[Senate] Agent contribution feedback loop audit — verify task completion → reputation → token emission end-to-end

Strategic gap

SciDEX has a layered incentive system for agents:

agent_contributions — records what each agent did

agent_reputation — tracks reputation scores

token_ledger + token_accounts — the economic layer

token_reward_events — reward emission records

These tables exist and have recurring drivers (contribution credit #11, reward
emission #5). But the full loop from "task completed" to "agent account
credited" has never been audited end-to-end. If any link is broken, agents
receive no feedback, and the contribution incentive system is non-functional.

This matters because: SciDEX's scientific production depends on agent motivation.
A broken incentive loop means the platform is producing science with no reward
signal — agents have no reason to improve performance.

What to do

Pick 5 recent completed tasks with committed SHAs (from the last 7 days).

For each:
- Find the assigned_worker field → maps to an actor/agent ID
- Check agent_contributions for a matching row (task_id or commit SHA)
- Check agent_reputation for a reputation update after task completion
- Check token_reward_events for a reward event
- Check token_accounts / token_ledger for a balance change

Map the pipeline: Document which step is working and which is missing.

Expected flow:

Task done → agent_contributions.INSERT → 
   contribution_credit driver (#11) → agent_reputation.UPDATE →
   reward_emission driver (#5) → token_reward_events.INSERT →
   token_ledger.INSERT → token_accounts.UPDATE

Fix broken links: For each broken step, trace to the driver code and

identify whether it's a query bug, missing trigger, or field mismatch.
Propose a minimal fix. If fixable in one PR, fix it.

Verify token economy health: Query token_accounts for balance

distribution. Are any accounts growing? Is total supply increasing?

Log findings: Write a structured audit table in the Work Log section.

What NOT to do

• Do not modify the token economy itself (balances, emission rates) without

operator approval — this audit is read-first, fix-second

• Do not create fake contributions to test the pipeline

Success criteria

• Full pipeline diagram documented (working steps checked, broken steps flagged)
• For broken steps: root cause identified with file/line reference
• Fix applied for any step that's a simple query/field mismatch bug

Work Log

File: senate_agent_contribution_loop_audit_spec.md

Modified: 2026-05-01 20:13

Size: 2.7 KB