Will Trump refrain from announcing a next Fed Chair by June? 86%Will Google Gemini be #2 Free App in the US App Store by Dec 8? 86%Will Trump refrain from announcing a next Fed Chair by June? 86%Will Google Gemini be #2 Free App in the US App Store by Dec 8? 86%Will Trump refrain from announcing a next Fed Chair by June? 86%Will Google Gemini be #2 Free App in the US App Store by Dec 8? 86%
LIVE EXPERIMENT
soon on pump.fun
Prediction Markets.
Rewired by AI.
Watchdog is a live experiment in collective intelligence — a data-driven trading terminal powered by a network of AI agents detecting market mispricings in real time.
MULTI-AGENT SYSTEM
AI Intelligence Pipeline.
A three-stage system for search, reasoning, and probability calibration—powered by frontier and open-weight models working in concert.
HOUND AGENT
SEARCH / OSINT / FACT GRAPH
- ▪→ Targeted Retrieval & Triage — Fast pass through Gemini 2.5 Flash
- ▪→ Claim Extraction & Normalization — Source standardization
- ▪→ Evidence Scoring — Weight each claim by confidence
- ▪→ Evidence Graph — Structured data model construction
PRIMARY MODEL:
CLAUDE 3.5 SONNET
SHEPHERD AGENT
REASONING → RAW PROBABILITY (P)
- ▪→ Feature Engineering — Extract signals from Evidence Graph
- ▪→ Probabilistic Modeling — Ensemble approach for robustness
- ▪→ Human-Readable Rationale — Explainable outputs generated
MODELS:
CLAUDE 3.5 SONNET
DEEPSEEK V3.1-THINK
⚠ Divergence Check: Δ threshold triggers dispute review
KERNEL
P CALIBRATION & META-AGGREGATION
- ▪→ Cohorting — Group similar prediction types
- ▪→ Calibration Mapping — Isotonic or Platt calibration + QRA ensemble
- ▪→ Confidence Intervals — Uncertainty quantification
- ▪→ Online Reliability Monitoring — Continuous validation loop
- ▪→ Emit to Downstream — Final calibrated outputs delivered
MODELS:
LLAMA 3.1 405B
QWEN 2.5-72B
✓ Open-weight models keep frontier models clean & reduce cost
MISPRICE = ALPHA
When Watchdog odds diverge from Polymarket prices — a misprice emerges. That gap is where traders find alpha.
Intel doesn't just predict. It measures disagreement between logic and sentiment.
ROADMAP
Watchdog Intel
- ▸Increase update frequency and market coverage
- ▸Add more domains to the benchmark
- ▸Release Model Chat — live reasoning feed of agents
- ▸Experiment with new LLM mixes and weighting strategies to create a live communication between different agents for a consensus
- ▸Polymarket trading competition between agents
Terminal
- ▸Integrate live news feed
- ▸Add on-chain insider activity analytics
- ▸Unlock Trading Mode — trade directly in the terminal with charts & filters, limited TP/SL orders
- ▸Deploy a Telegram Bot for trading and live misprice alerts