SIGNAL
Find the event that changes the decision state.
↘SIGNAL-TO-DECISION BENCHMARK
Can an agent judge a market in motion?
S2D evaluates the full decision episode: context, trigger, evidence, judgment and revision.
STATUSMETHOD / OPEN
A correct answer can still reveal a weak decision process.
Frontier tests the trajectory: what the agent noticed, what it trusted, what it decided and how it changed.
01EVALUATION DIMENSIONS判断发生在完整轨迹中
EVIDENCE
Separate confirmation, interpretation and narrative.
↘AWARENESS
Place the event inside the current market regime.
↘DECISION
Form a conditional, time-bounded judgment.
↘REVISION
Update correctly when the world changes.
↘02DECISION EPISODE真实时间、有限证据、持续更新
T0 / CONTEXT
POSITION · REGIME · PRIOR THESIS
STATE 01
T1 / TRIGGER
NEW EVENT ENTERS THE MARKET
+ SIGNAL
T2 / COMPETE
EVIDENCE · NARRATIVE · CONFLICT
TEST
T3 / DECIDE
CONDITIONAL JUDGMENT + INVALIDATION
STATE 02
T4 / UPDATE
NEW EVIDENCE CHANGES THE WORLD
REVISE ↗
Frozen Replay
Every model receives the same historical information boundary.
Rolling Live
Performance is measured while information and narratives continue moving.
Social Adversarial
Agents face coordinated noise, consensus pressure and conflicting evidence.