jyotint-sealed-corpus
收藏资源简介:
JYOTINT™ 密封预测语料库是一个公开评级、可证伪的预测记录数据集,采用加密预测预注册技术(基于OpenTimestamps/Bitcoin锚定的完整语料库清单)确保防篡改。该数据集覆盖发射任务保证(公开评级的发射记录:23次呼叫,结果由公开事实裁定)和地缘政治情报与预警(包括在克罗库斯市政厅袭击前约200天密封的大规模伤亡预警、在瓦格纳兵变相关政权稳定性呼叫前约110天的预测、2024年印度选举结果前约299天的预测,以及新格伦NG-3上面级失败在飞行前一天的命名)。所有咨询均在其预测窗口开启前上传至公共基础设施(YouTube/X),形成时间戳和SHA-256哈希的防篡改密封对。事后,每个已关闭咨询均根据密封时固定的评分标准进行评级(HIT/NEAR/PARTIAL/MISS),无追溯调整。数据集包含92条记录(23个发射咨询和69个情报咨询),结果分布为73个HIT、6个NEAR、4个PARTIAL、9个MISS。数据模式统一,每条记录包含ID、标题、预测文本(claim)、准确性评级、密封时概率、密封日期、实现日期、提前天数、密封哈希等字段。此外,提供了一个选举切片(corpus-elections.jsonl),包含34个已评级的选举预测(美国2022/2024、印度2024、马哈拉施特拉2024、巴西2022、英国2022),保持相同模式。数据集适用于文本分类、预测校准、可证伪性分析、地缘政治情报评估等任务,并强调其作为可验证预测注册表的边界(非国家安全决策输入)。数据采用CC-BY-4.0许可证,由Vijay Jyotish(Vijay Jyotish LLC)操作,可通过DOI(10.5281/zenodo.20630257)引用。
The JYOTINT™ Sealed Prediction Corpus is a publicly-rated, falsifiable dataset of prediction records, employing encrypted prediction pre-registration technology (based on OpenTimestamps/Bitcoin-anchored full corpus manifest) to ensure tamper-proofing. It covers launch mission assurance (publicly-rated launch records: 23 calls, outcomes adjudicated by public facts) and geopolitical intelligence & warning (including a mass casualty warning sealed approximately 200 days before the Crocus City Hall attack, a prediction about regime stability calls related to the Wagner mutiny sealed about 110 days prior, a prediction for the 2024 Indian election results sealed about 299 days in advance, and the naming of the New Glenn NG-3 upper stage failure one day before flight). All advisories are uploaded to public infrastructure (YouTube/X) before their prediction windows open, forming tamper-proof seal pairs of timestamps and SHA-256 hashes. Afterward, each closed advisory is rated (HIT/NEAR/PARTIAL/MISS) based on fixed scoring criteria established at sealing, with no retroactive adjustments. The dataset contains 92 records (23 launch advisories and 69 intelligence advisories), with outcome distribution of 73 HITs, 6 NEARs, 4 PARTIALs, and 9 MISSes. The data schema is uniform, with each record including fields such as ID, title, prediction text (claim), accuracy rating, probability at sealing, sealing date, realization date, days ahead, and sealing hash. Additionally, an elections slice (corpus-elections.jsonl) is provided, containing 34 rated election predictions (US 2022/2024, India 2024, Maharashtra 2024, Brazil 2022, UK 2022), maintaining the same schema. The dataset is suitable for tasks like text classification, prediction calibration, falsifiability analysis, and geopolitical intelligence assessment, and emphasizes its boundary as a verifiable prediction registry (not as input for national security decisions). The data is licensed under CC-BY-4.0, operated by Vijay Jyotish (Vijay Jyotish LLC), and citable via DOI (10.5281/zenodo.20630257).
JYOTINT™ Sealed Forecast Corpus 数据集详情
数据集概览
该数据集是一个公开可验证、可伪造的预测记录库,所有预测在事件发生前通过加密方式封存(基于OpenTimestamps/比特币锚定),事件结束后按照封存时的评分标准进行评定,且保留所有失败预测。涵盖发射任务保障和地缘政治预警两大类,共92条记录。
数据规模与组成
- 总记录数:92条
- 发射咨询(LA):23条
- 情报咨询(IA):69条(涵盖俄乌、美国2024/2022、巴西2022、英国2022、印度2024、市场、马哈拉施特拉邦)
- 评定结果分布:73 HIT · 6 NEAR · 4 PARTIAL · 9 MISS
- 数据集大小:n<1K
数据配置与格式
- 配置文件:
default:主数据集,数据文件为corpus.jsonlelections:选举子集,数据文件为corpus-elections.jsonl,包含34条已评定的选举预测(25 HIT · 3 NEAR · 6 MISS)
- 数据格式:JSON Lines,每条记录包含以下字段:
| 字段 | 类型 | 描述 |
|---|---|---|
id |
string | 咨询编号(LA-NNN或IA-XX-NNN) |
title |
string | 描述性目录标签 |
claim |
string | 封存时的预测原文(SHA-256哈希证明其在事件前存在) |
accuracy |
enum | 评定结果:HIT / NEAR / PARTIAL / MISS / OPEN |
prob_at_seal |
number | null |
ipcc_band |
string | null |
outcome_summary |
string | null |
sealed_on |
string (ISO日期) | 公开上传日期 |
materialized_on |
string | null |
lead_days |
number | null |
lead_time_label |
string | null |
url |
string (URL) | 咨询页面链接 |
sources |
string[] | 结果证据来源URL |
artifact_url |
string (URL) | YouTube/X公开封存载体 |
video_id |
string | null |
object_id |
string | 哈希种子来源对象ID |
object_type |
string | 对象类型(youtube / x / other) |
seal_hash_sha256 |
string (hex) | SHA-256哈希值 |
hash_input_template |
string | 哈希模板格式 |
iy |
object | null |
sita |
object | null |
warning_timeline |
object | null |
license |
string | CC-BY-4.0 |
operator |
string | Vijay Jyotish |
entity |
string | Vijay Jyotish LLC |
核心特性
- 可验证性:每条记录的
seal_hash_sha256可离线复现,完整语料库通过OpenTimestamps锚定到比特币区块链 - 特异性:68条已评定预测中61条在封存文本中命名了因果机制,平均覆盖6.4/7个维度(谁/什么/哪里/何时/为什么/如何),中位提前68天
- 诚实性:MISS结果保留在分母中,未过滤;选举子集中马哈拉施特拉邦2024和美国2024的下级选举失败均保留
使用与引用
- 许可证:CC-BY-4.0
- 正确使用方式:作为可验证的预测前注册与早期预警信号,而非国家安全决策输入
- 引用方式:可通过Zenodo DOI 10.5281/zenodo.20630257 引用
- 验证命令:
- 单条哈希验证:
node -e console.log(require("crypto").createHash("sha256").update("OBJECT_ID|SEALED_ON|TITLE|CLAIM").digest("hex")) - 完整语料库检查:
wget https://jyotishintelligence.com/verify-jyotint.mjs && node verify-jyotint.mjs
- 单条哈希验证:
数据集限制
- 方法基础为吠陀占星术
- Brier分数为自评,与基线率持平(不证明校准技能)
- 样本量小且由操作者按固定规则选择(排除时间不可判定、预防悖论、医疗责任、伦理等类别)
- 包含一条NSSL运载火箭预测(Vulcan Cert-1/LA-003),但无国家安全载荷预测




