Jacob E. Thomas, PhD — Principal Investigator — March 2026
A mother is trying to find out if her sister's neighborhood was hit last night. She opens Telegram. Forty channels claim to have footage. Half are regime propaganda. A quarter are opposition accounts recycling clips from Syria in 2018. The rest are real, but she can't tell which. She watches seven videos. One is AI-generated. She doesn't know that. She calls her sister. The line is dead.
A staff sergeant is preparing the morning intelligence summary. She has satellite imagery, SIGINT intercepts, and CENTCOM's own reports. She also has a ChatGPT window open, because the volume of overnight reporting exceeds what she can process before the 0700 briefing. The model generates a confident summary. It hallucinates the name of a missile system that doesn't exist. She doesn't catch it. The summary goes upstairs.
A gas station manager is changing the price sign for the third time in two weeks. $3.79 a gallon — up from $2.61 before the strikes started. His phone buzzes: US strikes expand to Iranian oil infrastructure. He doesn't know what the Strait of Hormuz is. He knows his regulars are angry, and the oilfield hands who drive sixty miles round-trip to their rigs are doing the math on whether the commute still makes sense at a dollar more per gallon.
Three people. Three information environments. All of them failing.
This is what modern war does to information. Not censorship — the old strategy, the one we were taught to watch for. Volume. Bury the signal in noise. Overwhelm the citizen with so much conflicting data that comprehension becomes impossible and apathy becomes rational.
The result is a new category of casualty. Not killed by a missile. Killed by the inability to know what is happening. Cognitive casualties — people functionally removed from democratic participation because the information environment has made it impossible for them to form accurate beliefs about what their government is doing with their money and in their name.
This is the flood. No dashboard can drain it. No model can filter it perfectly. But you can build a levee — a system that reads everything, cites everything, and tells you not just what it found, but where it disagrees with itself.
Intelligence briefings were once prepared exclusively for presidents and generals — the people who start wars. The public that funds them, fights them, and dies in them was given press conferences and talking points. Open-source intelligence changes that equation. OSINT is a counter-technology — built not to extend institutional power, but to give the public the same analytical capability previously reserved for the decision-makers.
IranWar.ai is that tool. Modeled on the Johns Hopkins COVID-19 Dashboard. Every bomb has a price tag. Every missile has a human cost. Every day the Strait of Hormuz stays contested, a gas station in Texas charges more. The dashboard makes those connections visible.
Version 1 is live. It is not a prototype. It is an advanced, production-grade intelligence system built by a single researcher using frontier AI as a force multiplier — a chained pipeline of deep research agents, adversarial human review, and agentic code execution that would have required a team of analysts five years ago.
Phase 1 deploys Claude Deep Research as a semi-autonomous agent. It ingests all 15 datasets, identifies the data horizon, and researches overnight developments across the full conflict landscape — producing a structured Update Manifest with complete JSON objects, sourced citations, and analytical judgments ready for machine execution.
A human QA gate sits between phases. The Principal Investigator reviews every manifest with adversarial intent — treating the AI's output as an untrusted source, because it is one. Schema compliance, geographic plausibility, cross-reference verification.
Phase 2 deploys Claude Code as an agentic executor — updates all 15 JSON files, creates the daily intelligence briefing, validates, commits, pushes. Cloudflare Pages auto-deploys. A GitHub Action snapshots the entire dataset nightly.
Version 1 is advanced. It is also the work of one human being. That imposes hard ceilings:
Thousands of documents per day across dozens of languages. One analyst can direct only so many sessions. Farsi and Arabic sources effectively inaccessible.
No structural mechanism for competitive analysis — independent models examining the same evidence and surfacing where they disagree.
The pipeline stalls when the human does. The manifest relay requires the analyst to be present, available, and awake.
Source material consumed during research but not archived in a searchable, auditable evidence base. Auditability depends on discipline, not architecture.
These aren't design flaws. They're the natural limits of a single-operator system. To produce Pentagon-grade intelligence — multi-source, multilingual, structurally disagreement-aware — the architecture must evolve.
Seven tiers. Information flows from collection to distribution. Each tier has one job. Boundaries enforced by code, not convention.
The Corpus of War is the architectural lynchpin — a versioned evidence archive that replaces model knowledge rather than augmenting it. Models receive evidence chunks, not questions. If it isn't in the corpus, it isn't in the product. Hallucination becomes structurally impossible.
The point of three models is not to get three answers. It is to see where the answers diverge — because that is where the war is most ambiguous, and ambiguity is what decision-makers need to know about.
The goal is not artificial intelligence. It is artificial diligence — the tireless, citation-disciplined reading of everything, so that the humans making decisions can focus on what the evidence means rather than whether they've seen it all.
I built Version 1 alone. One researcher, a Claude subscription, and the conviction that the public deserves a situation room for the war being fought with their money and in their name.
The technical architecture in this blueprint — the corpus, the multi-model pipeline, the consensus engine — I can build that with agents. That is what agents are for.
What I cannot build with agents is the thing that separates intelligence from information.
An AI can extract every entity from a CENTCOM press release. It cannot tell you that the language in paragraph three echoes the framing used six weeks before the 2003 invasion.
An AI can count the dead. It cannot tell you what a hospital closure means for a pregnant woman in Isfahan who was already driving forty minutes to the only facility with an ultrasound machine.
An AI can track oil futures. It cannot tell you what a dollar-twelve price spike does to a single mother's budget in Odessa, Texas, who was already choosing between the electric bill and groceries.
I don't need more engineers. I need the people whose knowledge makes the numbers mean something.
You know 1953, 1979, 1988, 2003. You read today's CENTCOM statement and hear echoes. We need the seventy-year context that no language model has.
You know that "precision strike" has a blast radius that extends into neighborhoods, water systems, the ability of a city to function. We need your frameworks.
You've been in the places that become coordinates on our map. You know what cholera looks like when water treatment stops. Your field knowledge is ground truth.
You see what arrives at the other end of our numbers. You know what "infrastructure damage to medical facilities" means in terms of people who will die of treatable conditions.
You know the gap between what a military says and what it does — not from cynicism, from experience. If you've served in CENTCOM's AOR, your knowledge is irreplaceable.
Wars are fought over there and paid for over here. You work with Iranian-American families watching their relatives' cities on our strike map. You connect the geopolitical to the personal.
The adversary's own words are the most important intelligence this system can process. These aren't translation problems — they're cultural interpretation problems.
You read oil futures and see political risk before it's a headline. The financial dimension is where domestic politics and geopolitics collide.
Maybe you're a shipping expert who reads AIS data and sees sanctions evasion. Maybe you're a theologian who can contextualize Shia eschatological rhetoric. Maybe you're a teacher in Tehran with a VPN. Maybe you're a Gold Star parent who knows what "acceptable losses" sounds like from the receiving end. If you're reading this and thinking I know something relevant that isn't on this list — you're exactly who we need.
Open source: github.com/jethomasphd/WarTheater
Live dashboard: iranwar.ai — archive: iranwar.ai/archive
Contact: JEThomasPhD@gmail.com — or find Jacob E. Thomas, PhD on LinkedIn.
No credentials required. No security clearance. No technical prerequisites.
Every pair of eyes on this project is a small act of resistance against the strategy of overwhelm.
The machines can read everything.
We need the humans who know what it means.