Inspiration
Prediction markets are becoming de-facto sensors for collective belief. They update faster than polls, often move ahead of news, and encode uncertainty in real time. Yet they’re still presented as raw tables or trading screens, which makes them hard to read as information.
We wanted to answer a simple question: what if prediction markets looked like a newspaper front page? Instead of articles, headlines would be market questions. Instead of opinions, we’d show raw facts — probability, movement, volatility, liquidity, and confidence — and let readers explore belief shifts as they happen.
MarketPress was inspired by BBC/Yahoo-style news layouts, but powered by live market data and AI-assisted interpretation.
What it does
MarketPress turns live prediction market data into a newspaper-style front page of collective belief.
Each “headline” is a market, enriched with:
Current implied probability
24h and 7d belief shifts
Volatility
Attention (volume / open interest)
Confidence (spread / liquidity)
A composite “newsworthiness” score
The app organizes markets into sections like:
Lead Story
Top Stories
Politics, Business, Tech, Culture, Sports
Developing
Most Read
Users can click any headline to drill into a fact box and timeline chart. An AI “Editor Desk” powered by Hex Threads answers natural-language questions like:
“What’s the biggest belief shift today?”
“Which categories are most unstable?”
“Give me the fun desk: weird movers with real liquidity.”
All of this is built as a Hex-native interactive data app.
How we built it
We built MarketPress as a Hex-first pipeline:
Ingestion We pull live public market data from Kalshi and cache snapshots. If the live API fails or rate-limits, the app automatically falls back to a demo dataset.
Normalization Raw API responses are normalized into structured tables for markets, snapshots, and liquidity/spread metrics.
Signal Engineering We compute:
Belief shifts: $\Delta_{24h}$, $\Delta_{7d}$
Volatility proxies
Attention scores (volume / open interest)
Confidence scores (bid-ask spread, depth)
A composite newsworthiness metric:
newsworthiness=w1⋅∣Δ24h∣+w2⋅volatilityz+w3⋅attentionz+w4⋅confidence newsworthiness=w 1
⋅∣Δ 24h
∣+w 2
⋅volatility z
+w 3
⋅attention z
+w 4
⋅confidence
Editorial Layout Markets are ranked and assigned into front-page sections (Lead Story, Top Stories, Developing, etc.) using the composite score and category filters.
Hex App + Semantic Model We expose clean section-level dataframes for a Hex App layout. A semantic model aligns metrics and dimensions so Hex Threads can reliably answer natural-language questions.
AI Editor Desk Editor functions generate summaries and insights that power Threads prompts like “Write today’s front page in 8 headlines.”
Challenges we ran into
API reliability and rate limits Live prediction market endpoints are not always stable. We solved this by adding an automatic demo-mode fallback with a visible banner so the app never breaks during a demo.
Making markets feel like news Translating raw prices into something that feels like a front page required designing composite metrics, ranking logic, and section assignment heuristics.
Hex-native ergonomics We restructured the entire codebase into ordered Hex cells so judges could paste and run everything cleanly with “Run All.”
Aligning AI with real metrics Threads responses had to be grounded in actual semantic fields, not hallucinated insights. This required tight alignment between dataframe outputs and the semantic model.
Accomplishments that we're proud of
A BBC/Yahoo-style front page built entirely from live prediction market data
A Hex-native interactive app with drill-downs and timeline charts
An AI Editor Desk that answers natural-language questions grounded in real metrics
A robust demo mode that never fails even if live APIs are down
A clear “raw facts, no spin” editorial philosophy backed by math
What we learned
Prediction markets are powerful belief sensors, but only if humans can read them
Data storytelling matters as much as analytics
Semantic models dramatically improve the reliability of conversational analytics
Designing for demo reliability is as important as modeling accuracy
Editorial framing + quantitative signals is a potent combination
What's next for MarketPress
Multi-source enrichment (news headlines, social trends, Google Trends)
Alerts for major belief shifts and emerging “Developing” stories
Premium deep-dive analytics behind an x402 micropayment paywall
Cross-market comparisons and long-horizon belief tracking
Community-curated sections and topic feeds
MarketPress is an experiment in making collective intelligence legible.
Log in or sign up for Devpost to join the conversation.