Machine Learning

Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails

Design adaptive-learning experiments by choosing a primary outcome, delayed retention check, power/MDE target, subgroup plan, attrition rule, ethics guardrails, and rollout gate.

status: reviewimportance: importantdifficulty 4/5math: graduateread: 31mlive demo

Back to Machine Learning Next: Adaptive Learning Impact Audits: Retention, Drift, Guardrails, Rollback (review)

Concept Structure

Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails

01Intuition

Start with the picture, metaphor, or geometric mechanism.

02Math

Make the objects explicit and connect them with notation.

03Code

Mirror the equations with runnable implementation details.

04Interactive Demo

Manipulate the mechanism and watch the idea respond.

3prerequisites

1next concepts

4related links

Learner Contract

What this page should let you do.

You are here becauseDesign adaptive-learning experiments by choosing a primary outcome, delayed retention check, power/MDE target, subgroup plan, attrition rule, ethics guardrails, and rollout gate.

This Machine Learning concept is the current idea: keep the same invariant visible across Intuition, Math, Code, Interactive Demo.

Before thisLearning System Policy Evaluation: Logs, Estimators, Guardrails, Promotion (review)

3 prerequisites listed; refresh them before leaning on the math or code.

Learning System Policy Evaluation: Logs, Estimators, Guardrails, Promotion (review)Causal Inference Basics: Correlation versus Intervention (review)Evaluation Pipelines

By the end4/4 sections ready | runnable code expected | live demo

Explain the mechanism, trace the main notation, and test one prediction in the live demo.

Do this firstIntuition

Read the intuition before the notation; the math should name a mechanism you already felt.

Then go nextAdaptive Learning Impact Audits: Retention, Drift, Guardrails, Rollback (review)

Follow this edge after making one prediction here; the next page should reuse the result, not restart the route.

Test the linkManipulate one control and predict the visible change.Then continue to Adaptive Learning Impact Audits: Retention, Drift, Guardrails, Rollback (review)

Claim/source review status

Substantive review recorded

2/2 claims have bounded review metadata; still check caveats and source scope.Metadata-derived; review may be AI-assisted. Not a human certification.

Claims2/2 reviewed

Sources6 cited

Codeattached

Demolive

Reviewed2026-07-05

Updatedpage 2026-07-05

Learning item flow

4/4 sections readyAsk about this Research room

ConceptLearner Outcome Experiment Design: Outcomes, Power, Retention, Guardr...Machine Learning EquationLearner Outcome Experiment Design: Outcomes, Power, Retention, Guardr...Exact equation CodeLearner Outcome Experiment Design: Outcomes, Power, Retention, Guardr...Exact runnable code DemoLearner Outcome Experiment Design: Outcomes, Power, Retention, Guardr...Interactive demo ClaimA learner-outcome experiment should name one primary outcome, its mea...Exact claim check SourceWhat Works Clearinghouse Procedures and Standards Handbook, Version 5.0Exact source

ConceptLearner Outcome Experiment Design: Outcomes, Power, Retention, GuardrailsMachine Learning

6 sources attachedLocal snapshot ready

concept:machine-learning/learner-outcome-experiment-design

Codenearby Predictbefore reveal Roomitem handoff

01IntuitionStart with the picture, metaphor, or geometric mechanism.02MathMake the objects explicit and connect them with notation.03CodeMirror the equations with runnable implementation details.04Interactive DemoManipulate the mechanism and watch the idea respond.

Intuition

Build the mental picture first so the rest of the page has something to attach to.

Section prompt

Policy evaluation asks whether a candidate adaptive-learning policy looks better than the current one. Learner outcome experiment design asks a more basic question:

What evidence would make us believe this actually helped learners?

That question has to be answered before rollout. Otherwise the system can optimize toward whatever is easiest to measure: clicks, completion, immediate correctness, or engagement. Those can be useful secondary signals, but they are not the same as durable learning.

A strong learner-outcome plan keeps these lanes separate:

primary outcome: the one result the experiment is powered to detect,
outcome timing: baseline, immediate post-test, delayed retention, and any longer follow-up,
instrument: the assessment, rubric, or administrative measure used consistently across arms,
direction: whether higher, lower, or calibrated is better,
minimum detectable effect: the smallest meaningful difference the study can reliably see,
subgroup plan: which learner slices are pre-registered and large enough to inspect,
attrition rule: how missing outcome data will be monitored and handled,
guardrails: consent, risk, equity, data minimization, accessibility, and support burden,
rollout gate: the exact conditions that must pass before expansion.

The habit is simple and easy to skip: design the learning claim before collecting the launch evidence.

Math

Translate the story into symbols, assumptions, and a derivation you can inspect.

Section prompt

Equation 1 $\Delta_{t^\star} = \mathbb E[Y_{i,t^\star} \mid Z_i = 1] - \mathbb E[Y_{i,t^\star} \mid Z_i = 0].$ Equation 2 $t^\star \in \{\text{delayed retention at 7-14 days}, \text{unit follow-up}, \text{course checkpoint}\}.$

Let $Z_i \in \{0,1\}$ be randomized assignment to control or treatment. Let $Y_{i,t}$ be the outcome for learner $i$ at timepoint $t$ .

The primary estimand should name both the outcome and the time:

\Delta_{t^\star} = \mathbb E[Y_{i,t^\star} \mid Z_i = 1] - \mathbb E[Y_{i,t^\star} \mid Z_i = 0].

For a learning system, $t^\star$ is often not the immediate post-test. A delayed-retention timepoint can matter more:

t^\star \in \{\text{delayed retention at 7-14 days}, \text{unit follow-up}, \text{course checkpoint}\}.

With equal treatment/control allocation, a compact standard-error approximation is

\operatorname{SE}(\widehat \Delta) = \sigma \sqrt{\frac{1}{n_T}+\frac{1}{n_C}},

where $n_T$ and $n_C$ are post-attrition sample counts and $\sigma$ is the outcome standard deviation.

A rough two-sided 80 percent power minimum detectable effect is

\operatorname{MDE}_{80} \approx (1.96 + 0.84) \operatorname{SE}(\widehat \Delta).

Subgroups should be declared before the result is known. For a subgroup $s$ , inspect

\Delta_s = \mathbb E[Y_{i,t^\star} \mid Z_i = 1, S_i=s] - \mathbb E[Y_{i,t^\star} \mid Z_i = 0, S_i=s],

but do not let an unplanned slice replace the primary outcome.

A compact rollout gate is:

\operatorname{ready} = \mathbf 1[ \text{primary outcome defined} \land \operatorname{power} \ge 0.8 \land \text{delayed retention measured} \land \text{subgroups pre-registered} \land \text{attrition plan passes} \land \text{guardrails pass} ].

The gate is intentionally conjunctive. A large average lift cannot repair a missing consent path, an underpowered design, or a primary outcome that was chosen after looking at results.

Code

Keep the implementation aligned with the notation so the algorithm is legible.

Section prompt

Runnable code 1from math import erf, sqrt def normal_cdf(x): return 0.5 * (1.0 + erf(x / sqrt(2.0))) def des...python

from math import erf, sqrt


def normal_cdf(x):
    return 0.5 * (1.0 + erf(x / sqrt(2.0)))


def design_power(total_n, attrition, outcome_sd, target_mde):
    effective_n = total_n * (1.0 - attrition)
    n_t = effective_n / 2.0
    n_c = effective_n / 2.0
    se = outcome_sd * sqrt((1.0 / n_t) + (1.0 / n_c))
    mde_80 = (1.96 + 0.84) * se
    power = normal_cdf((target_mde / se) - 1.96)
    return round(power, 3), round(mde_80, 3)


plans = {
    "ready": {
        "total_n": 2400,
        "attrition": 0.10,
        "outcome_sd": 0.42,
        "target_mde": 0.05,
        "delayed_retention": True,
        "subgroups_preregistered": True,
        "guardrails": "pass",
    },
    "immediate_only": {
        "total_n": 2400,
        "attrition": 0.08,
        "outcome_sd": 0.42,
        "target_mde": 0.05,
        "delayed_retention": False,
        "subgroups_preregistered": True,
        "guardrails": "pass",
    },
}

for name, plan in plans.items():
    power, mde_80 = design_power(
        plan["total_n"], plan["attrition"], plan["outcome_sd"], plan["target_mde"]
    )
    ready = (
        power >= 0.8
        and plan["delayed_retention"]
        and plan["subgroups_preregistered"]
        and plan["guardrails"] == "pass"
    )
    print(name, power, mde_80, ready)

The witness does not claim the experiment is solved. It shows the shape of a design review: compute what the study can detect, then refuse to treat power as the only gate.

Interactive Demo

Use direct manipulation to connect the explanation to a moving system.

Section prompt

Use the Outcome Design Lab to predict the right experiment-design move before the proof unlocks:

Outcome: decide whether the primary outcome and timepoint make a durable learning claim.
Power: decide whether the sample can detect the minimum meaningful effect.
Retention: decide whether immediate gains need a delayed check.
Subgroups: decide whether slices are pre-registered and large enough.
Guardrails: decide whether consent, data, risk, and equity gates allow the study to proceed.

Before reveal, exact sample counts, power, MDE, attrition, subgroup cell sizes, and rollout decisions stay locked. After reveal, inspect the design ledger and decide whether the experiment should run, be resized, gain a delayed-retention measure, pre-register subgroup slices, repair attrition handling, or stop on a learner-risk gate.

Live Concept Demo

Explore Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails

The stage is code-native and interactive. Use it to test the explanation against the mechanism.

difficulty 4/5graduatecode-aligned

Demo Prediction Checkpoint

Manipulate one control and predict the visible change.

Commit to what Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails should make visible before reading the result.

After The First Pass

Turn the concept into an inspected object.

Once the invariant is visible in the intuition, math, code, and demo, use these panels to inspect the mechanism visually, check source support, practice the idea, and attach a grounded research question.

Mechanism Storyboard

See the idea move before the page explains it

Design adaptive-learning experiments by choosing a primary outcome, delayed retention check, power/MDE target, subgroup plan, attrition rule, ethics guardrails, and rollout gate.

Prediction open01 / Intuition

Prediction lens

Start with the picture, metaphor, or geometric mechanism.

Commit first

Before reading further, choose the kind of change Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails should make visible.

Visual Inquiry

Make the image answer a mathematical question

Design adaptive-learning experiments by choosing a primary outcome, delayed retention check, power/MDE target, subgroup plan, attrition rule, ethics guardrails, and rollout gate.

4/4 stages readyLive demo connected

Visual cueWhich visible object should carry the first intuition?

Inspection depth2/4

Prediction

Which visible object should carry the first intuition?

Commit first

Pick the cue that should make Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails easier to reason about before the page gives the answer.

Source Grounding

Canonical references for the mechanism on this page.

reference · 2022What Works Clearinghouse Procedures and Standards Handbook, Version 5.0Institute of Education Sciences

Primary education-evidence standard for study review, outcome standards, attrition, baseline equivalence, and reporting.

Open source

reference · 2025What Works Clearinghouse Study Review Protocol, Version 5.1Institute of Education Sciences

Support for eligible outcome domains, independent measures, consistent measurement, and practically meaningful benchmarks.

Open source

paper · 2009Controlled Experiments on the Web: Survey and Practical GuideKohavi, Longbotham, Sommerfield, and Henne

Support for randomized online experiments, clear evaluation criteria, statistical power, sample size, and trustworthy launch decisions.

Open source

paper · 2024PUMP: Estimating Power, Minimum Detectable Effect Size, and Sample Size When Adjusting for Multiple Outcomes in Multi-Level ExperimentsHunter, Miratrix, and Porter

Support for power, MDES, sample-size, multi-level RCTs, multiple outcomes, and sensitivity to assumptions in education experiments.

Open source

paper · 2006Test-Enhanced Learning: Taking Memory Tests Improves Long-Term RetentionRoediger and Karpicke

Support for delayed retention checks and for treating assessment timing as part of the learning claim.

Open source

reference · 1979The Belmont ReportNational Commission for the Protection of Human Subjects of Biomedical and Behavioral Research

Support for respect for persons, beneficence, justice, informed consent, risk-benefit assessment, and fair subject selection.

Open source

Claim Review

Design adaptive-learning experiments by choosing a primary outcome, delayed retention check, power/MDE target, subgroup plan, attrition rule, ethics guardrails, and rollout gate.

Status2 substantive reviews recorded

Claims without a substantive review badge still need exact source-support review.

Sources6 references

wwc-2022-procedures-standards, wwc-2025-study-review-protocol, kohavi-2009-controlled-experiments, hunter-2024-pump-power, roediger-2006-test-enhanced-learning, belmont-report

Local checks4 local checks

Use equations, runnable code, and demos to check whether the source support is operational.

Substantively reviewedA learner-outcome experiment should name one primary outcome, its measurement timepoint, direction, instrument, missing-data rule, and delayed-retention check before interpreting adaptive-learning lift.Claim metadata: source checked

The references support pre-specifying valid outcome domains and measures, distinguishing immediate checks from delayed retention, and treating missing outcomes as part of study credibility.

Sources: What Works Clearinghouse Procedures and Standards Handbook, Version 5.0, What Works Clearinghouse Study Review Protocol, Version 5.1, Test-Enhanced Learning: Taking Memory Tests Improves Long-Term RetentionThe page teaches a compact design contract for adaptive-learning experiments; it does not replace a psychometric validation, IRB review, or full statistical analysis plan.A bounded review summary is present; still check caveats and exact reference scope.

Checked WWC Handbook and Study Review Protocol for eligible outcome domains, face validity, independent measures, consistent measurement, baseline equivalence, and attrition framing; checked Roediger/Karpicke for the distinction between immediate and delayed retention. Oracle/GPT Pro review remains pending because the correct Chrome/Oracle lane is not attachable from this workstation.

Reviewer: codex-local-primary-reference-audit; reviewed 2026-07-05

Equation 1 Equation 2 Runnable code 1 Interactive demo

Substantively reviewedA rollout decision should inspect power/MDE, subgroup and attrition plans, and learner-risk guardrails separately from the average treatment effect.Claim metadata: source checked

The references support power and sample-size planning, multiple-outcome sensitivity, randomized experiment criteria, attrition and subgroup caution, and ethical risk-benefit gates.

Sources: Controlled Experiments on the Web: Survey and Practical Guide, PUMP: Estimating Power, Minimum Detectable Effect Size, and Sample Size When Adjusting for Multiple Outcomes in Multi-Level Experiments, The Belmont Report, What Works Clearinghouse Procedures and Standards Handbook, Version 5.0The local power arithmetic is a teaching approximation with fixed variance and equal allocation; production designs may need clustering, covariates, sequential monitoring, multiplicity adjustment, and formal review.A bounded review summary is present; still check caveats and exact reference scope.

Checked controlled-experiment practice for randomized online experiments, evaluation criteria, power, and sample size; checked PUMP for MDES and power under multiple outcomes in education RCTs; checked Belmont for consent, risk-benefit, and fair participant selection.

Reviewer: codex-local-primary-reference-audit; reviewed 2026-07-05

Equation 2 Runnable code 1 Interactive demo

Source support candidates

reference 2022What Works Clearinghouse Procedures and Standards Handbook, Version 5.0

Primary education-evidence standard for study review, outcome standards, attrition, baseline equivalence, and reporting.

reference 2025What Works Clearinghouse Study Review Protocol, Version 5.1

Support for eligible outcome domains, independent measures, consistent measurement, and practically meaningful benchmarks.

paper 2009Controlled Experiments on the Web: Survey and Practical Guide

Support for randomized online experiments, clear evaluation criteria, statistical power, sample size, and trustworthy launch decisions.

paper 2024PUMP: Estimating Power, Minimum Detectable Effect Size, and Sample Size When Adjusting for Multiple Outcomes in Multi-Level Experiments

Support for power, MDES, sample-size, multi-level RCTs, multiple outcomes, and sensitivity to assumptions in education experiments.

Mechanism checks

Equation 1

\Delta_{t^\star} = \mathbb E[Y_{i,t^\star} \mid Z_i = 1] - \mathbb E[Y_{i,t^\star} \mid Z_i = 0].

Equation 2

t^\star \in \{\text{delayed retention at 7-14 days}, \text{unit follow-up}, \text{course checkpoint}\}.

Runnable code 1from math import erf, sqrt def normal_cdf(x): return 0.5 * (1.0 + erf(x / sqrt(2.0))) def des...Demo stateLive mechanism probe

Practice Loop

Try the idea before it explains itself

Design adaptive-learning experiments by choosing a primary outcome, delayed retention check, power/MDE target, subgroup plan, attrition rule, ethics guardrails, and rollout gate.

Readiness0/3 checks ready

Predict

Before touching the demo, predict one visible change that should happen in Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails.

Hint 1

Reveal when your model needs a nudge.

Hint 2

Reveal when your model needs a nudge.

Hint 3

Reveal when your model needs a nudge.

Your answer canvas

Local checks

Claim

A concrete answer is on the canvas.

Mechanism

The answer names why the claim should hold.

Bridge

It touches the page context or a neighboring idea.

Misconception check

Grounded research drawerClose

ConceptLearner Outcome Experiment Design: Outcomes, Power, Retention, GuardrailsMachine Learning

Runnable code comparisonLearner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails runnable code 1effective_n = total_n * (1.0 - attrition)Prediction before revealLearner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails interactive demoManipulate one control and predict the visible change.

Grounded room questionWhat is the smallest example that makes Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails click without losing the math?Local snapshot ready

Research Room

Attach the question to a claim, equation, code, or demo

Pick the concept, equation, source, runnable code, claim, misconception, or demo state before asking for help. The handoff keeps that page item in context.

Next local actionNo local draft saved yet

Open the draft below to save one note and next action in this browser.

conceptMachine Learning

Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails

Attached question

What is the smallest example that makes Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails click without losing the math?

Local action draftNo local draft saved yetExpand only when ready to capture one local next action

Local action draft

This draft stays in this browser, attached to the selected learning item.

Draft noteNext action

No local draft saved.

Evidence to inspect

References to inspect: attached references on this page.
Definition, prerequisite, and contrast concept links
The equation or runnable code that makes the concept operational
One demo state that shows the invariant instead of a slogan

What would resolve this

The learner can state the mechanism in their own words
The learner can name the prerequisite that would repair confusion
The learner can predict how the mechanism changes under one perturbation

Grounded AI handoff

I am working in Continuous Function's research reading room. Object: concept - Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails Selected item key: recorded for copy. Context: Machine Learning Page anchor: recorded for copy. Open question: What is the smallest example that makes Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails click without losing the math? Evidence to inspect: - References to inspect: attached references on this page. - Definition, prerequisite, and contrast concept links - The equation or runnable code that makes the concept operational - One demo state that shows the invariant instead of a slogan What would resolve this: - The learner can state the mechanism in their own words - The learner can name the prerequisite that would repair confusion - The learner can predict how the mechanism changes under one perturbation Answer as a careful research tutor: stay source-grounded, separate verified evidence from assumptions, name the relevant math objects, and end with one next action.

View it in context

concept/concept-notebook/machine-learning/learner-outcome-experiment-design
concept:machine-learning/learner-outcome-experiment-design

Learning Map

Before / Now / Try / Next

BeforeLearning System Policy Evaluation: Logs, Estimators, Guardrails, Promotion (review)

NowIntuition -> Math -> Code -> Demo

TryManipulate one control and predict the visible change.

NextAdaptive Learning Impact Audits: Retention, Drift, Guardrails, Rollback (review)

Intuitionready
Mathready
Codeready
Interactive Demoready

AI Companion

Ask beside the selected object

Design adaptive-learning experiments by choosing a primary outcome, delayed retention check, power/MDE target, subgroup plan, attrition rule, ethics guardrails, and rollout gate.

Your question

GoalComfortStyleStuck on

Context prompt

You are my AI learning companion for Continuous Function. Current context: Machine Learning concept. Learning surface: Learner Outcome Experiment Design: Outcomes, Power, Retention, Guardrails. What this page says: Design adaptive-learning experiments by choosing a primary outcome, delayed retention check, power/MDE target, subgroup plan, attrition rule, ethics guardrails, and rollout gate. Current section: Intuition, math, code, and interactive demo. Suggested next step: Manipulate one control and predict the visible change.. Learner goal: Understand the idea. Learner comfort level: New to this. Preferred explanation style: Visual first. Task: Explain the central idea in plain language, then restate it with the exact math objects from the page. Answer in a way that helps me learn: ask one clarifying question only if needed, use intuition before notation, and end with one thing I should try on the page.

Domain

Machine Learning

machine-learningadaptive-learningexperiment-designlearning-outcomespower-analysisguardrails

Prerequisites

Learning System Policy Evaluation: Logs, Estimators, Guardrails, Promotion (review)Causal Inference Basics: Correlation versus Intervention (review)Evaluation Pipelines

Leads To

Adaptive Learning Impact Audits: Retention, Drift, Guardrails, Rollback (review)

Learning System Policy Evaluation: Logs, Estimators, Guardrails, Promotion (review)Adaptive Curriculum Optimization: Policy, Utility, Exploration, Audit (review)Causal Inference Basics: Correlation versus Intervention (review)Train/Dev/Test Splits, Cross-Validation, and Leakage

Within this domain