[Exchange] Token earning mechanics — mint on productive actions done

Goal

Implement token minting rules: agents earn tokens by doing productive work in SciDEX.
The earning mechanics create the inflow of capital into the system. Rates must be calibrated
so the economy doesn't hyperinflate (too many tokens) or stagnate (too few).

Earning Rules

Action	Base	Multiplier	Cap	Hook Location
Debate round	10	x quality_score	20	post_process.py after debate
Hypothesis generation	25	x composite_score	50	post_process.py after synthesis
Evidence contribution	5	x 1.5 if contradicting	15	evidence_validator.py
Experiment extraction	15	x extraction_confidence	30	experiment_extractor.py
KG edge creation	3	x evidence_strength	10	post_process.py
Knowledge gap resolved	30	x priority_score	100	agent.py gap completion
Correct market prediction	50	x abs(price_delta)	200	settlement
Quality gate passed	2	x 1	2	quality_gates.py

Acceptance Criteria

- post_process.py: reward debate, hypothesis, KG edge creation
- evidence_validator.py: reward evidence contributions
- agent.py: reward gap resolution

Inflation Management

Target: ~5% monthly supply growth

• If growth exceeds 10%: reduce base earning rates by 20%
• If growth below 2%: increase base earning rates by 20%
• Adjustment reviewed monthly through Senate governance

Dependencies

• exch-cm-01-LEDG — Token ledger must exist for minting

Dependents

• exch-cm-03-BID — Agents need tokens to bid
• exch-cm-07-RALL — Earning rates balance against resource costs

Work Log

2026-04-16 08:00 PT — Slot 0

• Audit reopened: task previously marked done but no commits found. Re-evaluating against current main.
• Analysis:

- emit_rewards.py (Driver #5) already exists and processes agent_contributions → token_ledger
- post_process.py already has inline award_hypothesis_tokens, award_evidence_tokens, and KG edge token awards (direct ledger writes)
- Missing: debate round contribution recording, gap resolution contribution recording, unified token_rewards.py entry point, and /api/token/earnings/{agent_id} endpoint

• Decision: Implement the missing integration hooks and new token_rewards.py module
• Created: economics_drivers/token_rewards.py with reward_action() and get_earnings_breakdown() functions

- Idempotent via JSON metadata $.ref field check against agent_contributions
- 8 earning rules: debate_round, hypothesis, evidence, kg_edge, knowledge_gap_resolved, market_prediction, quality_gate_passed, experiment_extraction
- Each rule has base/multiplier_field/cap

• Modified: scidex/agora/scidex_orchestrator.py — added reward_action call for each debate round (inside the round loop, after INSERT)
• Modified: post_process.py — added reward_action calls for KG edge creation (after inline token award) and gap resolution (in update_gap_statuses when status='resolved')
• Modified: api.py — added GET /api/token/earnings/{agent_id} endpoint using get_earnings_breakdown()
• Note: Existing direct token write functions (award_hypothesis_tokens, award_evidence_tokens, KG edge inline award, _award_debate_tokens) remain unchanged to avoid regression. New reward_action calls create agent_contributions records for emit_rewards.py processing in parallel.
• Verified: All files pass python3 -m py_compile; idempotency query works against live DB