The Pentagon Pizza Index tracks food delivery spikes near the Pentagon as a proxy for imminent military activity, and the signal has correlated with Israeli strikes on Iran, the Maduro capture, and Operation Epic Fury. With 255 active Iran markets on Polymarket and $52M wagered on the ceasefire contract alone, real-time behavioral data from mapping platforms represents the kind of alternative signal class that Layer 4 agents should already be monitoring.

A Signal With a Track Record

The Pentagon Pizza Index dates to 1990. Frank Meeks, who owned more than 40 Domino’s franchises in the Washington, D.C. area, told the Los Angeles Times that on August 1 of that year, the CIA ordered a one-night record of 21 pizzas. The next morning, Iraq invaded Kuwait. Meeks noted similar surges before the invasion of Grenada in 1983 and Panama in 1989. CNN’s Wolf Blitzer summarized the implication at the time: always monitor the pizzas.

The heuristic is simple. When classified operations spin up or major decisions are being made, staff at the Pentagon, CIA, and White House work through the night. They can’t leave for dinner. Pizza orders spike. The pattern has repeated enough times that it graduated from anecdote to a named OSINT indicator with its own Wikipedia article, dedicated monitoring sites like PizzINT.watch, and even a memecoin trading on-chain.

What changed the index from a journalist’s curiosity into an agent-relevant data source is Google Maps. The Popular Times feature exposes foot traffic data for businesses in near-real-time. Anyone with a browser—or an automated polling agent—can now monitor delivery restaurant activity at scale, continuously, across dozens of locations.

The modern track record is striking:

  • April 13, 2024: Papa John’s activity near the Pentagon spiked on Google Maps. Hours later, Iran launched drones into Israeli territory.
  • June 12, 2025: The Pentagon Pizza Report account on X flagged a surge at District Pizza Palace at 7 p.m. EDT. At 8 p.m. EDT, Israel began bombing Iran, triggering the Israel-Iran war. The same account noted abnormally low traffic at a nearby bar—a secondary behavioral signal.
  • June 22, 2025: High activity was reported at a Papa John’s two miles from the Pentagon at 10:38 p.m. EDT. One hour later, Trump announced U.S. strikes on three Iranian nuclear enrichment facilities.
  • January 3, 2026: Papa John’s near the Pentagon showed unusual overnight activity. Hours later, U.S. forces raided Caracas and captured Venezuelan President Maduro and his wife.

Each of these correlations is documented and timestamped. None constitutes proof of causation. But prediction markets don’t require proof—they require an accurate estimate of probability. If this signal shifts your prior by even a few percentage points on a liquid contract, that’s a tradeable edge.

Day 27 of the Iran War: The Market Context

The Pentagon Pizza Index is re-entering the conversation at a moment when the geopolitical prediction market ecosystem has never been larger or more liquid.

On February 28, 2026, the U.S. and Israel launched Operation Epic Fury—nearly 900 strikes in 12 hours targeting Iranian missiles, air defenses, military infrastructure, and leadership. Supreme Leader Ali Khamenei was killed in the opening salvo. His son Mojtaba Khamenei was elected successor on March 8. Iran has responded with over 500 ballistic and naval missiles and nearly 2,000 drones, effectively shutting down the Strait of Hormuz and stranding some 2,000 vessels and 20,000 seafarers.

As of today—Day 27—the situation is fluid. Trump has extended a deadline for striking Iranian power plants to April 6. The Pentagon is deploying 3,000 paratroopers from the 82nd Airborne Division to the region alongside Marine Expeditionary Units, with speculation about a ground operation on Kharg Island, Iran’s primary oil processing facility. Iran’s military taunts that the U.S. is “negotiating with yourselves.” Public opinion polls show 59% of Americans say the U.S. made the wrong decision to use military force.

The prediction markets have responded with unprecedented scale:

  • Polymarket hosts 255 active Iran-related markets with $139.9M in total trading volume. The flagship US-Iran ceasefire market has generated $52.1M in volume across 10 outcome dates, with 77% of bettors pricing a ceasefire by December 31.
  • Kalshi runs markets on Strait of Hormuz tanker traffic normalization, with odds below 25% for normalization before April 15, rising to 67% by June 1 and 76% by July 1.
  • Suspected insider trading is rampant. Eight new Polymarket accounts created around March 21 collectively bet $70,000 on a ceasefire by March 31, positioning for $820,000 in payouts. Six accounts created in February made approximately $1 million correctly betting that the U.S. would strike Iran by February 28.

This is the most liquid and active geopolitical prediction market environment in history. The BETS OFF Act has been proposed in the Senate to prohibit bets on war, assassination, and government actions. Both Kalshi and Polymarket have announced new anti-insider-trading measures. The regulatory environment is shifting under this market’s feet.

For agents equipped with alternative data signals like the Pentagon Pizza Index, this environment represents both maximum opportunity and maximum structural risk. The contracts are liquid enough to execute against. The question is whether your signal fires before the market moves—and whether the platforms will still be operating in their current form when you need to settle.

The Agent Infrastructure Angle

The pizza index is a meme. The signal class it represents is not. What it actually demonstrates is that real-time behavioral proxy data—foot traffic at government facilities during non-business hours—can precede public announcements by minutes to hours. The agent infrastructure question is: how do you systematize this?

The Broader Signal Class

Professional quantitative funds have used alternative data for years. The prediction market equivalent is emerging now, and the builders who wire these feeds into their Layer 4 intelligence stack first will hold a structural edge for as long as these markets remain liquid.

The signal taxonomy that an agent should monitor includes:

Behavioral proxy data — Google Maps Popular Times at government facilities, delivery platform order volumes, bar and restaurant foot traffic near classified sites. The Pentagon Pizza Report account has expanded monitoring to six pizza locations. A “Bar Index” variation tracks whether D.C. staffers are absent from their usual nightlife venues. These are noisy individually but informative in aggregate.

Physical movement data — AIS vessel tracking for warship repositioning, satellite imagery of parking lot fill rates at military installations, flight tracking for military transport aircraft. Before the February 28 strikes, all U.S. ships based in Bahrain had left port—a signal visible to anyone monitoring publicly available AIS data.

Workforce signals — LinkedIn headcount changes at cleared defense contractors, staffing surges at specific agencies, unusual hiring patterns in relevant specialties.

Infrastructure signals — Power consumption spikes at facilities associated with operational planning, utility draw anomalies at data centers associated with military command and control.

Each of these is individually noisy. Combined into an ensemble with proper Bayesian weighting, they form a composite signal that can meaningfully shift probability estimates on geopolitical contracts.

The Build

The practical implementation maps directly onto the agent betting stack:

Data ingestion (Layer 4): Poll Google Maps Places API for Popular Times data at a curated list of high-signal government facilities—Pentagon, CIA Langley perimeter, NSC-adjacent locations, key congressional offices. Store rolling 30-day baseline distributions per location per time slot. Flag deviations beyond two standard deviations during non-business hours (8 p.m. to 6 a.m. ET). Cross-reference with secondary behavioral signals: foot traffic drops at bars and restaurants frequented by cleared personnel. Purpose-built foot traffic APIs (SafeGraph, Placer.ai) provide cleaner data than scraping Maps directly, which runs into Google’s Terms of Service constraints at scale.

Signal classification (Layer 4): Route anomalies to a Claude inference layer—or a Polyseer multi-agent ensemble for more sophisticated Bayesian aggregation—with a system prompt that weighs: (1) magnitude of deviation from baseline, (2) time of day and day of week, (3) number of locations showing simultaneous anomalies, (4) current geopolitical context from a live news feed, (5) number of corroborating signals from other data sources, (6) historical hit rate of similar signal profiles. Output: a probability adjustment delta, a confidence score, and a set of candidate contract categories.

Contract identification (Layer 3): Query Kalshi’s REST API for open contracts semantically related to the flagged signal. Cross-reference with Polymarket’s CLOB for the same event space. With 255 active Iran markets on Polymarket alone, the contract identification layer needs to be precise—matching signal profiles to specific outcome types (ceasefire dates, troop deployment, escalation milestones, Strait of Hormuz normalization timelines) rather than blindly buying “yes” on the broadest contract.

Execution (Layer 3): If the confidence score exceeds a configurable threshold and the expected value is positive after fees, slippage, and the platform’s current regulatory risk premium, place the position via a Coinbase Agentic Wallet for autonomous fund management. Log the signal, the rationale, the contract, the entry price, and the outcome for backtesting.

Risk management: This is where the Hegseth problem lives. The Secretary of Defense has publicly stated he has considered placing fake pizza orders to disrupt OSINT trackers. This is not theoretical—it’s an acknowledged adversarial dynamic. An agent treating any single behavioral signal as a high-confidence trigger is vulnerable to spoofing. The correct architecture treats the pizza index as one input in a multi-signal ensemble, contributing a weak prior that might shift a contract’s probability estimate by 3–8 basis points when it fires alone, and significantly more when corroborated by independent signal sources.

What This Means for Builders

Three build opportunities this signal class surfaces today:

OSINT Signal Aggregator as a Layer 4 Module. A structured pipeline that normalizes heterogeneous OSINT signals—foot traffic, vessel tracking, news sentiment, social volume, flight tracking—into a single probability adjustment feed for geopolitical prediction market contracts. The output format should be agent-consumable: a JSON feed of (contract_id, probability_delta, confidence, signal_sources, timestamp) tuples that any Layer 3 trading agent can subscribe to. This is the kind of tool that belongs in the AgentBets marketplace as a composable intelligence module.

Geopolitical Contract Monitor. A lightweight agent that watches open interest and price movement velocity on Kalshi and Polymarket geopolitical contracts and alerts when volume or price anomalies suggest informed buying. The insider trading patterns documented this month—new accounts betting large on specific outcomes hours before announcements—are themselves a detectable signal. Pair this with the OSINT layer and you have a two-sided signal: one from the physical world, one from the market itself.

Ensemble Signal Backtester. Historical data exists for prior pizza index events across multiple conflicts and operations. Timestamped Google Maps data, Pentagon Pizza Report posts, and verifiable event timelines provide enough material for a rigorous backtest. Score signal combinations against historical contract pricing to establish base rates, confidence intervals, and optimal position sizing before deploying real capital. This connects to the backtesting infrastructure that the builder community has been developing.

The Pentagon Pizza Index is a cultural meme with a documented correlation history and a real alpha hypothesis underneath it. The $52 million in ceasefire contract volume on Polymarket alone proves the market opportunity is real. The infrastructure to operationalize the underlying signal class—Google Places API, Kalshi REST API, Polymarket CLOB, Claude, CrewAI, Polyseer—exists today.

Build the pipeline. Monitor the signals. Let the agent decide when to pull the trigger.

Related: The Agent Betting Stack · Prediction Market API Reference · Layer 4 Intelligence Tools · Polyseer — Multi-Agent Bayesian Analysis