WHAT IT ANSWERS
这张表适合回答什么
- 比较模型提示版本表现
- 查看多空 Call 胜率
- 分析前向收益漂移
DATASET · research
Daybreak Judge 前向评估
research.daybreak_judge_forward_eval
按模型和提示版本评估已登记多空 Call 的胜率与前向漂移；不把 watch 或 no-trade 决策混入准确率。
字段27
关键字段4
更新口径producer_schedule
时间列
eval_date单页上限5000
分页窄筛选
PIT不支持
规模<1K
GRAIN & TIME
一行代表什么
- 业务粒度
- 每行代表“Daybreak Judge 前向评估”的一条记录；逻辑粒度由 eval_date × model_id × prompt_version × temperature 共同确定。
- 时间口径
- 以 eval_date 为主要查询时间轴；除下方限制特别说明外，时间戳按 UTC、日期按来源市场日历解释。
- 更新节奏
- 由对应生产任务或事件触发更新，不承诺固定日历频率；查询前先检查最新时间。
KEY FIELDS
先看这些字段
eval_date- eval date 对应的日历日期。
model_id- 用于识别 model 的业务标识。
prompt_version- 用于确定该表业务粒度或默认顺序的 prompt version 字段。
temperature- 用于确定该表业务粒度或默认顺序的 temperature 字段。
RECOMMENDED JOINS
怎样和其他表关联
该表主要按自身业务键使用，目前没有推荐的跨表主键联接。
HOW TO QUERY SAFELY
查询方法与已知限制
先用 catalog 确认当前契约；可按 catalog 中列出的 filters 进一步缩小范围；单页最多 5000 行，该表不提供稳定 cursor，结果过大时必须缩小标的、日期或其他筛选。
- 返回范围仍受 Friends 权限、单页上限和 catalog 中的筛选契约约束。
MACHINE CONTRACT
筛选、排序与分页
- 默认排序
eval_date, model_id, prompt_version, temperature- 唯一键
—- 读取模式
canonical_final_active- 最大时间窗
- 未单独限制
可用 filters
drifted→ drifted · integermodel_id→ model_id · stringprompt_version→ prompt_version · stringPOINT IN TIME
可复现读取
该资产没有 PIT 契约。普通 rows 端点仍按当前公开读模式返回数据；不要把它表述成历史时点快照。
COPY-READY CALLS
怎么调用
同一数据集可用 HTTP、Python SDK、CLI 或 MCP；Key 始终只通过环境变量/Bearer Header 传递。
HTTP
curl --fail-with-body -H "Authorization: Bearer $FRIENDS_DATA_API_KEY" "https://api.alphahubs.uk/api/v1/data/datasets/research/daybreak_judge_forward_eval/rows?limit=100&start=2026-08-01&end=2026-08-01"PYTHON SDK
from friends_data_api import FriendsDataClient, Settings
async with FriendsDataClient(Settings.from_env()) as client:
page = await client.get_dataset_rows('research', 'daybreak_judge_forward_eval', limit=100, start='2026-08-01', end='2026-08-01')
print(page.data)
if page.has_more:
print("next cursor:", page.next_cursor)CLI
friends-data get-dataset-rows research daybreak_judge_forward_eval --limit 100 --start 2026-08-01 --end 2026-08-01MCP TOOL CALL
{
"arguments": {
"dataset": "daybreak_judge_forward_eval",
"end": "2026-08-01",
"filters": {},
"limit": 100,
"namespace": "research",
"start": "2026-08-01"
},
"tool": "get_dataset_rows"
}FIELD DICTIONARY
27 个返回字段
descriptionSource 可在 JSON 中区分仓库注释、标准术语表和名称推导说明。
| # | 字段 | ClickHouse 类型 | JSON | 可空 | 角色 | 说明 |
|---|---|---|---|---|---|---|
| 1 | model_id | LowCardinality(String) | string | no | filter, default_order, warehouse_primary_key | Identifier for model.名称推导 |
| 2 | prompt_version | LowCardinality(String) | string | no | filter, default_order, warehouse_primary_key | Source-normalized field: prompt version.名称推导 |
| 3 | resolved_model | String | string | no | value | Source-normalized field: resolved model.名称推导 |
| 4 | model_fingerprint | String | string | no | value | Source-normalized field: model fingerprint.名称推导 |
| 5 | model_identity_source | LowCardinality(String) | string | no | value | Source-normalized field: model identity source.名称推导 |
| 6 | temperature | Float64 | number | no | default_order, warehouse_primary_key | Source-normalized field: temperature.名称推导 |
| 7 | eval_datedate | Date | string / date | no | time, default_order, warehouse_primary_key | Calendar date for eval date (YYYY-MM-DD).名称推导 |
| 8 | eval_tsUTC_datetime | DateTime64(3, 'UTC') | string / date-time | no | value | UTC timestamp for eval ts.名称推导 |
| 9 | eval_run_id | String | string | no | value | Identifier for eval run.名称推导 |
| 10 | n_total | UInt32 | integer | no | value | Count of total.名称推导 |
| 11 | n_forward | UInt32 | integer | no | value | Count of forward.名称推导 |
| 12 | n_dropped_preregistration | UInt32 | integer | no | value | Count of dropped preregistration.名称推导 |
| 13 | n_scored | UInt32 | integer | no | value | Count of scored.名称推导 |
| 14 | accuracy | Nullable(Float64) | number | yes | value | Source-normalized field: accuracy.名称推导 |
| 15 | cohort_is_stale | UInt8 | integer | no | value | Source-normalized field: cohort is stale.名称推导 |
| 16 | accuracy_muzzled | UInt8 | integer | no | value | Source-normalized field: accuracy muzzled.名称推导 |
| 17 | determinism_probe_repeats | UInt8 | integer | no | value | Source-normalized field: determinism probe repeats.名称推导 |
| 18 | determinism_probe_agreement | Nullable(Float64) | number | yes | value | Source-normalized field: determinism probe agreement.名称推导 |
| 19 | determinism_probe_status | LowCardinality(String) | string | no | value | Source-normalized field: determinism probe status.名称推导 |
| 20 | drifted | UInt8 | integer | no | filter | Source-normalized field: drifted.名称推导 |
| 21 | drift_delta | Nullable(Float64) | number | yes | value | Source-normalized field: drift delta.名称推导 |
| 22 | drift_tolerance | Float64 | number | no | value | Source-normalized field: drift tolerance.名称推导 |
| 23 | prev_accuracy | Nullable(Float64) | number | yes | value | Source-normalized field: prev accuracy.名称推导 |
| 24 | drift_reason | String | string | no | value | Source-normalized field: drift reason.名称推导 |
| 25 | version | UInt64 | integer | no | provenance | Monotonic row version used for correction-aware reads.标准术语表 |
| 26 | version_tsUTC_datetime | DateTime64(3, 'UTC') | string / date-time | no | provenance | UTC timestamp associated with the row version.标准术语表 |
| 27 | is_deleted | UInt8 | integer | no | value | Soft-delete marker; public serving excludes current tombstones.标准术语表 |
没有匹配字段
换一个字段名、类型或角色。
结构来自 Friends 只读身份可见的真实 serving relation；权限、过滤、排序、PIT 与分页来自可执行 DatasetSpec。Schema observed at 2026-10-06T18:08:52Z。