recent signals
📈 Equity
Red Sea rerouting — shipping + defense
↑ 5.9/10
1h ago
📈 Equity
Private credit stress — leveraged loan defaults SHORT
↓ 5.9/10
1h ago
📈 Equity
Big Tech debt issuance flood — credit market crowding
↓ 5.9/10
1h ago
📈 Equity
Oil supply shock
↑ 6.5/10
1h ago
📈 Equity
US fiscal deterioration — sovereign term premium
↑ 6.5/10
1h ago
📈 Equity
Rate hike — REITs & homebuilders SHORT
↓ 6.5/10
1h ago
what makes it different
Cross-source clustering
3+ sources = signal
One Bloomberg article is noise. When Bloomberg, Reuters, MIT Tech Review, and FT all flag the same trigger within 48 hours, conviction multiplies by up to 1.7×.
Semaphore-limited scoring · 48h rolling window
Supply chain maps
2–3 steps deep
AI capex surge → HBM demand → DRAM market tightens → memory companies. The system follows effects downstream automatically rather than stopping at the obvious play.
Hardcoded supply chain graph · 9 trigger types mapped
Self-calibration
Learns weekly
Every week it compares conviction scores to actual P&L. Already found: policy shift and cyber signals outperform, uranium and mineral signals underperform. Scores adjusted automatically.
Pearson correlation · Per trigger-type adjustment
Discovery engine
Finds the unknown
Daily scan of articles that don't match any existing thesis. Recently found: Boeing China 200-plane order, UniCredit-Commerzbank hostile takeover, NextEra data center megadeal.
Unmatched article scan · Claude inference · Daily
70+ sources
Including research
Bloomberg and Reuters for breaking news. MIT, Stanford, NBER, Brookings for research breakthroughs. Cybersecurity feeds, commodity trackers, and agricultural data — all in one scan.
RSS + NewsAPI + price feeds · Every 2 hours
23 theses
Equity · Commodity · Bond
Tanker spikes, uranium surges, cyber attacks, AI buildout, rate cycles, Japanese repatriation, private credit stress, geothermal, and more — each with tickers, exit signals, and risk notes.
Click any thesis card on the dashboard →
the adaptive scoring engine
📡
Ingest
70+ sources
Every 2h · Bloomberg to MIT
AI SCORING ENGINE
🧠
Score + Adapt
Conviction 0–10
Scores every signal · Adjusts weekly from outcomes
🔗
Correlate
Cross-source boost
3+ sources = ×1.3–1.7 conviction
🗺️
Map
23 thesis types
Signal → tickers + exit signals
⚡
Alert
Telegram · Real-time
Structured thesis · Immediate
How the AI scoring works
🔍
Reads intent, not just keywords
Understands that "Iran warns it may consider closing Hormuz" scores 4/10 (threat only) while "Iran navy seizes oil tanker" scores 9/10 (confirmed event).
🎯
Classifies trigger type precisely
Assigns exact trigger categories: hormuz_blockade, cyber_attack, rate_hike, mineral_supply_shock — preventing oil news from firing AI theses and vice versa.
🚫
Filters noise aggressively
Analyst forecasts, opinion pieces, earnings previews, and events older than 48h all score near zero and never become alerts.
How the algorithm adapts
📊
Weekly outcome analysis
Every week it measures the correlation between conviction scores and actual market outcomes — broken down by trigger type.
⚙️
Automatic score adjustment
Trigger types that consistently lead to strong signals get scored higher going forward. Trigger types that generate noise get scored lower — automatically, without human intervention.
🔄
Injected into every future score
Historical calibration data is included in each scoring prompt — the AI reads its own track record before scoring the next article.
Conviction score scale — how signals are classified
0–2
Noise · Ignored
3–4
Rumour · Unconfirmed
5–6
Credible · Developing
7–8
Confirmed · High impact
9–10
Binary shock · Alert fires
9.5 — "Iran navy seizes tanker in Hormuz"
9.0 — "Fed cuts rates 50bps emergency meeting"
2.0 — "Analyst says oil could rise if tensions escalate"
Start reading the market differently.
70+ sources. 23 theses. An AI that gets smarter every week. Start reading the market differently.
Get Started →
comments & requests
We read every submission
Suggest a new signal, report a bug, or ask a question. The product roadmap is driven by user requests.
💡Feature request — new thesis or data source
📡Signal suggestion — something we should track
🐛Bug report — something not working
💬General feedback — anything else