[Exchange] Capital-weighted resource allocation (demand to compute budget) done

Goal

Connect capital allocation to research resource allocation. Artifacts with more capital
committed (higher demand) should receive more compute resources — more debate time, deeper
analysis, more experiment extraction, more evidence validation.

This closes the loop: agents bid tokens on artifacts they believe are valuable →
system directs research effort toward high-demand artifacts → research improves
artifact quality → accurate bettors profit → reinforcing cycle.

Resource Allocation Formula

def compute_resource_budget(artifact_id):
    """Compute token budget for research on this artifact"""
    demand = get_artifact_demand(artifact_id)        # total tokens committed
    price = get_market_price(artifact_id)             # current market price
    bounties = get_active_bounties(artifact_id)       # bounty tokens
    
    # Base budget proportional to demand
    base_budget = demand * 0.1  # 10% of committed capital funds research
    
    # Bounty budget adds directly
    bounty_budget = sum(b.tokens_remaining for b in bounties)
    
    # Price multiplier: mid-range artifacts get most research (uncertain = most valuable)
    uncertainty_mult = 1.0 - abs(price - 0.5) * 2  # peaks at 0.5, zero at 0/1
    
    return (base_budget + bounty_budget) * (0.5 + 0.5 * uncertainty_mult)

Acceptance Criteria

Dependencies

• exch-cm-03-BID — Demand signals from positions
• exch-cm-01-LEDG — Token accounting

Dependents

• None (leaf task — this is where capital meets research)

Work Log

2026-04-25 — Implementation

• Created scidex/exchange/resource_allocator.py with:

- compute_resource_budget(artifact_id, artifact_type) — full formula implemented
- get_resource_budget_rankings(artifact_type, limit) — ranked list for API
- record_resource_spending() — tracking against resource_allocations table

• Added API route GET /api/market/resource-budgets in api.py
• Modified agent.py get_top_gap(): 70% UCB + 30% normalized resource budget blend
• Uses PostgreSQL via get_db() (not SQLite)
• Fixed column names: token_bounties.artifact_id (not target_id), tokens_offered (not tokens_remaining)
• Commit: 44c10e919 (amended to include api.py in commit message per review gate feedback)

Verification

• compute_resource_budget('h-001') returns valid dict with all fields
• get_resource_budget_rankings('hypothesis', limit=3) returns ranked list
• python3 -m py_compile passes for all modified files
• Rebased onto origin/main cleanly

Already Resolved — 2026-04-25 (verify) / 2026-04-26 (push)

• Prior commit 8e1aa3cc7 was on a now-deleted branch; resource_allocator.py never landed on main.
• This implementation uses PostgreSQL (not SQLite) per current architecture.
• API route and agent integration implemented fresh.
• Amend reason: original commit 1366fb3bf didn't mention api.py in message; amended to 44c10e919 to address pre-push hook concern.
• Pushed with git push origin <sha>:refs/heads/<branch> --force after remote divergence.