Lesson overview | Previous part | Next part

Online Experimentation and AB Testing: Part 4: Inference for AB Tests

4. Inference for AB Tests

Inference for AB Tests is the part of online experimentation and ab testing that turns the approved TOC into a concrete learning path. The subsections below keep the focus on Chapter 17's canonical job: measurement, reliability, uncertainty, and decision support for AI systems.

4.1 Difference in means and proportions

Difference in means and proportions is part of the canonical scope of online experimentation and ab testing. In this chapter, the object under study is not merely a dataset or a model, but the full online randomized experiment: the items, prompts, outputs, graders, uncertainty statements, and decision rules that turn model behavior into evidence.

The basic mathematical pattern is an empirical estimator. For a model or system $m$ evaluated on items $z_1,\ldots,z_n$, the local estimate is written

$$\widehat{\operatorname{ATE}} = \frac{1}{n}\sum_{i=1}^n Y_i(1)-Y_i(0).$$

The formula is intentionally simple. The difficulty lies in deciding what counts as an item, which loss or score is meaningful, whether the items are independent, and whether the estimate answers the real product or research question. For difference in means and proportions, those choices determine whether the reported number is evidence or decoration.

A useful invariant is that every evaluation claim should be reproducible as a tuple $(m,\mathcal{T},\pi,g,\rho)$, where $m$ is the system, $\mathcal{T}$ is the task sample, $\pi$ is the prompt or intervention policy, $g$ is the grader, and $\rho$ is the aggregation rule. If any part of this tuple is missing, the number cannot be audited.

Examples of correct use:

• Report difference in means and proportions with item count, prompt protocol, grader version, and a confidence interval.
• Use paired comparisons when two models answer the same evaluation items.
• Inspect at least one meaningful slice before concluding that the aggregate result is reliable.
• Store raw outputs so future graders can be replayed without querying the model again.
• Document whether the metric is measuring capability, reliability, user value, or risk.

Non-examples:

• A leaderboard point estimate without sample size.
• A benchmark score produced with an undocumented prompt template.
• A model-graded result without judge identity, rubric, or agreement check.
• A robustness claim measured only on the easiest in-distribution examples.
• An online win declared before the randomization and logging checks pass.

Worked evaluation pattern for difference in means and proportions:

1. Define the evaluation population in words before writing code.
2. Choose the smallest metric set that answers the decision question.
3. Compute the point estimate and an uncertainty statement together.
4. Run a slice or paired analysis to check whether the aggregate hides structure.
5. Archive raw outputs, scores, and seeds before changing the prompt or grader.

For AI systems, difference in means and proportions is especially delicate because the same model can be used with many prompts, decoding policies, tools, retrieval contexts, and safety filters. The measured quantity is therefore a property of the system configuration, not just the base weights.

Implementation checklist:

• Use deterministic seeds for synthetic or sampled evaluation subsets.
• Print metric denominators, not only percentages.
• Keep missing, invalid, timeout, and refusal outcomes explicit.
• Prefer typed result records over loose CSV columns.
• Separate raw model outputs from normalized grader inputs.
• Track the smallest reproducible command that generated the result.
• Record whether the estimate is item-weighted, token-weighted, user-weighted, or domain-weighted.
• Write the decision rule before seeing the final score whenever the result will guide a release.

The mathematical habit to build is skepticism with structure. A score is not ignored because it is noisy; it is interpreted through the design that produced it. Difference in means and proportions is one place where that habit becomes concrete.

4.2 t tests and z tests

t tests and z tests is part of the canonical scope of online experimentation and ab testing. In this chapter, the object under study is not merely a dataset or a model, but the full online randomized experiment: the items, prompts, outputs, graders, uncertainty statements, and decision rules that turn model behavior into evidence.

The basic mathematical pattern is an empirical estimator. For a model or system $m$ evaluated on items $z_1,\ldots,z_n$, the local estimate is written

$$\widehat{\operatorname{ATE}} = \frac{1}{n}\sum_{i=1}^n Y_i(1)-Y_i(0).$$

The formula is intentionally simple. The difficulty lies in deciding what counts as an item, which loss or score is meaningful, whether the items are independent, and whether the estimate answers the real product or research question. For t tests and z tests, those choices determine whether the reported number is evidence or decoration.

A useful invariant is that every evaluation claim should be reproducible as a tuple $(m,\mathcal{T},\pi,g,\rho)$, where $m$ is the system, $\mathcal{T}$ is the task sample, $\pi$ is the prompt or intervention policy, $g$ is the grader, and $\rho$ is the aggregation rule. If any part of this tuple is missing, the number cannot be audited.

Examples of correct use:

• Report t tests and z tests with item count, prompt protocol, grader version, and a confidence interval.
• Use paired comparisons when two models answer the same evaluation items.
• Inspect at least one meaningful slice before concluding that the aggregate result is reliable.
• Store raw outputs so future graders can be replayed without querying the model again.
• Document whether the metric is measuring capability, reliability, user value, or risk.

Non-examples:

• A leaderboard point estimate without sample size.
• A benchmark score produced with an undocumented prompt template.
• A model-graded result without judge identity, rubric, or agreement check.
• A robustness claim measured only on the easiest in-distribution examples.
• An online win declared before the randomization and logging checks pass.

Worked evaluation pattern for t tests and z tests:

1. Define the evaluation population in words before writing code.
2. Choose the smallest metric set that answers the decision question.
3. Compute the point estimate and an uncertainty statement together.
4. Run a slice or paired analysis to check whether the aggregate hides structure.
5. Archive raw outputs, scores, and seeds before changing the prompt or grader.

For AI systems, t tests and z tests is especially delicate because the same model can be used with many prompts, decoding policies, tools, retrieval contexts, and safety filters. The measured quantity is therefore a property of the system configuration, not just the base weights.

Implementation checklist:

• Use deterministic seeds for synthetic or sampled evaluation subsets.
• Print metric denominators, not only percentages.
• Keep missing, invalid, timeout, and refusal outcomes explicit.
• Prefer typed result records over loose CSV columns.
• Separate raw model outputs from normalized grader inputs.
• Track the smallest reproducible command that generated the result.
• Record whether the estimate is item-weighted, token-weighted, user-weighted, or domain-weighted.
• Write the decision rule before seeing the final score whenever the result will guide a release.

The mathematical habit to build is skepticism with structure. A score is not ignored because it is noisy; it is interpreted through the design that produced it. t tests and z tests is one place where that habit becomes concrete.

4.3 Bootstrap confidence intervals

Bootstrap confidence intervals is part of the canonical scope of online experimentation and ab testing. In this chapter, the object under study is not merely a dataset or a model, but the full online randomized experiment: the items, prompts, outputs, graders, uncertainty statements, and decision rules that turn model behavior into evidence.

The basic mathematical pattern is an empirical estimator. For a model or system $m$ evaluated on items $z_1,\ldots,z_n$, the local estimate is written

$$\widehat{\operatorname{ATE}} = \frac{1}{n}\sum_{i=1}^n Y_i(1)-Y_i(0).$$

The formula is intentionally simple. The difficulty lies in deciding what counts as an item, which loss or score is meaningful, whether the items are independent, and whether the estimate answers the real product or research question. For bootstrap confidence intervals, those choices determine whether the reported number is evidence or decoration.

A useful invariant is that every evaluation claim should be reproducible as a tuple $(m,\mathcal{T},\pi,g,\rho)$, where $m$ is the system, $\mathcal{T}$ is the task sample, $\pi$ is the prompt or intervention policy, $g$ is the grader, and $\rho$ is the aggregation rule. If any part of this tuple is missing, the number cannot be audited.

Examples of correct use:

• Report bootstrap confidence intervals with item count, prompt protocol, grader version, and a confidence interval.
• Use paired comparisons when two models answer the same evaluation items.
• Inspect at least one meaningful slice before concluding that the aggregate result is reliable.
• Store raw outputs so future graders can be replayed without querying the model again.
• Document whether the metric is measuring capability, reliability, user value, or risk.

Non-examples:

• A leaderboard point estimate without sample size.
• A benchmark score produced with an undocumented prompt template.
• A model-graded result without judge identity, rubric, or agreement check.
• A robustness claim measured only on the easiest in-distribution examples.
• An online win declared before the randomization and logging checks pass.

Worked evaluation pattern for bootstrap confidence intervals:

1. Define the evaluation population in words before writing code.
2. Choose the smallest metric set that answers the decision question.
3. Compute the point estimate and an uncertainty statement together.
4. Run a slice or paired analysis to check whether the aggregate hides structure.
5. Archive raw outputs, scores, and seeds before changing the prompt or grader.

For AI systems, bootstrap confidence intervals is especially delicate because the same model can be used with many prompts, decoding policies, tools, retrieval contexts, and safety filters. The measured quantity is therefore a property of the system configuration, not just the base weights.

Implementation checklist:

• Use deterministic seeds for synthetic or sampled evaluation subsets.
• Print metric denominators, not only percentages.
• Keep missing, invalid, timeout, and refusal outcomes explicit.
• Prefer typed result records over loose CSV columns.
• Separate raw model outputs from normalized grader inputs.
• Track the smallest reproducible command that generated the result.
• Record whether the estimate is item-weighted, token-weighted, user-weighted, or domain-weighted.
• Write the decision rule before seeing the final score whenever the result will guide a release.

The mathematical habit to build is skepticism with structure. A score is not ignored because it is noisy; it is interpreted through the design that produced it. Bootstrap confidence intervals is one place where that habit becomes concrete.

4.4 CUPED preview

CUPED preview is part of the canonical scope of online experimentation and ab testing. In this chapter, the object under study is not merely a dataset or a model, but the full online randomized experiment: the items, prompts, outputs, graders, uncertainty statements, and decision rules that turn model behavior into evidence.

The basic mathematical pattern is an empirical estimator. For a model or system $m$ evaluated on items $z_1,\ldots,z_n$, the local estimate is written

$$\widehat{\operatorname{ATE}} = \frac{1}{n}\sum_{i=1}^n Y_i(1)-Y_i(0).$$

The formula is intentionally simple. The difficulty lies in deciding what counts as an item, which loss or score is meaningful, whether the items are independent, and whether the estimate answers the real product or research question. For cuped preview, those choices determine whether the reported number is evidence or decoration.

A useful invariant is that every evaluation claim should be reproducible as a tuple $(m,\mathcal{T},\pi,g,\rho)$, where $m$ is the system, $\mathcal{T}$ is the task sample, $\pi$ is the prompt or intervention policy, $g$ is the grader, and $\rho$ is the aggregation rule. If any part of this tuple is missing, the number cannot be audited.

Examples of correct use:

• Report cuped preview with item count, prompt protocol, grader version, and a confidence interval.
• Use paired comparisons when two models answer the same evaluation items.
• Inspect at least one meaningful slice before concluding that the aggregate result is reliable.
• Store raw outputs so future graders can be replayed without querying the model again.
• Document whether the metric is measuring capability, reliability, user value, or risk.

Non-examples:

• A leaderboard point estimate without sample size.
• A benchmark score produced with an undocumented prompt template.
• A model-graded result without judge identity, rubric, or agreement check.
• A robustness claim measured only on the easiest in-distribution examples.
• An online win declared before the randomization and logging checks pass.

Worked evaluation pattern for cuped preview:

1. Define the evaluation population in words before writing code.
2. Choose the smallest metric set that answers the decision question.
3. Compute the point estimate and an uncertainty statement together.
4. Run a slice or paired analysis to check whether the aggregate hides structure.
5. Archive raw outputs, scores, and seeds before changing the prompt or grader.

For AI systems, cuped preview is especially delicate because the same model can be used with many prompts, decoding policies, tools, retrieval contexts, and safety filters. The measured quantity is therefore a property of the system configuration, not just the base weights.

Implementation checklist:

• Use deterministic seeds for synthetic or sampled evaluation subsets.
• Print metric denominators, not only percentages.
• Keep missing, invalid, timeout, and refusal outcomes explicit.
• Prefer typed result records over loose CSV columns.
• Separate raw model outputs from normalized grader inputs.
• Track the smallest reproducible command that generated the result.
• Record whether the estimate is item-weighted, token-weighted, user-weighted, or domain-weighted.
• Write the decision rule before seeing the final score whenever the result will guide a release.

The mathematical habit to build is skepticism with structure. A score is not ignored because it is noisy; it is interpreted through the design that produced it. CUPED preview is one place where that habit becomes concrete.

4.5 Heterogeneous treatment effects

Heterogeneous treatment effects is part of the canonical scope of online experimentation and ab testing. In this chapter, the object under study is not merely a dataset or a model, but the full online randomized experiment: the items, prompts, outputs, graders, uncertainty statements, and decision rules that turn model behavior into evidence.

The basic mathematical pattern is an empirical estimator. For a model or system $m$ evaluated on items $z_1,\ldots,z_n$, the local estimate is written

$$\widehat{\operatorname{ATE}} = \frac{1}{n}\sum_{i=1}^n Y_i(1)-Y_i(0).$$

The formula is intentionally simple. The difficulty lies in deciding what counts as an item, which loss or score is meaningful, whether the items are independent, and whether the estimate answers the real product or research question. For heterogeneous treatment effects, those choices determine whether the reported number is evidence or decoration.

A useful invariant is that every evaluation claim should be reproducible as a tuple $(m,\mathcal{T},\pi,g,\rho)$, where $m$ is the system, $\mathcal{T}$ is the task sample, $\pi$ is the prompt or intervention policy, $g$ is the grader, and $\rho$ is the aggregation rule. If any part of this tuple is missing, the number cannot be audited.

Examples of correct use:

• Report heterogeneous treatment effects with item count, prompt protocol, grader version, and a confidence interval.
• Use paired comparisons when two models answer the same evaluation items.
• Inspect at least one meaningful slice before concluding that the aggregate result is reliable.
• Store raw outputs so future graders can be replayed without querying the model again.
• Document whether the metric is measuring capability, reliability, user value, or risk.

Non-examples:

• A leaderboard point estimate without sample size.
• A benchmark score produced with an undocumented prompt template.
• A model-graded result without judge identity, rubric, or agreement check.
• A robustness claim measured only on the easiest in-distribution examples.
• An online win declared before the randomization and logging checks pass.

Worked evaluation pattern for heterogeneous treatment effects:

1. Define the evaluation population in words before writing code.
2. Choose the smallest metric set that answers the decision question.
3. Compute the point estimate and an uncertainty statement together.
4. Run a slice or paired analysis to check whether the aggregate hides structure.
5. Archive raw outputs, scores, and seeds before changing the prompt or grader.

For AI systems, heterogeneous treatment effects is especially delicate because the same model can be used with many prompts, decoding policies, tools, retrieval contexts, and safety filters. The measured quantity is therefore a property of the system configuration, not just the base weights.

Implementation checklist:

• Use deterministic seeds for synthetic or sampled evaluation subsets.
• Print metric denominators, not only percentages.
• Keep missing, invalid, timeout, and refusal outcomes explicit.
• Prefer typed result records over loose CSV columns.
• Separate raw model outputs from normalized grader inputs.
• Track the smallest reproducible command that generated the result.
• Record whether the estimate is item-weighted, token-weighted, user-weighted, or domain-weighted.
• Write the decision rule before seeing the final score whenever the result will guide a release.

The mathematical habit to build is skepticism with structure. A score is not ignored because it is noisy; it is interpreted through the design that produced it. Heterogeneous treatment effects is one place where that habit becomes concrete.