The Faithful
Measure

FUNDAMENTALS 01 Understanding Users Through Evidence EPIGRAPH “The beginning of wisdom is this: Get wisdom, and whatever you get, get insight” (Proverbs 4:7). Concrete Situation and Opening Decision ILLUSTRATIVE CASE EPISODE Three weeks into the pilot, Elena asks Priya, the analyst, a plain question: is the service helping? Priya’s dashboard shows visits, new accounts and posted requests, and every line is rising. Leah, a volunteer, mentions that coordinators still telephone her to confirm deliveries. Before Priya reports anything to the sponsor, she must decide which question her analysis should answer first and what evidence could answer it. LEARNING OBJECTIVES By the end of this chapter, you should be able to define UX analytics and distinguish it from neighboring practices, classify evidence as behavioral or attitudinal and as quantitative or qualitative, explain what each kind can and cannot establish, describe the path from a decision to evidence and back, recognize the main categories of analytics instrument, and write an analysis brief whose assumptions remain visible. Key terms UX analytics; event; metric; dimension; behavioral evidence; attitudinal evidence; quantitative; qualitative; triangulation; analysis brief Core Explanation and Visual Anchors 1.1 What UX analytics is UX analytics is the disciplined use of evidence about people’s behavior to understand their experience of a product and to support a decision about it. The definition has three parts. The evidence is mostly behavioral: records of what people did. The subject is experience: whether people could accomplish what they came to do, and at what cost to them. The purpose is a decision: something the team will do differently depending on what it learns. Remove any part and the practice changes character. Evidence without a decision produces reports that nobody uses. A decision without evidence is an opinion with a budget. Evidence about behavior that never asks about experience can describe a product in detail while missing whether it serves anyone. The experience extends beyond the screen. In the illustrative community service, a volunteer accepts a delivery in the application, but the experience includes knowing that the coordinator has confirmed it, arriving at the right address, and learning whether the need was met. An analysis confined to screens will describe part of that experience and may mistake it for the whole. Several neighboring practices use similar tools. Organizations draw the boundaries differently, so ask what question a team is answering before relying on its label. TABLE 1.1 Compare neighboring practices Practice | Typical question | Typical unit Web analytics | How much traffic arrived, from where, and which pages did it view? | Sessions and pageviews Product analytics | Which features do people adopt, and do they return? | Users and events Business intelligence | How is the organization performing against its financial and operating plan? | Accounts, orders, revenue UX research | What do people need, and why do they act as they do? | Participants and observations UX analytics | Where does the experience help or hinder people, how often, and what should change? | People, tasks and journeys UX analytics borrows from each. It uses the instruments of web and product analytics, shares the purpose of UX research, and must eventually speak to the organization’s plan. Its distinct contribution is to connect measured behavior with the quality of a person’s experience. 1.2 A record is a trace of a person Analytics instruments collect records. An event is a record that something happened: a page was displayed, a button was pressed, the server confirmed a request. A metric is a number calculated from records, such as the count of confirmed requests or the proportion of visits that included one. A dimension is an attribute used to group records, such as device type, community or week. Chapter 4 develops these terms; for now, notice what kind of thing a record is. A record is a trace left by a person doing something for a reason. It preserves the action and a timestamp. It does not preserve the reason, the circumstances, or anything that happened away from the instrumented surface. When a coordinator telephones a volunteer to confirm a delivery, no event fires. The dashboard describes the instrumented part of the service, which may or may not be the important part. Hay puts the division of labor simply: quantitative analytics will generally tell you what people are doing, and qualitative methods help you learn why. (Hay 2017) The distinction guards against two errors. One treats a number as if it explained itself. The other dismisses numbers because they cannot explain themselves. A count of abandoned forms does not say why people left. It does say how many did, which an interview with five people cannot. Records also omit people. Those who never found the service, who declined tracking, or who gave up before the first instrumented step leave no trace at all. Every metric has a population it describes and a population it cannot see. State both. 1.3 Four kinds of evidence Evidence about experience varies along two dimensions. It may concern what people do or what people say: behavioral or attitudinal. It may answer how many and how much, or why and how: quantitative or qualitative. Rohrer’s widely used map of research methods arranges them on these axes. (Rohrer n.d.) Crossing the two dimensions gives four kinds of evidence. FIGURE 1.1 Two distinctions produce four kinds of evidence. Each supports a different sort of claim. Adapted from the dimensions in Rohrer’s landscape of user research methods. TABLE 1.2 What each kind of evidence can and cannot support Kind | Can support | Cannot establish alone Behavioral, quantitative | How many people reached a step, how often, and where they stopped | Why they stopped, or whether those who finished were well served Behavioral, qualitative | How a difficulty arises and what people try next | How common the difficulty is among all users Attitudinal, quantitative | How widely a stated view is held among those who answered | What people actually do, or the views of those who did not answer Attitudinal, qualitative | Reasons, expectations and the words people use | Prevalence, or behavior the person does not notice or recall What people say and what they do can differ without anyone being dishonest. People forget, summarize, and describe what they intend. Behavior has its own distortions: a person may complete a task through a workaround that the records present as success. Neither kind is the truth against which the other is checked. Each is a partial witness. UX analytics lives mainly in the behavioral and quantitative quadrant, which explains both its authority and its characteristic mistakes. Numbers drawn from thousands of sessions feel conclusive. They are conclusive about what they measure. 1.4 From a decision to evidence and back Useful analysis begins with a decision someone must make and works backward. What would we need to know to decide well? How will the terms in that question be defined? What evidence exists, and has it been checked? Only then does analysis begin, and it ends with a recommendation that states its limits. A later review asks whether the decision helped and what should be asked next. FIGURE 1.2 Analysis begins with a decision and returns to it. The dashed line marks the review that turns one analysis into the next question. Working in the other direction, from available data toward something interesting, is tempting because the data is already there. It tends to produce findings that are true, mildly surprising, and connected to nothing anyone can act on. Analytical questions come in four common types. They build on one another, and the later types inherit every weakness in the earlier ones. TABLE 1.3 Four types of analytical question Type | Question | Example from the community service | Developed in Descriptive | What happened? | How many posted requests were fulfilled within the agreed time? | Chapters 4–6 Diagnostic | Why did it happen? | At which step do unfulfilled requests stop, and what do people do there? | Chapters 7–10 Predictive | What is likely to happen? | Which volunteers are unlikely to return next month? | Chapter 11 Prescriptive | What should we do? | Which change to the confirmation step should be released? | Chapters 9 and 13 Most valuable work in a young service is descriptive and diagnostic. A prediction built on undefined measures inherits their problems and conceals them behind a more impressive method. MODEL IN DEPTH · Rewriting the question Weak question: “Is the service working?” It names no unit, no condition, no period and no comparison. Any rising line appears to answer it. Stronger question: “Among requests posted during the pilot, what proportion were fulfilled within the agreed time, and at which step did the unfulfilled requests stop?” The unit is a request. The condition is fulfillment within a stated time. The period is the pilot. The second clause directs attention to a place in the journey. The stronger question still depends on definitions Priya does not yet have. Who records that a request was fulfilled, and when? If coordinators confirm by telephone, does the application ever learn of it? Writing the question well exposes what must be settled before any number is reported. 1.5 Compare kinds of evidence before concluding Triangulation means examining a question with more than one kind of evidence and attending to where the kinds agree and disagree. Agreement raises confidence. Disagreement is often more useful, because it shows that one of the readings was wrong. MODEL IN DEPTH · One number, two readings In an illustrative week, 100 volunteer sessions reach the screen that follows acceptance of a request, and 70 of those sessions end there. The exit rate for that screen is 70 ÷ 100 = 70 percent. Reading A: The volunteer has accepted the request. The task is complete, and leaving is what success looks like. A high exit rate on a final screen is unremarkable. (Hay 2017) Reading B: The screen does not show whether the coordinator has confirmed the assignment. Volunteers leave uncertain, and coordinators telephone them to close the gap. The exit marks a point where the service hands its work back to people. The number cannot choose between these readings. Priya watches six coordinators work through a morning’s requests and reviews Elena’s call notes. Four of the six coordinators telephone the volunteer after an acceptance. Reading B has support, from a small observation that shows how the problem arises without establishing how common it is across the pilot. The exit rate was accurate. Its first interpretation was not. Priya’s next step is a measure that can count what she observed: the proportion of accepted requests followed by a coordinator’s confirmation inside the application. Notice the order. The quantitative record located a place worth examining. Qualitative observation supplied a mechanism. A new quantitative measure will establish prevalence. Analysts who are comfortable moving among kinds of evidence make fewer confident mistakes than those loyal to one. 1.6 Instruments, not answers Analytics tools are instruments. Each observes from a vantage point and has blind spots. It is more durable to learn the categories of instrument and the questions each can address than to learn a particular product’s menus. TABLE 1.4 Categories of analytics instrument Category | What it observes | Examples at the time of writing Web and product analytics | Events, sessions and users across a site or application | Google Analytics 4, Adobe Analytics, Amplitude, Mixpanel Behavior visualization | Clicks, scrolling and recorded sessions on individual pages | Microsoft Clarity, Contentsquare Experimentation | Outcomes under randomly assigned alternatives | Optimizely, VWO Feedback and survey | What people say, in context or afterward | Qualtrics, in-product survey tools Reporting | Combined measures presented for a decision | Looker Studio, Power BI, Tableau The third column will age fastest. Google retired Universal Analytics in 2023 and replaced it with Google Analytics 4, which records behavior through a different data model, so reports and even metric definitions changed. Hotjar, long a common heatmap tool, was merged into Contentsquare in 2025. Appendix F carries versioned walkthroughs so that the chapters can teach reasoning that outlasts a product name. Every instrument also loses data in ordinary operation. Visitors decline consent, browsers block scripts, large reports are sampled, and small groups are withheld to protect privacy. These are not defects to be ashamed of. They are properties to be stated whenever a figure is reported. FROM THE FIELD · analytics.usa.gov The United States government’s Digital Analytics Program, operated by the General Services Administration, publishes web traffic data for federal websites at analytics.usa.gov. The program reports covering more than 500 second-level domains and roughly 7,000 hostnames, and it offers the data for download. The site is useful for practice because real figures are available without an account. It is instructive for a second reason. Alongside the numbers, the program states that it does not track individuals, that visitor IP addresses are anonymized, and that the data is subject to sampling and should be read for high-level trends. An organization publishing its instrument’s limits next to its figures is modelling a habit this book will ask of you in every report. (U.S. General Services Administration n.d.) 1.7 An analysis brief makes the question inspectable An analysis brief is a one-page statement of what an analysis is for and what it will rely on. It is written before the analysis and revised as evidence arrives. Its value lies in exposing disagreement early, when changing course is cheap. Use the following fields: the decision and its owner; the analysis question; the people and the unit of analysis; working definitions; evidence available, by kind; evidence missing; known limits of the instruments; people affected by collecting the evidence; the next step; and the condition under which the question should be reconsidered. Early in a project many fields will hold hypotheses. Label them as such. MODEL IN DEPTH · Priya’s first brief Decision and owner: Whether to extend the pilot to two further organizations next quarter. Jonah decides, advised by Maya and Elena. Analysis question: Among requests posted during the pilot, what proportion were fulfilled within the agreed time, and at which step did unfulfilled requests stop? People and unit: Members who post, volunteers who accept, coordinators who confirm. The unit is a request, followed from posting to fulfillment. Evidence available: Application events for posting and acceptance (behavioral, quantitative). Elena’s fulfillment record, kept by hand (behavioral, quantitative, outside the application). One morning’s observation of coordinators (behavioral, qualitative). Evidence missing: Any record of telephone confirmations. Members’ own account of whether the need was met (attitudinal). Both are hypothesized to matter. Instrument limits: Acceptance is recorded when the button is pressed, which may differ from a confirmed assignment. Volunteers who use the service on two devices may be counted twice. People affected by collection: Requests can describe sensitive needs. The analysis will use request status and timing, and will not read request text. Next step: Reconcile two weeks of application events against Elena’s record before reporting any rate. A brief of this kind gives colleagues something to correct. Daniel can say how acceptance is actually logged. Elena can say whether her record is complete. Maya can challenge whether fulfillment within the agreed time is the outcome that matters. The analysis improves before it has begun. Faith and Practice Reflection FAITH & PRACTICE · Faith and Practice Reflection Elena’s question deserves a truthful answer because volunteers’ evenings and members’ needs stand behind it. Attention comes before technique: learn what happened to people before reporting what happened to numbers. A rising line can reassure a sponsor while a coordinator is still making telephone calls. Insight is worth more than a favorable chart. 1.8 Understanding begins with the person behind the record Proverbs 4 is a father’s instruction to his son to pursue wisdom above other gains. In Proverbs that wisdom begins with the fear of the LORD, and it concerns the whole conduct of a life. The passage is not an endorsement of data collection, and this chapter does not use it as one. What it does commend is that understanding should be sought deliberately and at cost, and that acquiring many things is different from gaining insight. An analyst surrounded by data knows the difference. (Proverbs 1:7; 4:1–9) The Christian account developed in Chapter 2 grounds the worth of persons in their creation in God’s image. A record can show that someone abandoned a form. It cannot show the hour they had available, the shared telephone they were using, or the need that brought them. The person exceeds every record the system keeps of them. Two practical consequences follow. Count in order to serve: ask who benefits from an analysis and whether the people observed are among them. And observe limits: that something can be collected is not yet a reason to collect it. Colleagues who do not share the Christian premise can hold both standards, and professional codes state similar duties toward the people a system affects. (Association for Computing Machinery 2018) The shared requirement is an honest account of what the evidence shows and whom it concerns. Actionable Application MODEL IN DEPTH · Worked scenario Priya rewrites “report that usage is growing” as “learn whether posted requests are fulfilled and where unfulfilled ones stop.” She lists the evidence she has by kind and notes that no event exists for telephone confirmations. She tells Elena that she cannot yet answer the question, explains why, and names the date by which the reconciliation with Elena’s record will be finished. Elena would have preferred a number. She receives something more useful: a statement of what is known, what is not, and when that will change. Project worksheet Complete the fields for a site or product you can observe, using the worked scenario as a guide. Keep each response specific enough to inform a decision. TABLE 1.5 Project worksheet Worksheet field | Complete for your product | Your response Decision and owner | What will be decided, and by whom? | ________________ Analysis question | What would the owner need to know? Name the unit, condition and period. | ________________ Evidence available | What do you have, and which of the four kinds is it? | ________________ Evidence missing | Which kind is absent, and what could it change? | ________________ Instrument limits | Whom or what can your instruments not see? | ________________ Next step | What will you check before reporting a figure? | ________________ Chapter summary UX analytics uses behavioral evidence to understand experience in support of a decision. Records are traces of people and omit reasons, circumstances and everyone the instrument cannot see. Four kinds of evidence support different claims, and comparing them catches confident misreadings. Instruments change; the questions they answer endure. An analysis brief makes the question and its assumptions available for correction. Practice and reflection APPLICATION · Guided practice A director reports: “Average time on our help page rose from two minutes to four. People are more engaged with our help content.” Identify the kind of evidence, give two competing interpretations, and name one additional kind of evidence that could distinguish them. Guidance: Time on page is behavioral and quantitative. People may be reading more carefully because the content improved. They may also be struggling to find an answer, rereading, or leaving the tab open while they telephone support. Observation of several people using the page, or a review of whether support contacts about the same topics rose or fell, would help distinguish the readings. “Engagement” is an interpretation placed on the number. It is not what the instrument measured. APPLICATION · Independent application Choose a live website or application that you can use as a visitor and return to throughout this book. Write a one-page analysis brief for one decision its owner plausibly faces. Label every unsupported statement as a hypothesis. Identify one kind of evidence you cannot obtain from outside the organization and explain how its absence limits what you may conclude. Review questions Explain why a record of behavior does not explain the behavior. Give an example in which what people say and what people do would differ without dishonesty. Distinguish a metric from a dimension. Explain why a prediction inherits the weaknesses of the descriptive measures beneath it. Describe a population that an analytics instrument on a website cannot see. YOUR DECISION ARTIFACT Retain your analysis brief. In Chapter 2 you will add a responsibility statement naming the people your analysis affects. In Chapter 3 you will connect the question to goals and measures, and in Chapter 4 to a measurement plan. The brief should change as your evidence improves. By Chapter 8 it becomes the opening page of an audit. Sources and further study The reflection draws on Proverbs 1:7 and 4:1–9. The chapter paraphrases and interprets these passages; it does not present its applications to analytics as part of the biblical text. Albert, Bill, and Tom Tullis. 2022. Measuring the User Experience: Collecting, Analyzing, and Presenting UX Metrics. 3rd ed. Morgan Kaufmann. Association for Computing Machinery. 2018. Code of Ethics and Professional Conduct. Its duties toward people affected by a system offer a nonsectarian comparison for the responsibilities discussed here. Hay, Luke. 2017. Researching UX: Analytics. SitePoint. Chapter 1 sets out the case for combining quantitative and qualitative evidence. Rohrer, Christian. n.d. “When to Use Which User-Experience Research Methods.” Nielsen Norman Group. Talluri, Sean. 2026. The Product Steward: Building for Human Flourishing in the Age of AI. Rev. ed. Chapters 1 and 11 introduce the community coordination service, outcomes, and measurement from the product manager’s position. U.S. General Services Administration, Digital Analytics Program. n.d. analytics.usa.gov: About this site.

FUNDAMENTALS 02 The Gospel and Faithful Measurement EPIGRAPH “A false balance is an abomination to the LORD, but a just weight is his delight” (Proverbs 11:1). Concrete Situation and Opening Decision ILLUSTRATIVE CASE EPISODE The sponsor’s quarterly report is due on Friday, and Priya’s dashboard has never looked better. Requests posted have tripled in three weeks. So have sessions. Minutes in the application have nearly quadrupled. Elena’s notebook records the same three weeks differently: the share of requests fulfilled in time has fallen each week, and her coordinators are on the telephone most evenings. Every figure on both pages is accurate. Priya must decide what an honest report says, and she finds herself asking a prior question: why does it matter so much that it be honest? LEARNING OBJECTIVES By the end of this chapter, you should be able to explain what Scripture’s teaching on honest weights concerns and why it bears on measurement; explain the Gospel through creation, fall, redemption and restoration; distinguish salvation from ethical or professional achievement; identify five ways an accurate number can mislead; apply a four-step rule for moving from a biblical passage to a professional decision; compare kinds of evidence in an agreement matrix; and write a responsibility statement for an analysis. Key terms Just weight; Gospel; image of God; self-deception; grace; repentance; common grace; vocation; denominator; definition drift; evidence agreement matrix; responsibility statement Core Explanation and Visual Anchors 2.1 What the honest-weights passages are about Picture a grain seller in an Israelite market town. She has a hand balance and a small bag of stones, each stone a standard weight. A farmer brings barley to sell. She sets her heavy stones on the pan, and his barley weighs less than it should. An hour later a widow comes to buy, the light stones come out of the same bag, and her silver buys less than it should. Neither customer can see the difference. Both go home poorer. Scripture returns to this small fraud with surprising persistence. The law forbids it twice, once among the commands that define love of neighbor. Proverbs condemns it four times. The prophets name it among the sins that bring judgment on a nation, and they are specific about who pays: the needy, who cannot afford to be cheated and have no means of checking. (Leviticus 19:35–36; Deuteronomy 25:13–16; Proverbs 11:1; 16:11; 20:10, 23; Amos 8:4–6; Micah 6:10–11) Three features of the practice explain the concern. The instrument was controlled by one party. The other party could not verify it. And each transaction, taken alone, looked entirely fair. A dishonest scale does not announce itself. It simply tilts, a little, every time. FIGURE 2.1 Dishonest weights worked because one party controlled an instrument the other could not check. The same structure describes a report whose reader cannot inspect its definitions. This chapter’s argument rests on that structure, and it does not require stretching the text. These passages concern honest measurement in dealings between people. An analyst’s report is a dealing between people. The sponsor who reads Priya’s dashboard cannot inspect her denominators any more than the widow could weigh the stones. Whoever controls the instrument owes the truth to the person who cannot check it. That is the whole of the claim. The proverbs do not tell Priya which metric to choose. 2.2 Five ways an accurate number misleads Outright fabrication is rare in analytics and easy to condemn. The ordinary temptation is subtler: to select, from several accurate figures, the one that flatters. The modern bag of weights holds five stones. TABLE 2.1 Five ways an accurate number misleads Device | How it works | The question that exposes it Denominator | The rate is calculated on a base that leaves out the failures: those who finished, those who stayed, those who answered | Of whom, exactly? Window | The period is chosen after the results are known | Why these dates? Exclusions | Inconvenient records are removed under a rule written afterward | Who was taken out, and when was the rule set? Definition drift | The label stays while its meaning moves, so that a trend spans two different measures | Did “active” mean the same thing in both periods? Silence | The report omits the people the instrument never saw | Who is not in this data at all? Silence deserves a second look, because nobody has to do anything wrong for it to occur. In the community service, members who decline analytics consent still post requests. The server records the request because the service cannot function otherwise. The analytics tool records nothing. A dashboard built on the tool will undercount requests, and will undercount them most in whichever community declined most often. Nobody lied. The instrument simply could not see those people, and the report did not say so. MODEL IN DEPTH · A truthful number, a misleading report Priya’s illustrative figures for three weeks: requests posted rose from 40 to 80 to 120. Requests fulfilled within the agreed time were 30, then 52, then 66. The rising story. Requests tripled: 120 ÷ 40 = 3. Fulfilled requests more than doubled: 66 ÷ 30 = 2.2. Both statements are accurate. The falling story. The share fulfilled in time was 30 ÷ 40 = 75 percent, then 52 ÷ 80 = 65 percent, then 66 ÷ 120 = 55 percent: a decline of 20 percentage points. Requests that went unmet rose from 10 to 28 to 54. These statements are accurate as well. What explains both. Two further organizations joined the pilot in weeks 2 and 3. Volume rose because the population grew. Coordinators confirm each assignment by telephone, so every added request lengthened someone’s evening, confirmations slowed, and more accepted requests fell through. Minutes in the application rose partly because volunteers kept reopening it to see whether anyone had confirmed them. A report that says “usage tripled” selects the heavy stones. A report that says only “fulfillment is collapsing” selects the light ones, and ignores that 36 more households were helped in week 3 than in week 1. The honest report gives the count, the rate, the denominator, the reason the population changed, and what the team will do. It is longer. It is also true. FIGURE 2.2 Two accurate pictures of the same three weeks, drawn on separate scales. A count can rise while the rate that matters falls. 2.3 Creation: the person behind the record Why should it trouble anyone that fifty-four households went unserved behind a rising line? Most people feel that it should. Christianity gives a reason. Genesis presents human beings as made in the image of God, which locates a person’s worth in the Creator’s act, prior to anything the person does, earns, clicks or completes. (Genesis 1:26–31) A row in a requests table is a trace of someone who bears that image. The row may record that a meal was requested at 6:40 on a Tuesday. It cannot record that she had waited two days before asking. Creation also dignifies the analyst’s work. Human beings are given a world to tend, and tending requires attention: noticing, naming, distinguishing, keeping account. (Genesis 2:15, 19–20) Careful observation belongs to the human vocation. Scripture shows counting as an act of care as well. When Jesus tells anxious disciples that the hairs of their heads are all numbered, the point is that the Father’s attention to them is complete, and so they need not fear. (Luke 12:6–7) The passage is about God’s care for his people, and it should not be pressed into a charter for data collection. It does show that to number something is not, in itself, to diminish it. Two boundaries follow. Persons exceed their records, so no metric measures a person’s worth. And creatures are finite, so not everything that can be observed should be. Rest and limit are written into the creation account alongside work. (Genesis 2:1–3) 2.4 The fall: the heart that reads the chart Genesis 3 describes humanity’s turn from God, and what follows at once is concealment: the man and woman hide, then shift blame. Scripture’s wider diagnosis includes everyone. “All have sinned,” Paul writes, and Jeremiah describes the human heart as deceitful beyond our own ability to fathom. (Genesis 3; Jeremiah 17:9; Romans 3:23) Sin is a disordered relationship with God that bends desire and judgment. It should not be reduced to error or poor performance. For an analyst the relevant fact is that self-deception precedes deception. John says that if we claim to have no sin, the ones we deceive are ourselves. (1 John 1:8) Priya is in little danger of inventing numbers. She is in real danger of feeling, quite sincerely, that the rising line is the true story, because it is the story she hoped for, the sponsor wants, and Elena needs. The favorable reading arrives first. It always does. Institutions compound the pull. A sponsor who funds on activity invites reports about activity. A team praised for good news learns which news to bring. No one need intend harm for the scale to tilt. A sober analyst therefore builds procedures that assume her own bias: questions and definitions written before the data is examined, a colleague invited to attack the conclusion, limitations stated in the report itself. Scripture also shows that counting itself can go wrong. David’s census of Israel drew judgment, although God had commanded censuses before. The text does not fully explain the difference, and readers should be cautious about supplying one; Joab’s protest suggests that the king’s motive was at issue. (2 Samuel 24; 1 Chronicles 21; compare Numbers 1) The narrow lesson is enough. The same act of numbering can serve care or serve pride, and the numbers alone will not tell you which. 2.5 Redemption: what God has done in Jesus Christ The Gospel is news about something God has done. The New Testament presents Jesus as the eternal Son of God who became truly human, lived the faithful life none of us has lived, died for sins, and was raised bodily from the dead. (John 1:1–14; 1 Corinthians 15:1–8) His death and resurrection deal with the guilt and estrangement that human effort cannot remove. Salvation is therefore a gift. Grace is God’s undeserved favor toward sinners. Faith is trust in Christ himself, as opposed to reliance on one’s record. Repentance is turning from sin toward God, which is more than regret at being found out. Paul holds the pieces together in a single passage: saved by grace, through faith, not as a result of works, and created in Christ for good works that follow. (Ephesians 2:1–10; Mark 1:14–15) The order matters. Works follow grace. They never purchase it. This guards the chapter against a serious misreading. Honest analytics cannot reconcile anyone to God. A career of scrupulous denominators does not atone for sin, and a Christian organization’s worthy mission does not excuse a misleading report. Professional integrity is a fitting response to grace. It is not a substitute for it. Grace does change what happens at the desk, in at least two ways. If a person’s standing before God rests on Christ and not on results, she can afford to read an unfavorable chart honestly, because her worth is not on the chart. And when she finds that last month’s figure was wrong, she can say so. John’s promise is that those who confess are forgiven and cleansed. (1 John 1:9) Forgiveness does not erase consequences or the duty to repair. It removes the need to hide. MODEL IN DEPTH · Correcting the record In week 5 Priya discovers that her week 3 report overstated acceptance of requests by volunteers. A release of the Android application had stopped sending the acceptance event for nine days, and she had filled the gap with an estimate that proved too generous. The technical response is to repair the figure from the server’s record, mark the affected dates in every chart, and add a check that compares event counts with server counts each week. The accountable response is to tell Jonah and the sponsor, in writing, what was wrong, by how much, and since when. A correction notice is brief: the original statement, the corrected statement, the cause, and what has changed so that it will not recur. Nothing requires her to do this. The sponsor would never have known. That is exactly the situation the honest-weights passages address. 2.6 Restoration: hope, and the limits it sets Christian hope looks to a renewed creation in which God dwells with his people and death, mourning and pain are no more. (Revelation 21:1–5) Its fulfillment belongs to God. That hope supports patient, unglamorous work in the meantime, including the work of helping a community service confirm deliveries reliably. It also sets limits on what such work may claim. Analytics cannot measure a person’s spiritual condition, and a ministry that treats engagement figures as a proxy for discipleship has confused two different things. A retention curve is not the kingdom of God. Better measurement will make the service more truthful about itself and more useful to the people it serves. Those are real goods. They are also modest ones, and an analyst who knows the difference is less likely to overpromise. 2.7 From a passage to a decision without skipping the reasoning Throughout this book, biblical passages are brought to bear on professional decisions by four steps. First, establish the passage’s subject in its own context. Second, state the principle it teaches. Third, investigate the circumstances to which the principle might apply. Fourth, propose an action and acknowledge where the analogy stops. The Product Steward develops the same rule for product decisions. (Talluri 2026) TABLE 2.2 The four steps applied to honest weights Step | Applied to Priya’s report Passage in context | Laws and proverbs forbidding dishonest weights in trade, where one party controls a measure the other cannot verify Principle | Whoever controls an instrument of measurement owes a truthful measure to those who rely on it Circumstances | The sponsor relies on Priya’s figures and cannot inspect units, windows, denominators or missing populations Bounded action | Report counts with rates, state the denominator and who is missing, explain the change in population. The passage does not choose the metric or the chart. Some passages reach a practice directly because the practice falls within their subject. Honest weights, truthful speech, presumption about the future and stumbling blocks before the blind are examples, and later chapters use them. Other passages offer only an analogy, which can illuminate so long as its limits are stated. Deuteronomy 19:15 requires two or three witnesses to establish a criminal charge. It is a rule for courts. This chapter borrows from it no more than an instinct: a serious claim should not rest on a single testimony. Resist the verse that merely sounds like analytics. Hear what a passage claims before recruiting it. A four-movement reflection for an analysis TABLE 2.3 A four-movement reflection for an analysis Movement | Question for the analyst | Boundary Creation | Whose traces are these, and what good is the analysis meant to serve them? | A person’s worth is not a quantity in the dataset Fall | Which reading do I want to be true, and who has checked it? | Naming a bias does not remove it Redemption | What would truthfulness and, where needed, correction require here? | Honest reporting does not earn standing with God Restoration | What modest good can this evidence support, and what can it not measure? | A healthy metric is not the fulfillment of Christian hope 2.8 Working beside people who believe differently The methods in this book were built by many hands, most of whom did not share the author’s faith. Christians have a category for this. Common grace names God’s generosity in giving insight and skill across humanity, distinct from the saving grace of the Gospel. It is a reason to receive a colleague’s correction with gratitude and to credit good work wherever it is found. A colleague of another conviction can oppose a misleading chart for reasons of her own, and professional bodies state duties very close to this chapter’s. The shared standard in the workplace and the classroom is reasoning that others can inspect. Assignments in this book ask you to explain and apply a position coherently. They never require you to profess it. FROM THE FIELD · A public standard for honest numbers The United Kingdom’s Office for Statistics Regulation publishes a Code of Practice for Statistics, now in its third edition, which sets standards for producing and using official statistics. It is organized around three pillars: Trustworthiness, Quality and Value. Its principles include “Be transparent,” “Be open about quality” and “Be clear.” The Code also carries standards for the public use of statistics by officials. One commitment reads: “Publish data used in public statements.” The regulator calls the wider habit intelligent transparency. A minister who quotes a figure should make it possible for a citizen to find the figure and check it. (Office for Statistics Regulation 2025) Set that beside the grain seller. A secular regulator, writing for a plural society, has arrived at the duty this chapter draws from Proverbs: the one who wields the number owes the means of checking it to the one who cannot. The American Statistical Association’s ethical guidelines press the same obligations on individual practitioners. (American Statistical Association 2022) The Gospel itself remains an invitation addressed to persons. If its claims are new to you, read the Gospel of Mark and the passages cited in section 2.5, and consider who Jesus says he is and why he calls people to repent and believe. A Christian you trust, or a local church, can help you weigh it. What Christianity commends is trust in Christ, which is something other than adopting a more careful reporting style. 2.9 Let the kinds of evidence question each other Chapter 1 distinguished four kinds of evidence and introduced triangulation. Here it becomes a working method with a moral edge. The commonest way to mislead with accurate data is to consult only the source that agrees with you. An evidence agreement matrix makes that harder. List the claims a report might make down the side and the available sources across the top. For each cell record one of three words: supports, contradicts, or silent. Then read the matrix. A claim supported by every source that speaks to it can be stated plainly. A claim that one source supports and another contradicts is a finding in its own right, and it identifies where to look next. A claim on which every source is silent should not be in the report. DATA BLOCK · Evidence agreement matrix · template CLAIM | EVENTS | SERVER | LOG | NOTES | SURVEY ------------------------------+-----------+----------+----------+----------+-------- Requests are rising | supports | supports | supports | silent | silent More people are being helped | silent | supports | supports | silent | silent The service works better | silent | CONTRA | CONTRA | CONTRA | mixed Volunteers are more engaged | supports? | silent | silent | CONTRA | CONTRA RULES 1. Fill the cells before drafting the headline. 2. "supports?" means the number fits but has a rival reading. Name the rival. 3. Every CONTRA needs a sentence in the report. 4. A row of silent cells is not a finding. Row four repays attention. Rising minutes in the application fit the claim that volunteers are more engaged. Priya’s field notes supply the rival reading: volunteers reopening a request again and again, waiting for a confirmation that arrives by telephone or not at all. The minutes are a trace of anxiety. Only someone who watched could have known. DATA BLOCK · Community service dataset · the three sources used in this chapter SOURCE GRAIN KEPT BY TIME SEES events one per event analytics tool UTC consenting members requests one per request application server UTC every request coordinator_log one per call coordinators, hand Central off-screen work FIRST EVENT NAMES object_action · past tense · snake_case invite_viewed fires before login; carries no member_id account_created consent_updated detail: analytics=granted | declined request_post_attempted client: the button was pressed request_posted client: server confirmed; may repeat on retry request_viewed request_accepted a volunteer takes a request volunteer_activated server-derived: first acceptance within 14 days DIMENSION RULES org_id ORG-A | ORG-B | ORG-C role member | volunteer | coordinator platform android | ios | web | server pilot_week derived from LOCAL time Appendix B holds the full dictionary. Chapter 4 teaches you to distrust it properly. Faith and Practice Reflection FAITH & PRACTICE · Faith and Practice Reflection God delights in a just weight. The proverb makes accuracy toward a neighbor a matter of worship, which dignifies every tedious hour spent reconciling two tables that ought to agree. Vocational diligence of this kind is rarely seen and seldom praised. Do it anyway. Fifty-four households are easier to overlook than a rising line, and someone in the room must be willing to go and count them. 2.10 The virtues of an honest measurer Methods can be taught in a semester. Character takes longer, and it determines what a practitioner is willing to check, report and repair. The table connects six virtues with concrete habits. None is exclusively Christian, and a reader of any conviction can practice them. The Christian claim is about their source and their end. TABLE 2.4 The virtues of an honest measurer Virtue | A concrete analytic habit | A question for examination Truthfulness | State unit, window, denominator and exclusions beside every rate | Which omitted fact would change the reader’s mind? Humility | Publish limitations in the report, not in an appendix nobody opens | Whose correction have I resisted? Justice | Ask who is missing from the data and who bears the cost of the gap | Whom can this instrument not see? Prudence | Match the strength of a claim to the strength of its evidence | What would I need before saying this more firmly? Courage | Bring the unfavorable finding to the person who least wants it | Which result have I been slow to share? Love of neighbor | Collect only what serves the people observed | Would they thank me for keeping this? Actionable Application MODEL IN DEPTH · Worked scenario Priya discards her first headline, “Usage tripled in three weeks.” Her second draft reads: “Requests tripled as two more organizations joined the pilot. The share fulfilled within the agreed time fell from 75 to 55 percent. Coordinators are confirming by telephone and cannot keep pace, and volunteers are left waiting. We propose an in-application confirmation step before adding further organizations.” Beneath it she lists the denominator for each rate and notes that members who declined analytics appear in the server record and not in the dashboard. Elena reads it twice. “It’s less exciting,” she says. “I’d rather send this one.” Project worksheet Add a responsibility statement to the analysis brief you began in Chapter 1. Keep each response specific enough that a colleague could disagree with it. TABLE 2.5 Project worksheet: the responsibility statement Worksheet field | Complete for your analysis | Your response People observed | Whose traces will you use, and did they agree to it? | ________________ People unseen | Whom can your instruments not see, and why? | ________________ Reader who cannot check | Who will rely on your figures without access to your definitions? | ________________ What you owe that reader | Which units, windows, denominators and limits will you state? | ________________ The reading you hope for | Which result do you want, and who will challenge it? | ________________ Limit on collection | What will you decline to collect or read? | ________________ Chapter summary Scripture treats honest weights as justice toward a neighbor who cannot check the scale, and an analyst’s reader stands in that neighbor’s place. Accurate numbers mislead through denominators, windows, exclusions, drifting definitions and silence about the unseen. The Gospel announces salvation by grace through faith in Christ; honest work follows that gift and cannot earn it. Grace frees an analyst to report unfavorable results and to correct the record. An evidence agreement matrix sets kinds of evidence against one another, and a responsibility statement names the people an analysis affects. Practice and reflection APPLICATION · Guided practice A ministry’s annual report states: “92 percent of participants said the program deepened their faith.” A footnote adds that the survey was offered at the final session. Identify the device from Table 2.1, explain what the figure can and cannot support, and rewrite the sentence honestly. Assume 140 people enrolled, 60 attended the final session and 50 completed the survey, of whom 46 agreed. Guidance: The device is the denominator, reinforced by silence. 46 ÷ 50 = 92 percent of respondents. Against those who enrolled the figure is 46 ÷ 140, about 33 percent, and the 80 people who stopped attending had no opportunity to answer. An honest sentence: “Of 140 who enrolled, 60 completed the program; 46 of the 50 who answered our closing survey said it deepened their faith.” Note also what the survey cannot measure at all. Agreement with a statement on a form is evidence about what people say. It is not a measurement of anyone’s faith. APPLICATION · Independent application For the site you adopted in Chapter 1, find one public claim its owner makes with a number: a count of customers, a satisfaction score, a rate of growth. State which of the five devices could be at work, what you would need to see to check the claim, and whether the owner has made that possible. Then complete the responsibility statement for your own analysis brief. Readers working from another ethical framework should state its premises with equal clarity and address the same evidence. The shared standard is reasoning that the people affected could examine. Review questions Explain why a dishonest weight was hard for its victims to detect, and identify the parallel in a modern report. Describe two accurate statements about the same data that leave opposite impressions. Explain grace, faith and repentance to someone with no church background. State why honest analytics cannot earn salvation, and what grace nevertheless changes in an analyst’s practice. Distinguish a passage that addresses a practice directly from one that offers only an analogy, with an example of each. Explain what a “contradicts” cell in an agreement matrix obliges a writer to do. YOUR DECISION ARTIFACT Attach your responsibility statement to the analysis brief from Chapter 1. In Chapter 3 you will decide which outcomes deserve measuring; the “reading you hope for” line will help you notice when a goal has been chosen because it is easy to hit. In Chapter 4 the “people unseen” line becomes a section of your measurement plan. Sources and further study The biblical argument draws on Genesis 1–3; Leviticus 19:35–36; Deuteronomy 19:15 and 25:13–16; Numbers 1; 2 Samuel 24; 1 Chronicles 21; Proverbs 11:1, 16:11, 20:10 and 20:23; Jeremiah 17:9; Amos 8:4–6; Micah 6:10–11; Mark 1:14–15; Luke 12:6–7; John 1:1–14; Romans 3:23–24; 1 Corinthians 15:1–8; Ephesians 2:1–10; 1 John 1:8–9; and Revelation 21:1–5. The chapter paraphrases and interprets these passages; it does not present its applications to analytics as part of the biblical text. American Statistical Association. 2022. Ethical Guidelines for Statistical Practice. Committee on Professional Ethics. Keller, Timothy, with Katherine Leary Alsdorf. 2012. Every Good Endeavor: Connecting Your Work to God’s Work. Dutton. Office for Statistics Regulation. 2025. Code of Practice for Statistics. Edition 3.0. UK Statistics Authority. Includes the Standards for the Public Use of Statistics, Data and Wider Analysis. Talluri, Sean. 2026. The Product Steward: Building for Human Flourishing in the Age of AI. Rev. ed. Chapter 2 presents the Gospel and the four-step rule in the setting of product decisions. Wolters, Albert M. 2005. Creation Regained: Biblical Basics for a Reformational Worldview. 2nd ed. Eerdmans.