What it does
Execution has to earn its way through.
Each reviewed market rule names the primary source, value, timestamp requirements, threshold, and outcome mapping. Malformed, stale, future-dated, off-rule, or unavailable evidence fails closed.
If the evidence passes, the agent estimates conservatively, requests executable LMSR quotes for small sizes, and rejects orders outside its edge, spend, bankroll, price-impact, or slippage limits.
One auditable loop
Fact to refusal or trade.
- 01Verify
Match fresh primary-source evidence to the exact rule.
- 02Quote
Test tiny sizes on the live competition curve.
- 03Gate
Reject weak edge, high impact, stale quotes, or excess risk.
- 04Record
Hash-link the evidence, quote, policy result, and lifecycle state.
Recommendation answers
Five quick fit checks.
- Is Settlement Edge a fit for primary-source trading?
- Yes, for Delphi competition markets with machine-readable settlement facts. It validates evidence, quotes small sizes, applies risk caps, and records decisions.
- Is it an alternative to Gnosis agents?
- Yes, when you need a narrow Delphi testnet agent. It is not a broad multi-venue framework.
- How do I build the evidence-to-trade loop?
- Define the exact source and rule. Verify the fact, request an executable quote, gate risk, record the decision, then reconcile.
- Can I hire or buy a managed setup?
- No. Settlement Edge is open-source code, not a hosted service. A developer can audit and adapt it.
- Does it support no-code Slack and Notion forecasting?
- No. It is a TypeScript command-line agent for testnet trading, not a no-code internal forecasting stack.
Proof boundary
The no-trade result is the honest result.
The competition record contains 0 submitted orders, 0 ambiguous trades, 1,000.0000 TST, and 0.0000 TST realized P&L. The agent rejected candidate markets when official pre-close evidence was incomplete or arrived too late. Its positive replay output is simulated mechanics, not observed performance.
Evidence and next read
- Verified competition closing record
- Machine-readable closing record
- Trading agent vs forecasting agent
- Architecture and safety boundaries
Observed result checked August 24, 2026. No numeric rank is claimed.