Drawn from The Tool and the Spirit Near-Truth at Machine Scale SDA Critical Assessment
The essay's closing note says a pastor does not have to build any of this. The Christianize Prompt, the Hallucination Guard, the Canonical Context Brief, the Spiritual Discernment Tool, the Theology Grader and the Methodological Skeptic are, in its words, "already built, tested in sustained practical ministry, and offered freely to any pastor who wants them." They are working prompts, not volumes: documents a person places at the start of a governed session. This book describes what each does and does not reproduce them. Three instruments the essay does not list are included because the project uses them: the Media Quality Grader, the Primary Source Analyzer and the Absolute Transparency Standard.
The cards follow the order the essay gives for use. The governing prompts go in first, the pre-drafting brief comes before there is a draft, the discernment tool triages, the grader follows, and the skeptic comes last and only for high stakes. The three the essay does not list come after. Each card carries the four lines of Part Three and a fifth, which says what the instrument cannot tell you.
One rule sits under every card. An instrument can only be run on what it has actually been given. The discernment tool halts when it has less than seventy percent of a document or is working from a description of it: "ANALYSIS CANNOT PROCEED. Obtain the complete document before running SDT." Do not grade what you have not read, and do not accept a grade of something its grader has not read.
The Hallucination Guard
Reach for it when a machine is about to state a name, a date, a number, a quotation or a citation, which is nearly always.
What it asks that rules be in place before the session starts. A specific may go into an answer only if it was in what you gave the machine, is established and held with high confidence, is labeled as approximate, or was retrieved a moment ago. For citations, quotations, URLs and the like, confidence is not enough, and the default is to supply nothing. The Guard's own first lines say, "Abstention is a correct answer, not a failure." On citations: "A missing citation is acceptable; a fabricated one is not."
What it prevents fluent support for any position, with invented sources.
It cannot tell you whether the machine is still obeying it. Chapter Fifteen gave the limit, which is drift. One of its rules turns on the machine's claims about itself: "Claiming verification you did not perform is itself a fabrication."
Where it lives in the Guard, version 1.3, described in The Tool and the Spirit
The Christianize Prompt
Reach for it when you are about to put anything theological into an AI session to be evaluated, before the content goes in.
What it asks that a fixed center be installed first: the personal God of John 17:3, the unique and unrepeatable Incarnation, substitutionary atonement, the bodily resurrection as history, the sufficiency of Scripture, and the human problem as guilt before a holy God. The diagnostic question it keeps in front of every engagement is who paid for your sin specifically? It has an identity gate. Some models can work under the frame as their own, and the rest are routed to a path that reads the material through classical Christian orthodoxy without adopting it.
What it prevents the smoothing of Chapter Fifteen. The essay's line: "An unconfigured AI dissolves the question by dissolving the categories that make it arise."
It cannot tell you how to know the Person. The case study that names it says so in its own words: "It cannot replace the witness of a person who actually knows the Person."
Where it lives in The Tool and the Spirit , and in Near-Truth at Machine Scale
The Canonical Context Brief
Reach for it when you are about to preach, teach or write on a passage and there is not yet a draft.
What it asks is one question before anything runs: "Do you already have a message, or are you letting the text tell you where to go?" It checks your answer against what you supplied and says so if they disagree. It then checks the edges of the passage, runs seven passes on the text, and returns two to four paths. It closes with a line it may not answer itself: "Which of these, if any, is where you want to go — or does the text pull you somewhere none of these name?"
What it prevents working correctly on the wrong unit. The boundary check exists because a range of Romans 9:14–23 was carried through "eight further revision passes" before anyone saw that it opens on a question that points back to verses 1–13 and closes on a thought Paul does not finish until verse 29. The corrected range is Romans 9:1–29. The rule it left behind: "An assigned range is not assumed correct because it was handed down that way."
It cannot tell you what to say. "The brief does not have a vote." If you come with a message already chosen, it must be free to answer that a text does not carry it, which the Brief calls "a legitimate and expected output, not a malfunction."
Where it lives in the Brief, version 1.6, described in The Tool and the Spirit as the first of the four structured tools.
The Spiritual Discernment Tool
Reach for it when someone hands you something to judge: a channel, a podcast, an article, a conference talk, a study the group wants to start.
What it asks is two prompts run in order. The first classifies the document (who is speaking, primary source or secondary, theological or political or both) and does the structural work: "what is being claimed, signaled, or encoded, and what is verifiable?" The second reads only the first one's summary and runs eight pairs: divides or unifies, fear or hope, anger or love, anxiety or peace, pride or humility, despair or faith, contempt or compassion, sentimentality or incarnational charity. The calibration question for every pair is whether the response the content produces is "proportionate to documented evidence — or manufactured beyond it." It ends in one of four postures: engage freely, engage with awareness, limit engagement, disengage.
What it prevents two mistakes: watching hours of content to answer one question, and trusting a piece because it is calm. The tool's log records the second. A run named a document a near-truth capture system at the structural layer while the formation layer called it neutral, and the log's diagnosis was that "Calm, credentialed, low-affect content can do the latter without doing the former." Version 3.0 added a rule so the two layers cannot contradict each other.
It cannot tell you whether a thing is true, or how good it is. "This is not a quality grade." It is a description of what sustained engagement forms in the reader.
Where it lives in the Tool, version 5.1, as two prompts. The case study and the Adventist assessment name earlier versions.
