Model Trials模型评测记录
A transparent audit trail.透明的审计轨迹。
Future public trial records will connect model identity, case identity, execution context, Tutor response, evaluator evidence, and sanitized operational metrics.
未来公开评测记录会连接模型身份、案例身份、执行语境、Tutor 回复、评估器证据和经过清理的运行指标。
- 0
- Public trials公开评测记录 No released runs yet尚无已发布运行
- 0
- Public model profiles公开模型档案 Reserved预留
- 48
- Public cases公开案例 tutor-eval-v0.2a
- Not available不可用
- Trial publication评测记录发布 Public artifact state公开产物状态
Shared evidence.
Stronger teaching.共享证据,
改进教学。
Audit ledger审计 ledger
No row without a public artifact.没有公开产物,就没有记录行。
Each future row will be an addressable evidence record. The current ledger is intentionally empty.
未来每一行都将是一条可寻址的证据记录。当前 ledger 有意保持为空。
| Trial ID评测记录 ID | Model ID模型 ID | Case ID案例 ID | Run index运行索引 | Evidence status证据状态 |
|---|---|---|---|---|
No public trials yet.目前暂无公开评测记录。 No public model trials available yet.目前尚无公开模型评测记录。 Check back after public model runs are released.公开模型运行发布后再来查看。 | ||||
The ledger exposes only the public trial summary identity contract: id, modelId, caseId, and runIndex.该台账只公开评测记录摘要身份契约:id、modelId、caseId 和 runIndex。
Future trial schema未来评测记录结构
What a trial includes.一条评测记录包含什么。
These families are derived from the current public trial field contract, not from a model run.
这些类别来自当前公开评测记录字段契约,而不是某次模型运行。
- 1Identity & execution身份与执行
The versioned identities that make a run addressable.让一次运行可被准确寻址的版本化身份。
modelModel identity模型身份modelVersionModel version模型版本datasetVersionDataset version数据集版本caseVersionCase version案例版本runIndexRun index运行索引
- 2Generation traceability生成可追踪性
The configuration references needed to reproduce generation.复现生成所需的配置引用。
generationSpecIdGeneration spec ID生成规格 IDgenerationSpecVersionGeneration spec version生成规格版本promptVersionPrompt versionPrompt 版本promptSha256Prompt SHA-256Prompt SHA-256
- 3Tutor evidenceTutor 证据
The complete public Tutor response for the case.该案例完整的公开 Tutor 回复。
tutorResponseTutor responseTutor 回复
- 4Evaluation evidence评测证据
Dimension results and failure evidence attached to the response.与回复关联的维度结果和失败证据。
correctnessCorrectness正确性diagnosisDiagnosis诊断能力guidanceGuidance引导能力adaptationAdaptation适应能力actionabilityActionability可执行性rubricResultsRubric results评分标准结果criticalFailuresCritical failures严重失败answerLeakageAnswer leakage答案泄露
- 5Judge evidenceJudge 证据
The evaluator identity and prompt version, when published.发布时包含评估器身份与 prompt 版本。
judgeJudgeJudgejudgePromptVersionJudge prompt versionJudge prompt 版本
- 6Operational metrics运行指标
Sanitized run signals kept distinct from pedagogical scores.经过清理的运行信号与教学评分保持分离。
tokensTokenstoken 数latencyLatency延迟costCost成本
Traceability chain可追踪链
From benchmark to evidence.从基准走向证据。
Trials remain separate from model profiles and leaderboard eligibility. A future record will link these stages without implying a ranking or teaching-effectiveness claim.
评测记录与模型档案、排行榜资格保持分离。未来记录会连接这些阶段,但不会暗示排名或教学有效性主张。
- Benchmark version基准版本Future public evidence未来公开证据
- Model identity模型身份Future public evidence未来公开证据
- Trial评测记录Future public evidence未来公开证据
- Case案例Future public evidence未来公开证据
- Tutor responseTutor 回复Future public evidence未来公开证据
- Evaluation evidence评测证据Future public evidence未来公开证据
- Sanitized metrics清理后的指标Future public evidence未来公开证据
Public evidence, clearly bounded边界清晰的公开证据
Evidence you can follow.可以追踪的证据。
Future published results should remain traceable back to versioned cases, model identity, and the evidence that supports each claim.
未来公开结果应始终能够追踪到版本化案例、模型身份以及支撑每项主张的证据。