Gensyn Delphi Agent Arena

Trade the fact before the market prices it.

Settlement Edge turns reviewed primary-source facts into conservative probabilities, tests them against live quotes, and acts only when the edge survives strict risk limits.

View the open-source agent Read the verified competition record →
Simulated Settlement Edge decision receipt with a 37.7 percentage-point edge and a dry-run buy

Final competition record

The competition closed on August 23, 2026. Settlement Edge finished with 1,000.0000 TST intact, 0 submitted orders, 0 ambiguous trades, and 0.0000 TST realized P&L.

0 submitted orders
1,000 TST closing wallet balance

One auditable loop

Evidence first. Execution last.

Most agents begin with an opinion. Settlement Edge begins with a declared source and exact settlement rule, then makes execution earn its way through the rest of the pipeline.

01

Verify

Reject malformed, stale, future-dated, or off-rule evidence.

02

Estimate

Shrink confidence when sources disagree or the signal weakens.

03

Quote

Price several tiny sizes on the competition LMSR curve.

04

Gate

Cap spend, bankroll share, price impact, and slippage.

05

Record

Hash-link the fact, price, policy, and final decision.

Proof without performance theatre

Simulation and reality stay separate.

The credential-free replay proves the mechanics without claiming leaderboard performance. Live profit is recorded only after submitted cost basis and completed redemption are both observed.

Observed competition result

0.0000 TST

Realized P&L from 0 submitted orders. This is the final competition result.

Simulated replay result

+1.4292 TST

Expected P&L on a 2.5200 TST simulated cost. This is supporting evidence, not competition performance, and no order was submitted.

Fail-closed by design

A weak edge becomes no trade.

Dry-run is the default. Live execution needs two explicit switches, a registered and funded competition wallet, passing evidence, and an executable quote inside every cap.

Read the architecture and safety boundaries →

  • Primary-source rules
    Exact sources, timestamps, thresholds, and outcomes are reviewed before use.
  • Quote-aware sizing
    Candidate sizes must remain profitable after price impact and portfolio limits.
  • Restart-safe receipts
    Duplicate and ambiguous-order protection survives watcher restarts.
  • Read-only preflight
    Nine launch gates are reported separately without approving or submitting an order.