CLM-CALIBRATION-001·hypothesis·open
Back to research agenda
Q-060M-001 · Truth & KnowledgeT-005 · World Models & Decision under Uncertaintyscientific
人类如何发现自己的世界模型正在失效，并校准‘我们可能错了’的概率？
This is a canonical research question in the X-LAB three-layer agenda. It connects upward to an era theme and millennium question, and downward to claims, evidence, experiments, systems, decisions and reality feedback.
Current evolution state
This state is deterministically derived from public / graph-safe authoritative projections. It is not a new source of truth or a scientific confidence score.
READINESS 0%
Research state
No evidence yetNo public Research OS evidence object linked yet.
Execution state
Not exportedMission Control has no public graph-safe mission projection for this question.
Decision state
Not exportedNo permission-safe YICE Decision has been exported for this question.
Reality state
Not exportedNo graph-safe YOUNION project, pilot, collaboration or outcome has been exported for this question.
Biggest current gap
No authoritative Research OS evidence object is publicly linked to this question yet.
Next critical move
Define or link the first evidence-bearing experiment or Research Package before making stronger claims.
Current claims & evidence
Claim status and confidence are provisional research judgments, not truth labels. Evidence may support, challenge or provide context.
Falsification-oriented experiment protocols
These are experiment designs, not results. Claim status can change only after preregistration, execution and a distinct result/evidence record.
EXP-CALIBRATION-CLAIM-001·CLM-CALIBRATION-001·proposed·preregistration required
World-Model Failure Early-Warning Backtest
选取具有已知结构突变的历史数据窗口，冻结预测模型，持续记录预测残差、结构断裂、跨源冲突和置信度校准指标，检验其是否能在显著性能崩溃之前发出稳定信号。
Primary metric
预警提前量 × 命中率 × 假阳性率的综合表现
Supports
一组预注册信号能跨多个历史窗口稳定早于模型性能崩溃出现，并保持可接受假阳性率。
Challenges
这些信号只能与崩溃同步或事后出现，或假阳性率高到无法用于决策。
Inconclusive
历史标签本身不可靠、模型选择偏差或数据泄漏使提前预警无法有效判定。
• 历史回测不等于未来可预测。
• 结构突变标签可能带有事后解释偏差。
Supported relationships
containsera_theme
World Models & Decision under Uncertainty
Era theme contains this canonical research question.
Exploreproposesclaim
Model failure may be detected early through prediction residuals, structural breaks, cross-source conflict and confidence miscalibration.
Research question proposes or tracks this explicit epistemic claim.
Explorerelated toprogram
Civilization World Model & Decision Systems
Q-060 anchors model calibration and failure detection.
ExploreCONTRIBUTE
Bring a resource that can move this question forward.
Experts, researchers, institutions, data, facilities, pilot environments and research funding can enter the ecosystem around this canonical ID.
Contribute to Q-060