2040 flash fiction competition. But most importantly congratulations to Jen Watkins our very own Veriphix CTO for winning with "Mel Pomene Now."

[John R Fuisz @jrfuisz]

NFL Betting Experiment - End of the Pre-season.

How did the LLM/AI do?

The pre-season has always been somewhat unpredictable. We asked the LLM/AI to analyze trends from 40 years of NFL games, most of which are regular season games. However, the pre-season is full of uncertainties. But just as in your business, where you apply LLM/AI, we face unknowns. So, how did we perform?

Jump to the free @veriphix substack to read the rest. Blame character limit. The substack is always free but sign up if you want to receive the notifications when something is published.

https://t.co/CQ8qgHlu1i