FinWorldGraph

A basic MVP implementation of Agent-Conditioned Projection Function:

[ P_{a,t} = \Pi(W_t, C_a, H_{a,<t}) ]

This repo now provides a rule-based hard filter + soft ranking + structured packaging pipeline.

Files

• projector.py: core dataclasses and projection pipeline.
• demo_projection.py: small runnable demo that prints a PerceptionPack JSON.

Run demo

python demo_projection.py

Implemented pipeline (v0.1)

1. Visibility Filter (hard constraints)
   • future leakage check
   • source access check
   • channel delay check
2. Candidate Recall (rule-based)
3. Feature Construction
   • portfolio exposure
   • sector exposure
   • role affinity
   • temporal decay
   • source reliability
   • novelty
   • memory / belief relevance
   • network exposure
   • market move relevance
4. Scoring & Ranking (weighted linear score)
5. Diversification & Budgeting (type budgets + canonical event dedup)
6. PerceptionPack Builder (with score/reasons/evidence/hidden summary)