ml-experiment-evaluation
Plan evaluation strategies for machine-learning product changes. Use when deciding between offline evaluation, interleaving, online A/B tests, multi-armed bandits, or model filtering for ranking, recommendation, search, personalization, or other ML-powered user experiences.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- ml-experiment-evaluation
- license
- MIT
- metadata
- { "tags": "practical-ab-testing,next-level-ab-testing,ab-testing,experimentation,machine-learning", "version": "0.1.0", "category": "Product Management", "displayName": "ML Experiment Evaluation" }
- description
- Plan evaluation strategies for machine-learning product changes. Use when deciding between offline evaluation, interleaving, online A/B tests, multi-armed bandits, or model filtering for ranking, recommendation, search, personalization, or other ML-powered user experiences.
- compatibility
- Codex, Claude Code, and other Agent Skills-compatible clients.
02 资源文件 5
- 参考文档guidelines.md1.7 KB
- 参考文档references/core/examples.md2.3 KB
- 参考文档references/core/knowledge.md3.2 KB
- 参考文档references/core/rules.md2.4 KB
- 参考文档workflows/choose-ml-evaluation-strategy.md2.5 KB
03 SKILL.md 正文
ML Experiment Evaluation
Use this skill to choose how to evaluate machine-learning product changes before they consume live experiment traffic or affect users. It focuses on offline evaluation, offline-online correlation, interleaving, model filtering, and when classic A/B testing or adaptive strategies are justified.
Source Traceability
Primary source: Next-Level A/B Testing by Leemay Nassery. Guidance is transformed and paraphrased from Chapter 4 on offline evaluation, offline-online correlation, multi-armed bandits, and interleaving for rankers.
Related skills:
experiment-sensitivity-optimizationfor reducing live variants and traffic.adaptive-experimentation-strategyfor bandits and dynamic allocation.ab-test-design-brieffor standard online A/B test planning.
Reference Routing
| Need | Read |
|---|---|
| ML evaluation concepts | references/core/knowledge.md |
| Selection and validation rules | references/core/rules.md |
| Evaluation strategy examples | references/core/examples.md |
| Step-by-step evaluation plan | workflows/choose-ml-evaluation-strategy.md |
Workflow
- State the model change and product decision.
- Identify the user harm or trust risk if a poor model reaches production.
- Choose the lowest-cost evaluation that can filter bad candidates.
- Check offline metrics and whether they correlate with online outcomes.
- Use interleaving when ranker comparison needs high sensitivity with fewer users.
- Escalate to online A/B testing or adaptive testing only when live evidence is needed and infrastructure can support it.
Output Format
# ML Evaluation Strategy
## Model Decision
[What model or ranking decision must be made.]
## Recommended Evaluation Path
[Offline only | Offline then A/B | Interleaving | A/B test | Adaptive strategy]
## Why
- Product risk:
- Offline signal available:
- Online evidence needed:
- Traffic or capacity constraint:
## Metrics
| Metric | Offline/Online | Role | Concern |
|--------|----------------|------|---------|
## Implementation Notes
- Data needed:
- Logging needed:
- Correlation check:
- Rollout guardrails:
Quality Bar
- Do not send poor offline candidates to live users just to get online evidence.
- Do not trust offline metrics until their relationship to online outcomes is understood.
- Do not use interleaving unless the product has a ranking or choice context where attribution can be logged.
- Do not recommend adaptive methods without checking data freshness, observability, and operational ownership.
04 同仓库更多 Skills
worktable-review-with-annotations
Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.
worktable-find-and-synthesize-context
Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.
worktable-create-or-update-docs
Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.
worktable-create-or-manage-records
Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.
worktable-create-or-manage-interactive-html
Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.
worktable-collaborate-threads
Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.