Mk 134

---
name: mk-134
description: "Sales Page Single-Variable Test Design. Use when you need one key difference compared per test; apply this method to the specified inputs and verify the deliverable"
---

# MK-134｜Sales Page Single-Variable Test Design

For a regular chat, upload `packs/mk-134.md`; it includes the shared rules, this method, and required references

## Purpose

This Skill delivers: **One key difference compared per test**

A page content check does not replace user behavior data; without performance data, do not assert the cause of declining conversions

## Required Inputs


Available raw records: current and proposed page blocks, customer/behavior evidence, assignment and exposure records, outcomes, and the existing evaluation design

## Method Steps

1. Build one distinguishable hypothesis from page content, customer questions, and behavior, naming the sole block to change and the primary outcome. Use existing pain-point evidence without requiring another survey
2. Keep product, price, eligibility, and traffic conditions comparable. Prefer simultaneous random allocation with a stable version per person. Nonrandom or before/after comparisons yield observed differences pending source/time confounding checks
3. Record the primary metric, economic floor, observation horizon, and stopping design before starting, using the brand's established evaluation design. If sample/uncertainty treatment or the smallest worthwhile difference is missing, propose a design and request that decision rather than inventing significance thresholds
4. Compare mature outcomes at the planned endpoint, not when a temporary lead appears. Stop and repair a version with a broken necessary interaction as a usability action, not a benefit winner
5. Recalculate or withdraw causal claims after delayed conversions, refunds, allocation errors, or concurrent price changes. If evidence does not support the intended decision, retain the original or gather more within the existing test allowance rather than forcing a winner

## Calculation and Interpretation

Conversion uses all eligible assigned visitors, including nonbuyers. A value objective uses net contribution per eligible visitor. Report raw counts, outcomes, horizon, and uncertainty treatment, with adopt/retain/repair and reversal conditions

Check example (hypothetical): A has ten orders from 100 visitors and B has 15 from 100; describe B's observed lead. Discovering that B also lowered price and used returning-customer traffic withdraws the headline-caused-lift claim and calls for a comparable single-change design

Decision rules authored by ThinkWithBlack, 2026-10-14; original course sources remain below

## Required Output Fields

| Order | Field | How to complete |
|---:|---|---|
| 1 | Hypothesis and original evidence | Use the supplied source, calculation, or explicit unknown |
| 2 | Control, variant, and single change | Use the supplied source, calculation, or explicit unknown |
| 3 | Held conditions and allocation | Use the supplied source, calculation, or explicit unknown |
| 4 | Primary metric, numerator, and denominator | Use the supplied source, calculation, or explicit unknown |
| 5 | Cost floor and observation window | Use the supplied source, calculation, or explicit unknown |
| 6 | Agreed end rule and uncertainty treatment | Use the supplied source, calculation, or explicit unknown |
| 7 | Mature results and confound checks | Use the supplied source, calculation, or explicit unknown |
| 8 | Adopt, retain, or repair decision | Use the supplied source, calculation, or explicit unknown |
| 9 | Missing evidence and reversal conditions | Use the supplied source, calculation, or explicit unknown |

## Method Sources

Original file: 行銷策略實戰學程Day4-佈局獲利 × 節奏變現 × 轉單收割

Source location: Nine-cell framework

Source basis: Derived from course material

Original URL: https://docs.google.com/presentation/d/1dq3xqYVUnWvUCpzZKI91yDDOzvkyG0uHy3EOsOjtItA/edit

English translation of source excerpt retained in the master table:

> Nine-cell framework