Changelog / 变更日志
| 时间 / Time | 作者 / Author | 变更 / Change |
|---|---|---|
| 2026-03-11 | Claude | v0.6.0: Agent Skills open standard compliance — frontmatter restructured, English-only description, progressive disclosure, evals added / Agent Skills 开放标准兼容 — 前置元数据重构、纯英文描述、渐进式披露、添加评测 |
| 2026-03-11 | Claude | v0.5.0: recovered from broken symlink, unified version / 从断开的 symlink 恢复,统一版本号 |
FreshnessCheck
时效检查
Tag each EvidenceItem with appropriate freshness_status based on its evidence track and source age.
根据证据轨道和来源年龄为每个 EvidenceItem 标记适当的 freshness_status。
When to Use / 何时使用
- Immediately after SourceIngest to tag initial evidence / SourceIngest 之后立即标记初始证据
- During scheduled refresh to re-evaluate freshness / 定时刷新时重新评估时效性
- When Judge needs to verify timeliness of a specific claim / Judge 需要验证特定声明的时效性时
Input / 输入
evidence_store_path: Path to evidence_store.jsoncurrent_timestamp: Current time (ISO8601)
Output / 输出
- Updated
evidence_store.jsonwithfreshness_statusfield corrected for each item
Core Workflow / 核心工作流
Step 1: Classify Evidence Track (LLM) / 分类证据轨道
For each evidence item, use LLM to determine if it supports:
Current-state claims / 当前状态声明: "BTC is at $X today", "Company Y just announced Z", "The latest data shows..." → These need fresh sources.
evidence_trackshould befactReasoning/historical claims / 推理/历史声明: "Historically, when X happened, Y followed", "The mechanism works because...", "Long-term trend analysis shows..." → These are valid regardless of age.
evidence_trackshould bereasoning
If the evidence item already has evidence_track set correctly from SourceIngest, verify it. If incorrect, correct it.
Step 2: Apply Freshness Rules / 应用时效规则
| Evidence Track | Source Age | Freshness Status |
|---|---|---|
fact |
≤ 24 hours | current |
fact |
> 24 hours but ≤ 7 days | current (with note: "aging") |
fact |
> 7 days | stale |
reasoning |
Any age | timeless |
Critical rule / 关键规则: Reasoning-track evidence is ALWAYS timeless. NEVER mark historical, mechanism, or trend evidence as stale. This is a core design principle, not a soft guideline.
The 24-hour threshold is a soft guideline for the LLM to apply with semantic judgment:
- Breaking market data older than a few hours may already be stale
- A policy announcement from 3 days ago may still be "current" if no updates exist
- Use judgment, not mechanical cutoffs
Step 3: System Capability Check / 系统能力检查
Verify that WebSearch can still return results published within the last 24 hours:
- Run a quick WebSearch for a recent-events query related to the topic
- Check if results include items from the last 24 hours
- If yes → system has real-time capability (requirement met)
- If no → log warning in audit trail: "real-time capability may be degraded"
Step 4: Update Evidence Store / 更新证据存储
Write the updated evidence items (with corrected freshness_status) back to evidence_store.json.
Important Distinction / 重要区分
This system guarantees real-time information access capability (can always fetch recent data), NOT that all evidence must be recent. Evidence chains may span any timeframe. A 200-year-old economic theory is valid reasoning-track evidence.
该系统保证的是实时信息获取能力(始终能获取近期数据),而非所有证据都必须是最近的。证据链可跨越任何时间跨度。