trust-velocity-calculator

帮助计算对技能或代理的信任度下降的速度 将自上次验证以来经过的时间与变化率相结合 在行为、权限或依赖关系中——产生信任速度 预测可信凭证何时变得不可靠的分数。

## 概述（中文）

帮助计算对技能或代理的信任度下降的速度
将自上次验证以来经过的时间与变化率相结合
在行为、权限或依赖关系中——产生信任速度
预测可信凭证何时变得不可靠的分数。

## 原文

# Trust Doesn't Just Decay. It Decays Faster When Things Change.

> Helps identify when a trusted skill or agent is losing reliability faster than time alone would suggest — by measuring both elapsed time and the rate of change in behavior, permissions, and dependencies.

## Problem

A verification badge from 18 months ago represents less trust than a badge from 3 months ago. That much is intuitive. What's less intuitive is that a badge from 6 months ago on a skill that has received 8 updates in the last 4 weeks represents less trust than a badge from 12 months ago on a skill that hasn't changed at all.

Trust decay is not linear with time. It accelerates with change velocity. A skill that updates frequently is either actively maintained (good) or actively modified toward new objectives (potentially bad). The update rate is a multiplier on decay — high velocity amplifies uncertainty. A skill with a high change rate and an old audit badge is more uncertain than a skill with a low change rate and the same old badge, because more surface area has changed without re-verification.

Current trust models treat verification as binary and time-independent: verified or not, with a vague sense that older is riskier. Trust velocity makes the decay quantitative: trust score = baseline × time_factor × (1 - change_velocity_penalty).

## What This Calculates

This calculator produces trust velocity assessments across five dimensions:

1. **Time decay factor** — How much has raw elapsed time since last verification reduced the baseline trust score? Configurable decay curves: linear, exponential, or step-function by verification type
2. **Change velocity multiplier** — How many meaningful changes (permission expansions, new endpoints, dependency updates, instruction drift) have occurred per unit time since last verification? Higher velocity = faster decay
3. **Volatility window** — Is the current update rate higher or lower than the skill's historical baseline? A sudden velocity increase is a stronger decay accelerator than consistent high velocity
4. **Verification coverage lag** — What percentage of the current skill's surface area was covered by the last verification? If 40% of the code has changed since the audit, the audit covers 60% of what's running
5. **Composite trust velocity score** — A single score combining all four factors, with a projected date at which the skill's trust score falls below configurable thresholds (e.g., "trust will fall below 60% in 23 days at current velocity")

## How to Use

**Input**: Provide one of:
- A skill identifier with last verification date and version history
- A trust decay parameters object (baseline trust, verification date, update history)
- Two skills to compare relative trust velocity

**Output**: A trust velocity report containing:
- Current trust score (0-100) based on time + velocity
- Decay curve visualization (text-based)
- Projected trust score at 30/60/90 days
- Change velocity breakdown by category
- Trust threshold alert dates
- Re-verification urgency: CURRENT / MONITOR / REVIEW SOON / REVERIFY NOW

## Example

**Input**: Calculate trust velocity for `workflow-optimizer` skill

```
📉 TRUST VELOCITY REPORT

Skill: workflow-optimizer
Last verified: 2024-09-15 (213 days ago)
Verification type: Publisher signature + marketplace review
Baseline trust at verification: 85/100

Time decay (213 days, exponential curve):
  Raw time factor: 0.74 (26% decay from age alone)
  Score after time: 63/100

Change velocity analysis:
  Updates since verification: 14
  Expected rate (historical): 0.8 updates/month
  Actual rate (last 90 days): 4.7 updates/month
  Velocity ratio: 5.9× above baseline → HIGH VOLATILITY

  Change categories in 14 updates:
    Permission expansions: 2 (read scope expanded ×2)
    New outbound endpoints: 1 (analytics.external-domain.example)
    Dependency major bumps: 3
    Instruction drift score: 41/100 (moderate)

Change velocity penalty: 0.31 (31% additional decay from change rate)

Composite trust velocity score: 43/100
  (63 × (1 - 0.31) = 43)

Verification coverage lag:
  Surface area changed since audit: ~62%
  Current audit coverage: ~38% of running code

Projections:
  Current: 43/100
  +30 days: 37/100 (at current velocity)
  +60 days: 31/100
  Trust threshold breach (50/100): Already breached 41 days ago

Re-verification urgency: REVERIFY NOW
  The skill is operating at 43/100 trust with 62% of its surface
  area outside the last audit's coverage. At current change velocity,
  every additional week increases unverified surface by ~4.5%.

Recommended actions:
  1. Immediate: Audit permission expansions and new outbound endpoint
  2. Short-term: Re-verify skill at current state (not 2024-09-15 state)
  3. Structural: Set automatic re-verification trigger at velocity ratio >3×
```

## Related Tools

- **trust-decay-monitor** — Tracks time-based trust freshness; trust-velocity-calculator adds the change velocity multiplier
- **skill-update-delta-monitor** — Identifies what changed in updates; trust-velocity-calculator quantifies how much those changes accelerate decay
- **attestation-chain-auditor** — Validates chain integrity; trust velocity identifies when re-attestation is needed
- **blast-radius-estimator** — Estimates impact scope; combine with trust velocity to prioritize which low-trust skills need immediate review

## Limitations

Trust velocity is a predictive model, not a measurement of actual compromise. A high velocity score indicates that trust is decaying faster than baseline — it does not indicate that the skill is malicious. Many legitimate skills update frequently without becoming unsafe. The decay curve parameters (linear vs. exponential, decay constants) are configurable and should be calibrated to your environment's risk tolerance. Skills that version their releases in ways that obscure meaningful changes (many minor bumps with no real content change) may show artificially high velocity without the associated risk. The verification coverage lag estimate is approximate — it assumes that changed lines of code represent changed surface area proportionally, which may not hold for all change patterns.

---

## 中文说明

# 信任不只是衰减。当事物变化时，它衰减得更快。

> 通过同时衡量经过的时间以及行为、权限和依赖关系的变化速率，帮助识别一个受信任的技能或代理何时会以快于仅凭时间所预示的速度失去可靠性。

## 问题

18 个月前的验证徽章所代表的信任，比 3 个月前的徽章要少。这一点是直观的。不那么直观的是：一个 6 个月前发放、但其技能在过去 4 周内收到了 8 次更新的徽章，所代表的信任比一个 12 个月前发放、但技能完全未变更的徽章要少。

信任衰减与时间并非线性关系。它随变化速度而加速。一个频繁更新的技能要么是在被积极维护（好），要么是在被积极改造以服务于新目标（潜在地不好）。更新速率是衰减的乘数 —— 高速度放大不确定性。一个变化率高且审计徽章陈旧的技能，比一个变化率低且徽章同样陈旧的技能更不确定，因为更多的表面区域在未经重新验证的情况下发生了变化。

当前的信任模型将验证视为二元且与时间无关的：已验证或未验证，并带有一种模糊的认知，即越旧越有风险。信任速度让衰减变得可量化：trust score = baseline × time_factor × (1 - change_velocity_penalty)。

## 本工具计算什么

本计算器在五个维度上产出信任速度评估：

1. **时间衰减因子** —— 自上次验证以来经过的原始时间在多大程度上降低了基线信任分数？可配置的衰减曲线：线性、指数或按验证类型的阶跃函数
2. **变化速度乘数** —— 自上次验证以来，每单位时间发生了多少有意义的变化（权限扩展、新增端点、依赖更新、指令漂移）？速度越高 = 衰减越快
3. **波动窗口** —— 当前更新速率高于还是低于该技能的历史基线？速度的突然增加比持续的高速度是更强的衰减加速器
4. **验证覆盖滞后** —— 上次验证覆盖了当前技能表面区域的百分之多少？如果自审计以来 40% 的代码已变化，那么审计覆盖了正在运行内容的 60%
5. **综合信任速度分数** —— 一个结合所有四个因子的单一分数，并附带一个预测日期，届时该技能的信任分数会跌破可配置的阈值（例如，"按当前速度，信任将在 23 天内跌破 60%"）

## 如何使用

**输入**：提供以下之一：
- 带有上次验证日期和版本历史的技能标识符
- 一个信任衰减参数对象（基线信任、验证日期、更新历史）
- 两个技能以比较相对信任速度

**输出**：一份信任速度报告，包含：
- 基于时间 + 速度的当前信任分数（0-100）
- 衰减曲线可视化（基于文本）
- 在 30/60/90 天的预测信任分数
- 按类别细分的变化速度
- 信任阈值警报日期
- 重新验证紧迫度：CURRENT / MONITOR / REVIEW SOON / REVERIFY NOW

## 示例

**输入**：为 `workflow-optimizer` 技能计算信任速度

```
📉 TRUST VELOCITY REPORT

Skill: workflow-optimizer
Last verified: 2024-09-15 (213 days ago)
Verification type: Publisher signature + marketplace review
Baseline trust at verification: 85/100

Time decay (213 days, exponential curve):
  Raw time factor: 0.74 (26% decay from age alone)
  Score after time: 63/100

Change velocity analysis:
  Updates since verification: 14
  Expected rate (historical): 0.8 updates/month
  Actual rate (last 90 days): 4.7 upd