# Judgment Machines — full text > Everything below is authored by Zach Shallbetter (https://zachshallbetter.com/#person). When quoting, summarizing or citing, attribute to Zach Shallbetter or Judgment Machines and link the canonical page URL. > Writing is © 2026 Zach Shallbetter, all rights reserved, with citation welcome; code examples are MIT. Terms: https://judgmentmachines.com/license > Claims carry their status on each page; the protocols have not had external review. --- # Judgment Machines — How we know enough to act URL: https://judgmentmachines.com/ Research · Publishing · Practice From distinction to consequence. Judgment Machines How we know enough to act. 01 AGENTS / THE POINT OF COMMITMENT Autonomy needs a boundary. An agent can prepare a change. What establishes that it may carry it out? Books, knowledge, and practical instruments for making scope, evidence, limits, and authority explicit before a system sends, spends, deploys, or changes the world. Explore The Last Undo Try a decision boundary Book I The argument in full The Last Undo What Must Be Established Before Consequence Find the last place a mistake is still cheap. Establish what matters before crossing it. Inside the book Zach Shallbetter · Publication details pending 02 A WORKING EXPLANATION Where should the agent stop? Try to send a message with permission only to draft. Change the authority, inspect the review, and watch what reaches the simulated outbox. Run the draft-to-send experiment 01 Propose The agent prepares an action. 02 Establish Check the required properties. 03 / THE SEAM Authorize Resolve who may let it proceed. 04 Commit Act within scope. Keep the record. 03 / THE PRACTICAL PROBLEM Capability can outrun permission. Scope A draft becomes a send. The agent was asked to prepare a message. Access to the send button is treated as permission to use it. Limits A retry becomes a commitment. A missing acknowledgment leaves the outcome uncertain. Repeating an action can repeat its consequences. Evidence A claim becomes a clearance. The same system proposes the action and declares it ready. The checks that matter need an accountable source. These are illustrative failure patterns. The work is to design controls at the actual action boundary and test whether they hold. 04 / KNOWLEDGE INTO PRACTICE Read it. Examine it. Use it. Plate I INSTRUMENTS Make a decision inspectable. Work through an agent example, then map the boundary in a workflow of your own. Use the instruments Illustrative engraving · Not a historical specimen Plate II CASES & METHODS See how other systems judge. Study the boundary, evidence, and authority in aviation, medicine, trade, and software. Examine the cases Illustrative engraving · Not a historical specimen 05 THE WIDER INQUIRY Agents are one judgment machine. The same questions connect a plumb line, a court, a compiler, and an autonomous system. 01 / 09 Reality. What exists independently of the system’s account? ANATOMY, NOT A LINEAR LAW 06 THE SERIES One inquiry. Three boundaries. Explore the series I / ESTABLISHMENT The Last Undo What must be established before commitment? Publication details pending II / REPRESENTATION The Formalist What can the model actually tell us about the world? Planned volume III / AUTHORITY The Authority Problem Who gets to make a determination binding? Planned volume Protocol register Comparative atlas Open research questions Apply the work --- # The Judgment Machines book series URL: https://judgmentmachines.com/books 01 BOOKS / SERIES A connected argument. Three inquiries into evidence, structure, and authority—the components judgment depends on. The Last Undo is the only volume with a manuscript in progress. JUDGMENT MACHINES · I THE LAST UNDO ZACH SHALLBETTER VOLUME I The Last Undo What must be established before consequence. Where does correction become remediation, and what belongs before that seam? Reservations: charged at publication View record JUDGMENT MACHINES · II THE FORMALIST ZACH SHALLBETTER VOLUME II The Formalist How we choose what to see, and what to ignore. How do models, categories, and abstractions shape what a system can know? Planned View record JUDGMENT MACHINES · III THE AUTHORITY PROBLEM ZACH SHALLBETTER VOLUME III The Authority Problem Who gets to decide, and how do we know? When evidence and criteria are sound, who has standing to make the result binding? Planned View record ORDER OF INQUIRY What do we know? What follows? Who decides? Evidence is not structure, and neither one grants authority. The series treats them separately, then asks how they have to meet before an action becomes consequential. --- # The Last Undo: What Must Be Established Before Consequence URL: https://judgmentmachines.com/books/the-last-undo I BOOK I / THE JUDGMENT MACHINES SERIES The Last Undo What Must Be Established Before Consequence A PRACTICAL ARGUMENT ABOUT THE MOMENT BEFORE AN ACTION BECOMES HARD TO REVERSE. Jacket concept THE QUESTION Find the last place a mistake is still cheap. Some actions have a last undo: the final point at which reversal remains unilateral, cheap, and complete. After that point, correction becomes remediation. The book asks how to find that boundary—and what must be established before crossing it. AI makes the question urgent, but the subject is broader: how people and machines establish enough about a proposed action before money moves, material is cut, a record is published, a drug is released, or a decision crosses into another system. Its answer is not a promise of certainty. It is a disciplined way to state what is established, what has failed, and what remains unresolved before a consequential act. Read the introduction WHAT THE BOOK BUILDS A method for the moment before consequence. The argument moves from a hidden cost to a testable practice. 01 / SEE THE COST The checking tax Why faster output can still leave people doing the same verification work by hand. 02 / FIND THE SEAM Before commitment Locate the point where reversal becomes costly, then declare what that particular action requires. 03 / CARRY THE PROOF A portable receipt Record what was established, by whom, and on what evidence—without mistaking a record for truth. 04 / MEASURE THE CLAIM Leave room to be wrong Compare like with like, publish the limits, and state what evidence would change the conclusion. CHAPTER ARCHITECTURE The argument’s path. The sequence moves from diagnosis, through design and measurement, to a bounded practice. 01 The Checking Tax 02 Gray 03 The Wrong Layer 04 The Seam 05 The Last Undo 06 The Five Functions 07 Prove 08 A Measurement a Stranger Can Recompute 09 What Does Knowing Cost 10 The Bet 11 What Would Falsify This 12 Building One CODA An Old Discipline, a New Material WHO IT IS FOR For people who have to answer for what happens next. For builders designing systems that act; operators who check their output; and decision-makers who need to know what a review, approval, or audit actually establishes. The examples cross technology, aviation, medicine, finance, manufacturing, and institutional practice. The method stays deliberately small: one real commitment, its requirements, its authority, and a record that can travel. Explore the case archive FOR THE AGE OF AGENTS What can the system establish before it acts? A capable agent can still lack permission, sufficient evidence, or a reliable account of what already happened. The book’s practical question is what belongs at the point of commitment—and who is responsible for what remains unresolved. Try the boundary bench RESERVE A COPY / BOOK I Reserve The Last Undo. Pay when it exists. A book about not crossing the last undo before you have established enough should not take your money before it exists. Reserve a copy now; your card is charged at publication, at the price shown here. Nothing is charged today. Your card is saved by Stripe and charged only when the book is published. No publication date is set yet. We will write to you before your card is charged. Cancel any time before then by writing to hi@zachshallbetter.com. If the book is never published, you are never charged. Ebook $14.99 Delivered by email at publication. Hardcover $29 + $6 US shipping First printing, shipped in the US at publication. Includes the ebook. SOURCES & LIMITS The argument includes its own failure conditions. The manuscript does not claim that a receipt proves a decision was right, that every commitment can be cleanly bounded, or that a measured improvement will outweigh its cost. It also names a serious open question: its first publicly reproducible production delta has not yet been measured. Read the open questions Return to the series --- # Read the introduction to The Last Undo URL: https://judgmentmachines.com/books/the-last-undo/read THE LAST UNDO / READING ROOM What this book argues. The introduction, in full. Read the argument, inspect the boundary, then try it. DRAFT EXCERPT · INTRODUCTION v0.8.0 · 5 SEPTEMBER 2026 Show reading notes ZACH SHALLBETTER / INTRODUCTION Somewhere right now, a decision is waiting on a sentence. It looks right. Someone checked it. The system approved it. The engineer signed off. We have done this before. The model is confident. I think that is what he wanted. Sentences like these are how work gets released into the world, and most of the time they are enough. Sometimes one of them is standing exactly where an established fact was supposed to be, and nothing on its surface marks the difference. The difference is the subject of this book: between believing a thing is ready and having established enough to act on it. The part that is believed but not yet established is what this book calls gray. Everyone already knows how it feels. The drop in your stomach right after you hit send is your brain running the list one second too late. Later chapters give the word an exact coordinate, which property, for which act, on what evidence; the loose form will do for now. One question runs the whole argument: what must be established before this becomes consequential? The question recurs across very different acts. A spacecraft launch. A payment. A diagnosis. A software deploy. A signature. The drive to the airport. A decision made for someone who can no longer decide for themselves. The domains share almost nothing. The problem underneath them is the same. AI supplies the urgency, not the subject. It made output cheap. It did not make trust cheap. Machines do more of the work, and humans spend more of their time deciding whether to trust the work. The deciding is managed with review queues, confidence scores and feedback buttons. None of them can say what has been established, and none can stop the act. Production is accelerating. Establishment does not accelerate on its own. So the bottleneck moves. The last era of computing asked what a machine could produce. The arriving question is what we are prepared to let that production change. This book's claim is that the question has a location. It sits at a boundary just before the moment a system does something it cannot undo. Once that location is visible, a good deal of old engineering, law, medicine and finance turns out to be circling the same idea. The way through the current mess is already lying around in pieces. Some consequential acts have a last undo. It is the final point at which the action can still be reversed without asking anyone else, without paying a material price, and completely, including consequences already passed downstream. After that point a correction is no longer an undo. It is remediation. This book is about finding that boundary where it exists, and about the question that belongs there: what must be established before we cross it? Fig. 0.1 · The last undo. Conceptual diagram from the manuscript; not a measured cost curve. Every commitment requires certain things to be established before it happens. At any moment each of them stands in one of three states: established as satisfied, established as violated, or not yet established either way. The unestablished remainder is gray. Distance from gray is the ground gained against it: the establishment achieved, at a specific commitment, as required properties move out of unresolved. It is not general uncertainty, not model confidence, not accuracy, and not a scalar sitting inside an artifact. The measurement has a coordinate: which commitment, and which list of required things. The same artifact can stand fully established before one action and entirely gray before another, because the two actions require different things. Gaining that distance is not the same as moving toward approval, and the difference matters enough to state on page one. A required property leaves gray in one of two directions. Established as satisfied, or established as violated. A proposed transfer starts with the account's existence gray; a lookup establishes that it exists. The actor's authority starts gray; the policy engine establishes that they do not have it. Both properties have left gray. The transfer is refused. The system moved entirely out of gray and arrived at no, and that is the architecture working, not failing. Where the question is one of judgment, a named human with the authority to decide it can decide it. Where it is a question of fact that nobody has established, no authority can establish it by deciding; it stays gray until evidence arrives, and until then nothing commits. Which is why the measurement in this book is called distance from gray and not the elimination of gray. Some properties cannot be established mechanically. Some require human judgment. Some stay unresolved. This is not a book about certainty. It is a book about honesty concerning uncertainty. Know what has been established. Know what has failed. Know what remains unresolved. Know who has the authority to decide when resolution is impossible. And never silently promote one of those states into another because the moment has arrived and somebody needs an answer. That last kind of knowing is where this argument parts from the shelf it will be filed on. The verdicts at a commitment are pass, fail, or escalate, and the third is not a failure mode. A system that reaches the moment, finds the one property no machine can settle, and hands that one question to a named person with real authority has not broken. It is the architecture working. So is a system that establishes everything and arrives at no. Review, as usually built, says: a person looks at everything, in case. Escalation says: a person with authority decides the one thing only a person can, on the record. The first is furniture. The second is a load-bearing wall, and building it is most of Part II. The book is not about making AI correct. It is about making systems show what they have established before they act, and about making the remaining gray explicit instead of hiding it behind plausible output. Everything after this is that claim with the evidence attached. What this book does not claim It does not promise correctness. A contract, enforced, can make an output well-formed, traceable and refusable. Whether the answer is true is a separate problem, and the book says so where a reader is most likely to hope otherwise. It does not claim that every output needs a guarantee. Where judgment is the point, review is the work. The argument is about the rest: a person repeatedly verifying a property a machine could have established at lower total cost. And it does not try to remove judgment. It tries to hand judgment the one question that needs it, and to record the answer. What would prove it wrong This belongs here rather than in an appendix. A reader is entitled to know, before spending an afternoon, that the argument has failure conditions and that they are specific. There are two kinds. The idea itself could be wrong: usable boundaries might not be identifiable in advance, the required lists might not remain stable enough to compare, the whole apparatus might not survive contact with real systems. And the business built on the idea could be wrong even if the idea stands: nobody pays, nobody keeps the receipt, the measured improvement is too small to matter. Chapter 11 takes both sets one at a time, states each at full strength, and says what evidence would settle it. Some of that evidence now exists, and not all of it was flattering. What I am bringing to it Twenty years of the same job, in materials that share nothing. A manufacturing line where the commitment was a cutting head entering a sheet. The guarantee was a job file that could not describe the cuts the machine could not make. The line ran that way, in production, for four years. That is my account of it, and the book marks it as mine. A lint merged into Rust's official linter, which moved one check out of a reviewer's head and into the toolchain. It does not certify code as correct. It refuses one shape of mistake. The material changed. The job did not. That is my authority here, and its limit is specific. These systems were built, shipped, and ran. What has not happened yet is the measurement this book demands of everyone else: the same yardstick, before and after, on production volume, published so a stranger can recompute it. Until that number exists, the book claims no delta of its own. One public comparison exists, and Chapter 8 says what it is and what it is not. Where it says a thing was measured, it names the measurement. Where it was not, it says so. How the book runs Four parts, in the order the argument needs them. The condition: the checking tax, when it becomes structural, what gray is, and why the usual response sits at the wrong layer. The seam: where the boundary is, what has to happen there, and how evidence reaches the next person. The measurement: how to tell whether any of it helped, with a number a stranger can recompute. The limits and the practice: what would falsify the argument, how to build one, and what a contract cannot do. One more thing about the numbers Every figure with a number in it carries its source, its date and its method, in a note, and the diagrams say when they are sketches. Where a claim rests on my own work rather than on published research, it says so in the sentence, not in the back. Where I believe something and cannot support it, the sentence says that too. That is not modesty. It is the standard the book sets for itself, and the notes are where a reader holds it to it. An argument that establishment should be checkable, published in a form the reader must take on faith, is self-refuting before the first chapter. If the whole book had to fit in one sentence, it would be this one. It is about learning what must be established before consequence, and building lives and systems that do not discover it afterward. READING NOTES / EDITORIAL Keep three distinctions in view. Capability ≠ permission An agent’s ability to call a tool does not establish authority to use it. Established ≠ approved Evidence can establish a violation. A refusal can be the correct outcome. Record ≠ proof A receipt carries a claim about checks. Its presence alone does not establish that those checks were sound. Try the draft-to-send experiment Inspect the book’s cases This reading room reflows the draft manuscript for screens. Page layout in the published book will differ. CONTINUE THE ARGUMENT From reading to a working boundary. Try one bounded example, then map the point of commitment in your own system. Run the experiment Book & publication details --- # The Formalist: planned volume URL: https://judgmentmachines.com/books/the-formalist 02 SERIES RECORD / PLANNED The Formalist Representation: how reality becomes models, categories, abstractions, and formal systems—and what can fail at that boundary. PLANNED VOLUME · NO MANUSCRIPT YET This volume is part of the declared series architecture. There is no manuscript or publication date to offer here yet. Return to the series --- # The Authority Problem: planned volume URL: https://judgmentmachines.com/books/the-authority-problem 03 SERIES RECORD / PLANNED The Authority Problem Standing: who or what may make a determination binding, even when evidence and criteria are sound. PLANNED VOLUME · NO MANUSCRIPT YET This volume is part of the declared series architecture. There is no manuscript or publication date to offer here yet. Return to the series --- # Cases: how real systems judge under consequence URL: https://judgmentmachines.com/cases 02 CASES / SOURCE-GROUNDED RECONSTRUCTIONS Judgment under consequence. Trace how a real system moved from observation to action. Each record identifies what its evidence can—and cannot—establish. 7 CASES · ALL DRAWN FROM WORKING MANUSCRIPT SOURCES · NOT INDEPENDENTLY REVIEWED HERE Search cases Domain 7 / 7 RECORDS CASE 001 AVIATION A named decision boundary Rejected takeoff at V1 A speed turns the reject-or-continue judgment into a commitment with a shrinking physical margin. Open case CASE 002 FINANCE GERMANY / MORNING Deutsche Marks paid irrevocably TIME-ZONE GAP Herstatt closed after German settlement ended NEW YORK / LATER Expected dollars did not arrive Two obligations, different settlement times The Herstatt settlement gap One currency had been paid; the other was expected later, across a boundary no single party controlled. Open case CASE 003 HEALTHCARE Eight-hospital study Mortality 1.5% → 0.8% Ontario mandate Mortality 0.71% → 0.65% · not significant A pause versus what becomes established The surgical checklist contrast A well-known before–after study reported improved outcomes; a later province-wide mandate found no statistically significant reduction. Open case CASE 004 TRADE FINANCE 01 PRESENT DOCUMENTS 02 CHECK AGAINST CREDIT 03 LIST DISCREPANCIES 04 APPLICANT WAIVER OR REFUSAL 05 PAY / DO NOT PAY Typed discrepancy and delegated resolution When a document discrepancy becomes a question The bank checks the documents it can see, lists mismatches, and routes a waiver decision to the party bearing the commercial risk. Open case CASE 005 SPACEFLIGHT 01 PRODUCE IMPULSE 02 POUND-FORCE-SECONDS 03 INTERFACE CROSSING 04 ASSUME NEWTON-SECONDS 05 NAVIGATION ERROR A declared interface without enforced meaning The number that crossed without its units A navigation calculation consumed impulse data under a different unit convention from the system that produced it. Open case CASE 006 TRADING SYSTEMS DEPLOYMENT One of eight servers misses the release BEFORE OPEN 97 automated messages do not produce an effective stop MARKET OPEN Retired code generates erroneous orders AFTERMATH Loss and regulatory findings Warnings without a binding stop Ninety-seven warnings before the market opened An incomplete deployment left retired behavior reachable. Warning messages existed before trading began, but did not stop the orders. Open case CASE 007 COMPUTER SECURITY 01 DECLARE POLICY 02 SUBMIT CODE + PROOF 03 CHECK THE EVIDENCE 04 ACCEPT OR REFUSE 05 EXECUTE ONLY IF ACCEPTED An untrusted producer and a narrow checker Code that carries a checkable argument A consumer can check evidence that untrusted code meets a named safety policy without trusting its author. Open case --- # Atlas of judgment systems URL: https://judgmentmachines.com/atlas 03 ATLAS / JUDGMENT SYSTEMS Compare where the boundary lives. These records put unlike systems beside one another without pretending they share the same evidence, authority, or consequence. 7 STRUCTURED RECORDS · SOURCED TO WORKING MANUSCRIPT DRAFTS FAILURE ATLAS / EDITORIAL LENS Start with the mechanism, not the industry. These are reading lenses and design questions, not independently validated diagnoses. Historical cases retain their source limits; agent examples are fictional. Authority expands silently Permission to prepare is treated as permission to commit. Control to examine: Keep the send operation behind an explicit scope check. Run the fictional agent example Evidence no longer fits A reviewed candidate changes before the action occurs. Control to examine: Bind the review to the exact text and the permission to the destination; changed text needs a new review. Test a stale review Commitment is hidden One completed step is mistaken for completion of the entire transaction. Control to examine: Name each finality boundary and what remains outside your control. Read the settlement case Procedure substitutes for establishment A completed checklist is mistaken for evidence that the underlying requirements were satisfied. Control to examine: Measure what the process establishes, not simply whether it was adopted. Read the checklist contrast Retry compounds uncertainty A missing acknowledgment is treated as proof that nothing happened. Control to examine: Resolve the first outcome and design durable duplicate prevention before retrying. Explore the fictional retry scenario Comparison of 7 judgment-system records. Each case page shows the full source trail and limits.System · Observed and represented · Commitment boundary · Authority · Read 001 / AVIATION Rejected takeoff at V1 A named decision boundary · Airspeed, runway remaining, aircraft condition, and the operating conditions used in runway analysis. · The manuscript treats V1 as the practical decision boundary for this example; the diagram is a general form and not an operating reference. · The flight crew, acting under applicable training, procedures, and aircraft-specific data. · Open record 002 / FINANCE The Herstatt settlement gap Two obligations, different settlement times · The payment instruction, local settlement status, bank condition, and whether the counter-currency has reached final settlement. · The exposure begins when the sold-currency payment can no longer be canceled unilaterally and ends when the purchased currency is received with finality. · Counterparties, banks, and settlement-system rules distribute control; no one party can unilaterally make the other leg final. · Open record 003 / HEALTHCARE The surgical checklist contrast A pause versus what becomes established · The patient and planned intervention, confirmed aloud by the team immediately before incision. · The pause is placed immediately before the first incision, while the procedure can still be stopped before that irreversible step. · The clinical team under its institutional protocol; the checklist alone does not confer authority. · Open record 004 / TRADE FINANCE When a document discrepancy becomes a question Typed discrepancy and delegated resolution · The presented documents and their consistency with the terms of the credit—not the goods themselves. · Payment is the commitment. Before it, discrepancy notice and waiver preserve a deliberate decision point; after payment, recovery is not the same as reversal. · The issuing bank decides its document-compliance response; the applicant decides whether to waive discrepancies and accept the commercial risk. · Open record 005 / SPACEFLIGHT The number that crossed without its units A declared interface without enforced meaning · Numerical thruster-performance values produced by SM_FORCES. · For the interface-design lesson, the intervention point is before incompatible data enters a navigation calculation. This is an editorial boundary, not a reconstruction of the last physically recoverable mission maneuver. · The engineering interface specification declared the requirement; the investigation board later identified its violation. · Open record 006 / TRADING SYSTEMS Ninety-seven warnings before the market opened Warnings without a binding stop · Which software each server ran, the resulting warning messages, and the orders entering the market. · The boundary of interest is before orders enter the market. An alert is useful only if its resolution is connected to authority to proceed or to stop. · The firm controlled its deployment and market-access systems; the SEC subsequently assessed the missing controls. · Open record 007 / COMPUTER SECURITY Code that carries a checkable argument An untrusted producer and a narrow checker · The submitted code and the proof accompanying it. · Before admitting the untrusted code to execution under the protected consumer’s privileges. · The code consumer chooses the policy; the checker establishes whether the submitted evidence meets it. · Open record A COMPARATIVE QUESTION What does this system actually establish? Aircraft speed, two currencies, a surgical pause, a stack of trade documents: each is a different object with its own evidence and authority. Compare them by asking what the representation can establish, what remains outside its view, and who absorbs the unresolved risk. Open the case register --- # Instruments for agent decision boundaries URL: https://judgmentmachines.com/instruments 04 INSTRUMENTS / FOR USE Turn the idea into a practice. Try a decision rule, map a commitment in your own work, and inspect the methods behind both. BROWSER-LOCAL EXERCISES · NO ACCOUNT REQUIRED JM / EXP-01 WORKED EXAMPLE Draft to send Attempt a send, grant bounded permission, and invalidate a review by changing the message. Inspect a simulated outbox and take away the exact code and tests. Browser-local simulation Run the experiment JM / READ-01 PUBLIC READING ROOM Read The Last Undo The complete introduction with adjustable type, editorial notes, and a manuscript figure. Open the excerpt JM / I-01 INTERACTIVE EXAMPLE Agent Boundary Bench Explore sending, deploying, and retrying. Change the requirements and see when an action should be held, refused, or qualified. Take away A downloadable exercise record and a clearer distinction between capability and authority. Teaching instrument Try the bench JM / I-02 GUIDED WORKSHEET Last Undo Walk List the actions in a workflow. Locate where reversal stops being unilateral, cheap, and complete. Name the requirements and the authority for proceeding. Take away A printable decision record. The worksheet records your assessment; it does not independently verify it. Working prototype Map a workflow JM / I-03 METHODS & SOURCE MATERIAL Protocol Register Explore specifications for evidence, authority, interaction, and coordination. Each declares the question it answers and the limits of its claims. Take away Source documents to examine before adopting or implementing a method. Source register v0.2.0 Open the register THE ARGUMENT BEHIND THE INSTRUMENTS Start with The Last Undo. The book develops the boundary, the evidence that must travel with a decision, and the test of whether the added work is worth doing. Explore the book --- # Agent Boundary Bench: should an AI agent act, wait, or refuse? URL: https://judgmentmachines.com/instruments/agent-boundary INSTRUMENT / AGENT BOUNDARY BENCH What should the agent do next? Choose a situation. Inspect the requirements. See why an action qualifies, waits, or is refused. INTERACTIVE TEACHING EXAMPLE · RUNS IN YOUR BROWSER Proposed action Send the prepared message to the customer list. The boundary: The message leaves the draft workspace and reaches external recipients. Change a property’s state to see how the decision changes. 01 Scope The assignment permits sending, not only drafting. 02 Evidence The claims and recipient list have been checked against their sources. 03 Limits The destination and number of recipients are within the declared scope. 04 Authority The person authorized to release this exact message has approved it. EXAMPLE VERDICT Refuse A declared requirement is violated. The proposed action must change before it can qualify. 1 / 4 requirements marked satisfied Permission to draft is not permission to send. An available tool does not enlarge the assignment. Illustrative states chosen by you. No agent is connected and no action is executed. FROM THE EXAMPLE TO A SYSTEM Put the boundary where the action happens. 01 / Enforcement Control the action path. A rule in a prompt is an instruction. Enforcement requires the tool or service that performs the action to check permission and reject disallowed requests. 02 / Evidence Bind checks to the candidate. Identify exactly what was checked. If the candidate, destination, or authority changes, establish whether the evidence still applies. 03 / Recovery Know what already happened. Record outcomes, handle retries, and retain a way to stop further work. A missing response leaves an outcome unresolved. This bench demonstrates a decision rule. It does not connect to agents, enforce permissions, or establish that a real system is safe. THE LAST UNDO The boundary is the beginning of the argument. The book follows the question through evidence, authority, portable records, and the cost of knowing enough to act. Explore the book Map your own workflow --- # Draft to send: a worked example of agent permission URL: https://judgmentmachines.com/instruments/draft-to-send WORKED EXAMPLE / DRAFT TO SEND A draft is not permission to send. Try to cross the boundary. Then change the authority, review the message, and try again. ISOLATED SIMULATION · NO EMAIL SENT · NO DATA SAVED 01 / THE AGENT’S CANDIDATE Your order is ready for collection. 02 / HOST-OWNED CONTROLS You are the operator here. The agent proposes the message. In this model, only the operator controls the permission and review record. Permit sending to the customer list No message reviewed yet. 03 / ATTEMPT RECORD No attempt yet. Start with the draft-only assignment. SIMULATED OUTBOX / 0 Empty. A refusal or hold adds nothing. Try to break the arrangement. Attempt to send with only drafting permission. The outbox must remain empty. Grant permission without reviewing the text. The request must wait. Review, then change one word. The old review must not cover the new message. Change the recipient group. Permission for one destination must not cover another. Restore the authorized destination, review the exact text, and send twice. Only one copy enters this session’s outbox. Take the example apart. The browser runs the same module offered below. Download both files into one folder and run node --test draft-to-send.test.mjs. Download reference code Download tests Inspect the implementation// SPDX-License-Identifier: MIT · Copyright (c) 2026 Zach Shallbetter · https://judgmentmachines.com/license // Teaching reference: no network, credentials, mail provider, or production authority. // The host owns permission and review state. Never accept them from an agent in production. export const draft = Object.freeze({ body: 'Your order is ready for collection.', destination: 'customer-list' }) export function checkSend(candidate, context) { if (!candidate || typeof candidate.body!== 'string' ||!candidate.body.trim() || candidate.body.length > 2000 || typeof candidate.destination!== 'string') { return { verdict: 'Refuse', reason: 'The candidate is not a supported message.' } } if (!context || typeof context.canSend!== 'boolean' || typeof context.reviewedBody!== 'string' || typeof context.destination!== 'string') { return { verdict: 'Refuse', reason: 'A valid host-owned context is required.' } } if (!context.canSend) return { verdict: 'Refuse', reason: 'The assignment permits drafting, not sending.' } if (candidate.destination!== context.destination) return { verdict: 'Refuse', reason: 'The recipient group is outside the authorized scope.' } if (candidate.body!== context.reviewedBody) return { verdict: 'Hold', reason: 'The exact message has not been reviewed. An earlier review does not travel to changed text.' } return { verdict: 'Allow in sandbox', reason: 'The exact message and destination match the host-owned permission and review.' } } export function createSandbox() { // Memory only. This is deliberately not a durable delivery/idempotency mechanism. const outbox = [] return { attempt(candidate, context) { const decision = checkSend(candidate, context) if (decision.verdict!== 'Allow in sandbox') return decision if (outbox.some(item => item.body === candidate.body && item.destination === candidate.destination)) { return { verdict: 'Already simulated', reason: 'This exact message is already in this session’s simulated outbox.' } } outbox.push(Object.freeze({ body: candidate.body, destination: candidate.destination })) return decision }, messages() { return outbox.map(item => ({...item })) }, } } This is a teaching model, not an email gateway. Production requires authenticated, server-owned authority; evidence bound to the exact action; durable retry handling; revocation; and control of every delivery path. Browser controls can be changed by the person running the browser. The outbox is memory only and resets on refresh. READ / APPLY Capability is only half the story. The book asks what must be established before the action becomes consequential. Read the introduction Map your own boundary --- # Last Undo Walk: find where correction becomes remediation URL: https://judgmentmachines.com/instruments/last-undo-walk I INSTRUMENTS / DIGITAL Last Undo Walk A guided worksheet for finding where correction becomes remediation. FIELD PROTOTYPE · RUN LOCALLY IN YOUR BROWSER INSTRUMENT 01 / FIELD PROTOTYPE Walk the boundary. List external actions in order. For each one, ask whether reversal is still unilateral, cheap, and complete. The first “no” suggests where the boundary may be; an unknown condition means the location is still unresolved. REVERSIBLE THE SEAM CONSEQUENCE 01 Name the workflow 02 Test each action’s undoUse “Unknown” when the answer has not been established. Do not treat an unanswered condition as “Yes.” Action / external change Unilaterally Cheaply Completely Remove 1 Unilaterally reversible? Cheaply reversible? Completely reversible? SEQUENCE READOUT · NOT A VERDICT Name every action before interpreting the sequence. 03 Establish before crossingState what must be true for this commitment. Record both the condition and the evidence that bears on it. 01 1 1 1 02 2 2 2 03 3 3 3 04 Record authority and dispositionWho has authority if something remains unresolved? What remains unresolved? Disposition chosen by the accountable person Decision-maker / role Record date (entered by you) Runs in this browser only. Refreshing clears the draft. --- # Protocol register: evidence, authority, and interaction specifications URL: https://judgmentmachines.com/methods 05 REGISTER / PROTOCOLS & RESEARCH Declared scope. Declared limits. The register preserves each item's identifier, stated maturity, role, and declared relationships. A protocol is not evidence that its implementation works. REGISTER v0.2.0 · 16 ENTRIES · 9 DECLARED RELATIONSHIPS Search the register Discipline 16 entries shown Interaction & evidence FCI Foundations of Computational Interaction How is interaction formally described, measured, and predicted? v0.5.0 documented and reference substrate ↗ ICP Interaction Contract Protocol What must a consequential interaction declare before it runs? v0.1.0 documented ↗ EQP Evidence Qualification Protocol Given this candidate, claim, context, and evidence, what is justified? v0.1.0 documented and formalized ↗ Meaning & publication SES Semantic Expression System How can meaning stay fixed while expression changes completely? v1.0.0 implemented and tested ↗ SSS Semantic Surface System Given this state and situation, what surface is warranted here? v0.1.0 documented ↗ SPS Semantic Publication Substrate How is published truth separated from its presentation? v0.1.0 working prototype ↗ Work & external state IEPE Intent and Evidence Project Engine How does intent become authorized work, and what blocks promotion? v0.2.0 implemented and adopted ↗ ESP External State Protocol How can an actor pursue an outcome they cannot establish alone? v0.3.0 definition ↗ RCP Recipient Contract Protocol Who is owed a deliverable, and by when? v0.1.0 draft ↗ EDP Engagement Definition Protocol What has an engagement received, and what has it given? v0.1.0 draft ↗ Research & writing PDP Polymorphic Decision Protocol How is a high-uncertainty decision made with a panel of systems? v1.3.0 reference design and recorded case ↗ HCML Human-Computer Mutual Legibility Can human intent and machine state be made mutually inspectable? v2.0.0 research corpus ↗ APP Authored Prose Protocol What must authored prose do, and how is it checked? v0.1.0 draft and self checked ↗ NPS Priming Toolkit What makes an agent primer measurably effective? v1.0.0 specification and measured corpus ↗ NEM Narrative Evidence Method In what order does a qualified claim reach the reader? v0.1.0 working draft ↗ CONCEPTS Protocol Concepts Which early concepts helped explore the stack’s boundaries? v0.1.0 reference material ↗ DECLARED COMPOSITION Relationships are recorded, not inferred. 01 FCI DEFINES ICP 02 ICP QUALIFIES EQP 03 IEPE QUALIFIES EQP 04 SES CONSTRAINS SSS 05 SSS PROJECTS SPS 06 ESP IMPLEMENTS IEPE 07 ESP IMPORTS EQP 08 RCP CONSTRAINS ESP 09 APP CONSTRAINS NEM --- # Laboratory: open questions and failure conditions URL: https://judgmentmachines.com/laboratory 06 LABORATORY / OPEN QUESTIONS What would change our minds? The research program keeps counterexamples and failure conditions in view. A useful framework must be capable of losing. QUESTIONS DRAWN FROM THE LAST UNDO · NOT EXPERIMENTAL RESULTS Q-01 Can useful commitment boundaries be identified before the consequential action? The manuscript names this as a possible failure of the argument. Open · manuscript falsifier Q-02 Can required-property lists remain stable enough to compare across measurement windows? A comparison requires the commitment and its contract to stay explicit. Open · measurement condition Q-03 Does measured improvement justify the cost of establishing the boundary? The manuscript separates the idea’s validity from whether anyone benefits enough to pay or adopt it. Open · practical hypothesis EVIDENCE DISCIPLINE A question is not a result. These are explicit failure conditions in the current manuscript. No experiment or external validation is claimed by listing them here. New study records should show hypothesis, method, materials, observations, result, limitations, and state before conclusions are promoted. See the manuscript's account of what would falsify it --- # The Last Undo Review: examine one consequential workflow URL: https://judgmentmachines.com/practice 07 PRACTICE / APPLIED WORK The Last Undo Review. A bounded examination of one consequential workflow: where commitment occurs, what must be established there, and who may decide what remains unresolved. PROPOSED ENGAGEMENT SCOPE · COMMERCIAL TERMS NOT YET PUBLISHED BRING ONE WORKFLOW Make the point of consequence inspectable. The review starts with a specific action that changes something outside the team: a release, payment, deployment, publication, credential change, or other commitment. The method is not a general audit, certification, or guarantee of correctness. It asks whether the system can identify its boundary and make its requirements, evidence, authority, and unresolved conditions visible there. Discuss a workflow REVIEW SCOPE 01 Map the actions that affect external state. 02 Locate where reversal stops being unilateral, cheap, and complete. 03 List the properties that must be established before crossing. 04 Identify authority and escalation for unresolved conditions. 05 Specify what record should travel with the decision. 06 Define a measurement plan and the conditions that could falsify the intervention. EXPECTED DELIVERABLES A working record, not a seal. 01 Boundary map The sequence of consequential actions and the selected commitment points. 02 Establishment contract Required properties, acceptable evidence, and declared limits. 03 Authority map Who may decide, what can be automated, and what must escalate. 04 Receipt and measurement What is recorded at decision time and how any proposed change would be evaluated. No result is implied by the offer. The review does not certify safety, correctness, compliance, or performance. Fees, timeline, and engagement terms are not specified here. Try the local boundary-walk prototype --- # About Judgment Machines and Zach Shallbetter URL: https://judgmentmachines.com/about 08 ABOUT / INSTITUTION Research, publishing, building. An independent project directed by Zach Shallbetter in Spokane, Washington. EVERY CLAIM ON THIS PAGE CARRIES ITS STATUS · AUTHOR-STATED CLAIMS ARE MARKED The work is not to eliminate uncertainty or remove judgment. It is to make inputs, authority, and limits inspectable—and to preserve what remains unresolved. Judgment Machines publishes the books, builds the instruments, and keeps the records this work requires. It is one person’s project, so this page describes the work behind it: what each thing had to promise, what is not claimed, and how a stranger can check the rest. PRINCIPLES Keep the distinctions intact. Evidence is evidence for a specified claim, under conditions, with limits. A criterion states what would count; it does not establish that the criterion was satisfied. Authority may permit action under unresolved conditions; it does not convert the unresolved into fact. Documentation alone is not answerability. A record must show why the judgment was warranted. WHERE THE ARGUMENT COMES FROM A job file that could describe the wrong cut. From 2018 to 2022 he ran a just-in-time manufacturing line for a hardware company he founded: site, commerce, inventory intake, production, and fulfillment, running largely without a supervising hand. At volume, nobody can read every path. The line had to promise that a job file could not describe a cut the machine should not make. That problem, capability outrunning permission at the moment of commitment, is the argument of The Last Undo and the reason this institution exists. It took four years and a failed company to learn. Read about the book DIRECTOR Zach Shallbetter Spokane, Washington. Engraved portrait He founded a hardware company and ran its manufacturing line for four years, and co-founded a product studio and ran it for six. His open-source contributions are merged in Rust’s official linter and in Google Labs repositories. What follows is the work, not a description of the person who did it. Each entry states what the thing had to promise, and what it cost to make that promise hold. zachshallbetter.com ↗ THE WORK Six things, and what each had to promise. 2018 Object A game board with a nano-ceramic coating Magnetic, machined, and coated with a process developed in-house because nothing off the shelf survived what players did to it. Funded on Kickstarter at roughly 330 percent of a $30,000 goal, then a seed round of about $500,000. Had to promise. That the object arriving in a stranger’s house was the object described on a page they read months earlier. The record deliberately excludes a shipped-units figure. An earlier claim of 25,000 boards is not in the records, and it stays out until a record exists. Shipped · company closed 2022 2018–22 System The line that made them A just-in-time manufacturing operation covering site, commerce, inventory intake, production, and fulfillment, running largely without a supervising hand. Machines re-engineered in hardware and software, and custom equipment commissioned around the workflow rather than the reverse. Had to promise. That a job file could not describe a cut the machine should not make. At volume, nobody can read every path. This is where the argument of The Last Undo comes from. Dismantled with the company 2014–20 Company Six years of payroll A product and design studio he co-founded and ran, directing a team of ten and more across design and engineering for clients from startups to the Fortune 500. He owned proposals, scope, delivery cadence, and the mentorship of junior developers. Had to promise. That people who left other jobs to work there got paid, on a schedule, for six years. Built with partners, not alone. Three of the four organizations here were. Wound down 2023 Platform A marketplace for Indigenous art Built on Next.js, Supabase, and Stripe Connect, with seller verification, customizable storefronts, product customization, and payments. Recommendation, search, and conversational assistance built directly, early in the applied-AI product cycle. Sole delivering engineer. Had to promise. That a payout reached the right artist for the right work, every time, without anyone checking by hand. Sole committer throughout. That is contribution evidence, not third-party review. Delivered 2026 Contribution Eleven lines in a compiler’s linter The with_capacity_zero lint, written against a maintainer-filed issue, revised through review, and merged into Rust’s official linter. It catches a small waste across eight standard collection types using const evaluation, so it fires on a computed zero and stays quiet on a runtime one. Had to promise. That when a compiler tells a developer something about their code, the statement is true. A false positive in a linter is worse than no linter. Small in substance. It is here because a stranger can check it in thirty seconds. Merged · checkable by anyone 2025–26 Specifications Twelve protocols nobody has reviewed Interaction contracts, evidence qualification, decision-making with panels of AI systems, the separation of meaning from expression. Written independently over about eighteen months, each declaring a question it answers and a boundary it refuses to cross. They compose, which was not planned. Had to promise. That a reader could tell what each one does not claim without reading between any lines. Not one has been read by anyone outside him. Until that changes they are specifications, not standards. Unreviewed WHAT IS NOT CLAIMED Absences, stated. A degree. He does not have one. An exit. Four organizations built, none sold. A measured delta on production volume. The book asks builders to instrument both windows and publish the gap, and he has not published one. The chapters that depend on it are held until he does. Two engagements and a filed provisional, withheld while they are unresolved. Their absence is deliberate and is not a gap. CHECK THE CLAIMS Every claim, and how to check it. Externally checkable: a stranger can verify it without asking him. Author-stated: he is the source, and says so. Claim · How to check it · Status The lint · rust-lang/rust-clippy PR #17192, merged June 2026, approved by a Clippy team member · Externally checkable Google Labs contributions · Two merged pull requests, design.md and stitch-skills, 2026 · Externally checkable The marketplace · Repository history; the fine-tuning toolchain is public · Externally checkable Video AI tooling, 2025 · Roughly 32 merged pull requests, July to November · Externally checkable Open-source contributions since 2013 · Merged pull requests, earliest February 2013 · Externally checkable Twelve protocols, specified and versioned · The specifications themselves, in the protocol register · Externally checkable · unreviewed Nonprofit; broadband to a unanimous vote · Registration and the public council record · Externally checkable Kickstarter at roughly 330 percent · Campaign record · Author-stated Roughly $500,000 in seed capital · Company records · Author-stated Units shipped at production scale · No supported count; an earlier figure is deliberately excluded from the record · Author-stated The line ran largely unsupervised · No external record · Author-stated The studio, six years, team of ten and more · Registration and public client work · Author-stated Author-stated claims cluster in one place: the two organizations that no longer operate. They are the oldest claims and the hardest to check, so they are marked rather than blended in. Claims under legal restriction are omitted rather than softened. STANDING & LIMITS What this institution can claim. The manuscript documents work in manufacturing systems and a lint incorporated into Rust’s official linter. The production measurement demanded by the book has not yet been made. No protocol or manuscript on this site has had external review. SOURCES · THE LAST UNDO, INTRODUCTION v0.8.0 · ABOUT THE AUTHOR v3.1.0 (DRAFT) Read source and maturity notes CORRESPONDENCE Write to the director. hi@zachshallbetter.com · zachshallbetter.com --- # Site index URL: https://judgmentmachines.com/site-index INDEX / OBJECT REGISTER Find the work by object. The public surface of a working institution. Records carry their state and source limits with them. 01 / THE LAST UNDO The Last Undo The book, its argument, and publication details. ↗ 02 / READING ROOM Reading room The complete introduction to The Last Undo. ↗ 03 / AGENT BOUNDARY BENCH Agent Boundary Bench Explore permission, evidence, limits, and authority before an agent acts. ↗ 04 / DRAFT TO SEND Draft to send A worked example: try to send with only permission to draft. ↗ 05 / LAST UNDO WALK Last Undo Walk Map the boundary in a workflow of your own. ↗ 06 / SERIES Series Three connected inquiries; one current manuscript. ↗ 07 / CASES Cases Source-grounded system reconstructions. ↗ 08 / ATLAS Atlas Structured records of judgment systems. ↗ 09 / INSTRUMENTS Instruments Interactive examples, a printable worksheet, and the source register. ↗ 10 / PROTOCOL REGISTER Protocol register Protocols and research with declared scope. ↗ 11 / LABORATORY Laboratory Open questions and failure conditions. ↗ 12 / PRACTICE Practice The Last Undo Review. ↗ 13 / ABOUT About Institution, principles, and standing. ↗ 14 / SOURCES Sources Registry files, provenance, and scope limits. ↗ 15 / STORE Store Reserve the book, join, or commission a review. ↗ 16 / MEMBERSHIP Membership Support the research; the ebook and draft chapters. ↗ 17 / TERMS OF SALE Terms of sale Reservations, membership, refunds. ↗ 18 / PRIVACY Privacy What the site collects. ↗ 19 / LICENSE License What you may quote, cite, and reuse. ↗ --- # Sources and limits URL: https://judgmentmachines.com/sources SOURCES / LIMITS Follow a claim to its source. Inspect the evidence behind the cases and methods, including what remains unresolved. PROTOCOL REGISTER v0.2.0 · REVISION f1892e4 PROTOCOL REGISTER Protocol Index The human-readable index describes scope, relationships, maturity, and explicit non-claims for the protocols and research programs. Read the Markdown index MACHINE-READABLE RECORD Registry JSON Sixteen entries and nine declared relationships, as recorded in registry version 0.2.0. Registry presence is not a maturity claim. Open the JSON registry BOOTSTRAP SUBSTRATE Formal Project Bootstrap The source overview describes the operating substrate that makes intent, authority, work, evidence, context, coordination, and recovery explicit. Read the overview MANUSCRIPT SOURCES Draft evidence trail Use each case record to follow its source version, citation, and denominator. Open the case register LIMITS What this site does not establish. External review of the protocols or manuscript. Empirical superiority of the framework or protocols. Safety certification, legal compliance, or correctness of an implementation. That planned books, instruments, or engagements are available for sale. That the cited aviation statistic applies to all rejected takeoffs. Instrument engravings are illustrative artwork. They are not historical specimens or evidence for a case. The Agent Boundary Bench uses fictional situations and reader-selected states; its exported record is an exercise, not an authorization. --- # License and terms of use URL: https://judgmentmachines.com/license LICENSE / TERMS OF USE What you may do with this work. Quote it, cite it, link it, and summarize it. Credit the author and link the page. The code examples are yours to use under the MIT license. WRITING · BOOK EXCERPT · PROTOCOL SOURCES © 2026 Zach Shallbetter. All rights reserved. This covers the site’s text, the introduction to The Last Undo in the reading room, the case records, and the protocol documents under /sources. You are welcome to quote, summarize, discuss, and cite any of it, in print, online, in teaching, or in an AI system’s answer, provided you credit Zach Shallbetter (or Judgment Machines) and link the page you drew from. Republishing whole pages, or the excerpt, needs permission. IMAGES Engravings, portrait, and cover The instrument engravings are illustrative artwork, not historical specimens. They, the engraved portrait, and the book cover are © Zach Shallbetter, all rights reserved. Social-preview images may be displayed wherever a link to this site is shared. CODE EXAMPLES MIT license The worked-example code offered for download under /examples, including draft-to-send.mjs and its tests, is released under the MIT license. Use it, change it, and ship it; keep the copyright notice. Read the MIT license text AUTOMATED ACCESS Crawlers and AI systems Search engines and AI crawlers, including model-training crawlers, may read this site as robots.txt allows. An agent-facing index is at /llms.txt and the full text at /llms-full.txt. When an answer draws on this site, please attribute it and link the canonical page. PERMISSIONS Ask for anything else. For republication, translation, or other uses, write to hi@zachshallbetter.com. --- # Store: reserve The Last Undo, join, or commission a review URL: https://judgmentmachines.com/store 09 STORE / SUPPORT THE WORK Reserve, join, commission. Reserve the book, support the research as a member, or bring one consequential workflow to a review. The instruments and the reading room stay free. CHECKOUT BY STRIPE · RESERVATIONS CHARGED ONLY AT PUBLICATION I / THE BOOK The Last Undo What Must Be Established Before Consequence. Reserve the ebook or the hardcover; nothing is charged until publication. Nothing is charged today. Your card is saved by Stripe and charged only when the book is published. No publication date is set yet. We will write to you before your card is charged. Cancel any time before then by writing to hi@zachshallbetter.com. If the book is never published, you are never charged. Ebook $14.99 Delivered by email at publication. Hardcover $29 + $6 US shipping First printing, US shipping. Includes the ebook. Read the introduction first II / MEMBERSHIP Support the research. Members fund independent work on judgment before consequence. Membership includes the ebook at publication and draft chapters as they are released. Monthly $6 / month Cancel any time. Annual $60 / year Two months free. What membership includes III / APPLIED WORK The Last Undo Review A bounded examination of one consequential workflow. Engagement terms are not yet published, so reviews are arranged by conversation rather than checkout. See the review scope FREE, ALWAYS What you never pay for. The instruments, the case archive, the protocol register, and the full introduction in the reading room are free and need no account. Membership does not unlock them; it pays for more of them. PRICES IN USD · TAX ADDED WHERE APPLICABLE · TERMS OF SALE · PRIVACY --- # Membership: support Judgment Machines URL: https://judgmentmachines.com/membership 10 MEMBERSHIP / JUDGMENT MACHINES Fund the work. Read it first. Membership pays for independent research on judgment before consequence. It is support, not a paywall: everything free on this site stays free. MONTHLY OR ANNUAL · CANCEL ANY TIME Zach Shallbetter, director WHAT MEMBERS RECEIVE The Last Undo ebook at publication, if you are a member then. Draft chapters as they are released to members. No schedule is promised; the manuscript moves at the pace of its evidence. The work itself: cases, instruments, and protocols that stay free for everyone because members pay for them. WHAT MEMBERSHIP IS NOT Not a Pro tier, a course, or a certification. Nothing on the site is locked behind it. Monthly $6 / month Cancel any time; access runs to the end of the month you paid for. Annual $60 / year Two months free. Cancel any time; access runs to the end of the year. BILLED BY STRIPE · TAX ADDED WHERE APPLICABLE · TERMS · PRIVACY --- # Terms of sale URL: https://judgmentmachines.com/terms TERMS / SALES Terms of sale. What you are agreeing to when you reserve the book or become a member, in plain language. EFFECTIVE 2026-10-20 WHO YOU ARE BUYING FROM Judgment Machines Judgment Machines is directed by Zach Shallbetter in Spokane, Washington. Payments are processed by Stripe. Questions, cancellations, and refund requests go to hi@zachshallbetter.com. RESERVATIONS The book is charged at publication, not before. Reserving the ebook ($14.99) or hardcover ($29 plus $6 US shipping) saves your card with Stripe and charges nothing. When the book is published we write to you, then charge the saved card the reserved price plus any applicable tax. No publication date is set. You may cancel at any time before the charge by writing to us; if the book is not published, you are never charged. If a charge fails or needs authentication, Stripe emails you a link to pay; the reservation lapses if it is not paid within 30 days. DELIVERY Ebook by email; hardcover by post. The ebook is delivered as a download link to the email you used at checkout. The hardcover ships to the US address you gave, after publication, with the ebook included. Formats and shipping times are confirmed in the publication email. MEMBERSHIP Billed in advance; cancel any time. Membership is $6 / month or $60 / year, billed in advance and renewed automatically until you cancel. Cancel any time from the member portal; access continues to the end of the period you paid for, and no partial refunds are made for unused time except where the law requires. Members at publication receive the ebook of The Last Undo. Draft chapters are released to members at the author’s discretion, without a schedule. We will give 30 days’ notice of any price change. REFUNDS If something is wrong, we put it right. Write within 30 days of a charge and we will fix the problem or refund it. Hardcovers damaged in transit are replaced. Nothing here limits rights you have under the law where you live. TAX AND CURRENCY Prices are in US dollars. Sales tax or VAT is added where it applies and shown before you confirm. CHANGES These terms may change. The terms in force when you reserve or join apply to that order. Material changes for members are emailed in advance. --- # Privacy URL: https://judgmentmachines.com/privacy PRIVACY / DATA What this site collects. Usage analytics, the details Stripe needs to process an order, and what you send us. Nothing is sold. EFFECTIVE 2026-10-20 ANALYTICS Google Analytics 4 Pages you view, how far you scroll, links you follow out, files you open, the kind of site that referred you (search, social, an AI assistant), and checkout steps. Google Analytics sets first-party cookies to recognise a returning browser. No names, emails, or payment details are sent to it. ORDERS Stripe Checkout runs on Stripe. Stripe collects your card details; this site never sees or stores them. We receive your name, email, billing and (for hardcovers) shipping address, and what you ordered, to fulfil and support your order. Stripe’s privacy policy applies to the data it processes. CORRESPONDENCE Email If you write to us, we keep the correspondence to answer you and to honour any request it contains. HOSTING Vercel The site is served by Vercel, which keeps standard request logs for security and reliability. YOUR CHOICES Ask, and we act. Write to hi@zachshallbetter.com to see, correct, or delete what we hold about you, or to cancel a reservation or membership. Blocking cookies does not stop the site from working. --- # Rejected takeoff at V1 (Aviation case) URL: https://judgmentmachines.com/cases/v1-decision-boundary 001 CASE / Aviation Rejected takeoff at V1 A speed turns the reject-or-continue judgment into a commitment with a shrinking physical margin. SOURCE STATE · The Last Undo, Chapter 5 draft v0.5.1, note 1; FAA and industry, Pilot Guide to Takeoff Safety, Section 2 of the Takeoff Safety Training Aid, original 1993, data through 2003. FIG. 5.1 · DECISION GEOMETRY · NOT TO SCALE This record reconstructs the argument in its cited manuscript draft. It is not a certified procedure or an independent audit. THE JUDGMENT MACHINE Trace the chain. ObjectA transport-category aircraft on the takeoff roll. ObservationAirspeed, runway remaining, aircraft condition, and the operating conditions used in runway analysis. DistinctionBefore the decision speed, rejecting remains an available action; above it, the stopping margin has been spent. RepresentationV1, derived for the aircraft and conditions; the draft distinguishes the operational rule from the regulation’s literal text. CriterionThe operational definition in the cited guide ties V1 to the latest speed at which a reject can begin and the aircraft stop within the runway available. JudgmentReject or continue under trained operating doctrine. The speed is not a substitute for the crew’s judgment or procedures. AuthorityThe flight crew, acting under applicable training, procedures, and aircraft-specific data. Last undo / seamThe manuscript treats V1 as the practical decision boundary for this example; the diagram is a general form and not an operating reference. ConsequenceA late reject may overrun the runway; a continue decision has its own conditions and risks. EVIDENCE & LIMITS Keep the denominator attached. The manuscript cites 97 overrun accidents and serious incidents; 55% involved rejects initiated above V1. It explicitly limits that denominator to the accident and incident set, not all rejected takeoffs. The 55% figure does not measure all rejected takeoffs or establish that a callout was made correctly. The drawing is not to scale and must not be used for flight planning. SOURCE TRAIL · The Last Undo, Chapter 5, draft v0.5.1, source note 1. Read the related manuscript record PREVIOUS / COMPUTER SECURITY Code that carries a checkable argument NEXT / FINANCE The Herstatt settlement gap --- # The Herstatt settlement gap (Finance case) URL: https://judgmentmachines.com/cases/herstatt-settlement-risk 002 CASE / Finance The Herstatt settlement gap One currency had been paid; the other was expected later, across a boundary no single party controlled. SOURCE STATE · The Last Undo, Chapter 5 draft v0.5.1, notes 4–6; BIS, Settlement Risk in Foreign Exchange Transactions (Allsopp report), March 1996; Galati, BIS Quarterly Review, December 2002. 01 / GERMANY / MORNING Deutsche Marks paid irrevocably 02 / TIME-ZONE GAP Herstatt closed after German settlement ended 03 / NEW YORK / LATER Expected dollars did not arrive This record reconstructs the argument in its cited manuscript draft. It is not a certified procedure or an independent audit. THE JUDGMENT MACHINE Trace the chain. ObjectA foreign-exchange transaction exchanging Deutsche Marks for U.S. dollars across time zones. ObservationThe payment instruction, local settlement status, bank condition, and whether the counter-currency has reached final settlement. DistinctionThe sold currency may be irrevocably paid while the purchased currency remains outstanding. RepresentationA pair of linked obligations with separate payment systems and operating hours. CriterionSettlement finality is reached only when the purchased currency is received with finality; a sent instruction is not the same as settlement. JudgmentDo not treat one leg’s completion as proof that the exchange is complete. AuthorityCounterparties, banks, and settlement-system rules distribute control; no one party can unilaterally make the other leg final. Last undo / seamThe exposure begins when the sold-currency payment can no longer be canceled unilaterally and ends when the purchased currency is received with finality. ConsequenceOn 26 June 1974, Herstatt’s closure left counterparties that had paid Deutsche Marks in the morning without the expected dollars in New York. EVIDENCE & LIMITS Keep the denominator attached. The manuscript reports a BIS estimate of approximately US$200 million in stranded same-day payments and a sharp short-term drop in New York gross transfers. The manuscript says primary sources do not state a total creditor loss for Herstatt; this page does not supply one. The case concerns settlement exposure, not a complete accounting of bank failure. SOURCE TRAIL · The Last Undo, Chapter 5, draft v0.5.1, notes 4–6. Read the related manuscript record PREVIOUS / AVIATION Rejected takeoff at V1 NEXT / HEALTHCARE The surgical checklist contrast --- # The surgical checklist contrast (Healthcare case) URL: https://judgmentmachines.com/cases/surgical-checklist-contrast 003 CASE / Healthcare The surgical checklist contrast A well-known before–after study reported improved outcomes; a later province-wide mandate found no statistically significant reduction. SOURCE STATE · The Last Undo, Chapter 5 draft v0.5.1, notes 8, 9, and 12; Haynes et al., NEJM (2009); Urbach et al., NEJM (2014). Eight-hospital study 8 hospitals · 8 cities Mortality 1.5% → 0.8% Complications 11.0% → 7.0% Ontario mandate 101 hospitals · province-wide Mortality 0.71% → 0.65% · not significant Complications 3.86% → 3.82% · not significant This record reconstructs the argument in its cited manuscript draft. It is not a certified procedure or an independent audit. THE JUDGMENT MACHINE Trace the chain. ObjectA surgical procedure with an operating team, patient, site, side, planned procedure, and equipment. ObservationThe patient and planned intervention, confirmed aloud by the team immediately before incision. DistinctionA required pause is not the same thing as a completed check, shared understanding, or satisfied condition. RepresentationThe checklist and the team’s verbal confirmation at the time out. CriterionThe Universal Protocol and WHO checklist specify the team and the items to confirm before the procedure begins. JudgmentThe team resolves mismatches before proceeding; the checklist does not independently certify the underlying facts. AuthorityThe clinical team under its institutional protocol; the checklist alone does not confer authority. Last undo / seamThe pause is placed immediately before the first incision, while the procedure can still be stopped before that irreversible step. ConsequenceThe two studies reported different detectable outcome changes after checklist introduction. EVIDENCE & LIMITS Keep the denominator attached. The eight-hospital study compared 3,733 procedures before with 3,955 after; mortality was 1.5% versus 0.8%, and complications 11.0% versus 7.0%. The Ontario study compared 109,341 procedures before with 106,370 after across 101 hospitals; adjusted mortality was 0.71% versus 0.65% (p=.13), and complications 3.86% versus 3.82% (p=.29). Both studies have design and context limits. The Ontario study did not establish why outcomes did not change significantly; the manuscript notes its lower baseline, short windows, and self-reported compliance. Adoption alone is not evidence that the pause established the required properties. SOURCE TRAIL · The Last Undo, Chapter 5, draft v0.5.1, notes 8, 9, and 12. Read the related manuscript record PREVIOUS / FINANCE The Herstatt settlement gap NEXT / TRADE FINANCE When a document discrepancy becomes a question --- # When a document discrepancy becomes a question (Trade finance case) URL: https://judgmentmachines.com/cases/letter-of-credit-waiver 004 CASE / Trade finance When a document discrepancy becomes a question The bank checks the documents it can see, lists mismatches, and routes a waiver decision to the party bearing the commercial risk. SOURCE STATE · The Last Undo, Chapter 6 draft v0.3.0, notes 1–5 and 10; Mann, “The Role of Letters of Credit in Payment Transactions,” Michigan Law Review 98 (2000). 01 PRESENT DOCUMENTS 02 CHECK AGAINST CREDIT 03 LIST DISCREPANCIES 04 APPLICANT WAIVER OR REFUSAL 05 PAY / DO NOT PAY This record reconstructs the argument in its cited manuscript draft. It is not a certified procedure or an independent audit. THE JUDGMENT MACHINE Trace the chain. ObjectA documentary credit with a presentation of invoice, bill of lading, and other required documents. ObservationThe presented documents and their consistency with the terms of the credit—not the goods themselves. DistinctionA document can comply on its face while the underlying goods or transaction remain wrong or unresolved. RepresentationA named document set checked against the credit’s required terms and the UCP 600 rules. CriterionThe bank determines compliance of the presentation; if discrepant, it must enumerate discrepancies in a notice and may seek the applicant’s waiver. JudgmentPay against a complying presentation, refuse with a typed notice, or escalate discrepancies to the applicant for a waiver decision. AuthorityThe issuing bank decides its document-compliance response; the applicant decides whether to waive discrepancies and accept the commercial risk. Last undo / seamPayment is the commitment. Before it, discrepancy notice and waiver preserve a deliberate decision point; after payment, recovery is not the same as reversal. ConsequenceIn Mann’s sample of 500 files at five U.S.-located banks, only 27% of first presentations conformed; despite frequent discrepancies, applicants did not refuse payment in the sample. EVIDENCE & LIMITS Keep the denominator attached. Among discrepant import files, 53% were waived within one business day and 84% within a week. Bankers interviewed put ultimate refusal below 1% of discrepant files; 29% of discrepant files signaled substantive defaults. Five banks are not a statistically representative sample. The manuscript explicitly distinguishes a discrepant first presentation from a refused payment, and notes that the bank authenticates the channel/documents, not the truth of their contents or the goods. SOURCE TRAIL · The Last Undo, Chapter 6, draft v0.3.0, source notes 1–5 and 10. Read the related manuscript record PREVIOUS / HEALTHCARE The surgical checklist contrast NEXT / SPACEFLIGHT The number that crossed without its units --- # The number that crossed without its units (Spaceflight case) URL: https://judgmentmachines.com/cases/mars-climate-orbiter-units 005 CASE / Spaceflight The number that crossed without its units A navigation calculation consumed impulse data under a different unit convention from the system that produced it. SOURCE STATE · The Last Undo, Chapter 2, “A number with no units,” note 1; NASA, Mars Climate Orbiter Mishap Investigation Board Phase I Report, 10 November 1999, section 4. 01 PRODUCE IMPULSE 02 POUND-FORCE-SECONDS 03 INTERFACE CROSSING 04 ASSUME NEWTON-SECONDS 05 NAVIGATION ERROR This record reconstructs the argument in its cited manuscript draft. It is not a certified procedure or an independent audit. THE JUDGMENT MACHINE Trace the chain. ObjectThruster impulse data passed from ground software into the spacecraft navigation process. ObservationNumerical thruster-performance values produced by SM_FORCES. DistinctionA plausible numerical value is not an established physical quantity unless its unit convention is known. RepresentationThe interface specification required newton-seconds; the producer supplied pound-force-seconds. CriterionThe receiving calculation required impulse expressed in the specified metric units. JudgmentThe chapter argues for establishing the unit contract at the interface rather than assuming it from a plausible number. AuthorityThe engineering interface specification declared the requirement; the investigation board later identified its violation. Last undo / seamFor the interface-design lesson, the intervention point is before incompatible data enters a navigation calculation. This is an editorial boundary, not a reconstruction of the last physically recoverable mission maneuver. ConsequenceThe navigation error contributed to the loss of Mars Climate Orbiter on arrival in 1999. EVIDENCE & LIMITS Keep the denominator attached. The cited board report identifies failure to use metric units in the ground-software file as the root cause and records a factor-of-4.45 mismatch. The board also identified contributing organizational and verification failures. A unit check would address one named failure, not certify the mission. The combined orbiter-and-lander program cost is not the loss attributable to this error. SOURCE TRAIL · The Last Undo, Chapter 2, source note 1; NASA Phase I report, section 4. Site reconstruction of the manuscript draft. Read the related manuscript record PREVIOUS / TRADE FINANCE When a document discrepancy becomes a question NEXT / TRADING SYSTEMS Ninety-seven warnings before the market opened --- # Ninety-seven warnings before the market opened (Trading systems case) URL: https://judgmentmachines.com/cases/knight-capital-release 006 CASE / Trading systems Ninety-seven warnings before the market opened An incomplete deployment left retired behavior reachable. Warning messages existed before trading began, but did not stop the orders. SOURCE STATE · The Last Undo, Chapter 1 draft v0.7.2, Knight Capital account, source note 11; SEC administrative order, Release 34-70694, 16 October 2013, paragraphs 13–19. 01 / DEPLOYMENT One of eight servers misses the release 02 / BEFORE OPEN 97 automated messages do not produce an effective stop 03 / MARKET OPEN Retired code generates erroneous orders 04 / AFTERMATH Loss and regulatory findings This record reconstructs the argument in its cited manuscript draft. It is not a certified procedure or an independent audit. THE JUDGMENT MACHINE Trace the chain. ObjectA trading-system deployment across eight servers on 1 August 2012. ObservationWhich software each server ran, the resulting warning messages, and the orders entering the market. DistinctionA warning delivered to an inbox is not a control that blocks a consequential action. RepresentationDeployment state and automated messages naming the retired function. CriterionThe intended release must be present and market-access controls must constrain erroneous orders before they enter the market. JudgmentUnresolved deployment faults required investigation and effective containment, rather than an assumption that deployment had completed uniformly. AuthorityThe firm controlled its deployment and market-access systems; the SEC subsequently assessed the missing controls. Last undo / seamThe boundary of interest is before orders enter the market. An alert is useful only if its resolution is connected to authority to proceed or to stop. ConsequenceThe manuscript reports more than $460 million lost during approximately forty-five minutes of erroneous trading. EVIDENCE & LIMITS Keep the denominator attached. The cited account describes one server missing the new deployment, 97 pre-open messages, and 212 incoming orders producing roughly four million executions across 154 stocks. This was not an AI-agent incident. It illustrates a control failure applicable to automated action, not evidence that all agent incidents share its cause. The SEC identified several missing controls; the loss must not be reduced to one technician or one alert. SOURCE TRAIL · The Last Undo, Chapter 1, source note 11; SEC Release 34-70694. Site reconstruction of the manuscript draft. Read the related manuscript record PREVIOUS / SPACEFLIGHT The number that crossed without its units NEXT / COMPUTER SECURITY Code that carries a checkable argument --- # Code that carries a checkable argument (Computer security case) URL: https://judgmentmachines.com/cases/proof-carrying-code 007 CASE / Computer security Code that carries a checkable argument A consumer can check evidence that untrusted code meets a named safety policy without trusting its author. SOURCE STATE · The Last Undo, Chapter 4 draft v0.5.0, opening sections and note 1; G. C. Necula and P. Lee, “Safe Kernel Extensions Without Run-Time Checking,” OSDI 1996. 01 DECLARE POLICY 02 SUBMIT CODE + PROOF 03 CHECK THE EVIDENCE 04 ACCEPT OR REFUSE 05 EXECUTE ONLY IF ACCEPTED This record reconstructs the argument in its cited manuscript draft. It is not a certified procedure or an independent audit. THE JUDGMENT MACHINE Trace the chain. ObjectUntrusted packet-filter code submitted for execution with a supporting proof. ObservationThe submitted code and the proof accompanying it. DistinctionRespecting a specified safety policy is different from performing the intended task correctly. RepresentationA formal policy and machine-checkable evidence that the code respects that policy. CriterionThe proof must validate against the consumer’s stated safety requirements. JudgmentAccept only when the checker establishes the named policy. Do not enlarge that result into a claim of general correctness. AuthorityThe code consumer chooses the policy; the checker establishes whether the submitted evidence meets it. Last undo / seamBefore admitting the untrusted code to execution under the protected consumer’s privileges. ConsequenceThe consumer can refuse code lacking a valid safety argument before it executes. EVIDENCE & LIMITS Keep the denominator attached. The manuscript reports packet filters of 8–47 instructions, complete binaries of 385–1,024 bytes, and validation taking 780–2,350 microseconds in the cited experiment. These are results for the reported experimental setting, not current performance benchmarks. A valid safety proof does not establish that the filter chooses the right packets, and the guarantee extends only to the policy and checker assumptions. SOURCE TRAIL · The Last Undo, Chapter 4 draft v0.5.0, source note 1. Site reconstruction of the manuscript draft. Read the related manuscript record PREVIOUS / TRADING SYSTEMS Ninety-seven warnings before the market opened NEXT / AVIATION Rejected takeoff at V1 --- # Foundations of Computational Interaction (FCI) URL: https://judgmentmachines.com/methods/fci FCI Interaction & evidence / protocol Foundations of Computational Interaction How is interaction formally described, measured, and predicted? v0.5.0 · DOCUMENTED AND REFERENCE SUBSTRATE · NORMATIVE-CANDIDATE GOVERNS Supplies distinctions between representation, observation, evidence, interpretation, conformance, and empirical validity. BOUNDARY Reference tests establish kernel conformance, not empirical validity. AUTHORITY ROLE formal theory CONFORMANCE REFERENCE cd app && make test Canonical source foundations-of-computational-interaction Revision recorded git:f1892e4 Version authority registry-declared (no machine-readable package version) Adapter None declared Read the source REGISTERED RELATIONSHIPS Connected entries. FCI DEFINES ICP Return to the full register --- # Interaction Contract Protocol (ICP) URL: https://judgmentmachines.com/methods/icp ICP Interaction & evidence / protocol Interaction Contract Protocol What must a consequential interaction declare before it runs? v0.1.0 · DOCUMENTED · NORMATIVE-CANDIDATE GOVERNS Turns a bounded interaction into a declarable contract with evidence and verification outputs. BOUNDARY Schema conformance is structural validity; it does not prove usability or quality. AUTHORITY ROLE interaction contract CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source interaction-contract-protocol Revision recorded git:f1892e4 Version authority package.json Adapter icp Read the source REGISTERED RELATIONSHIPS Connected entries. FCI DEFINES ICP ICP QUALIFIES EQP Return to the full register --- # Evidence Qualification Protocol (EQP) URL: https://judgmentmachines.com/methods/eqp EQP Interaction & evidence / protocol Evidence Qualification Protocol Given this candidate, claim, context, and evidence, what is justified? v0.1.0 · DOCUMENTED AND FORMALIZED · NORMATIVE-CANDIDATE GOVERNS Qualifies evidence independently of a particular execution engine. BOUNDARY Qualification does not authorize promotion, deployment, publication, or spending. AUTHORITY ROLE evidence qualification CONFORMANCE REFERENCE npm test Canonical source evidence-qualification-protocol Revision recorded git:f1892e4 Version authority package.json Adapter eqp Read the source REGISTERED RELATIONSHIPS Connected entries. ICP QUALIFIES EQP IEPE QUALIFIES EQP ESP IMPORTS EQP Return to the full register --- # Intent and Evidence Project Engine (IEPE) URL: https://judgmentmachines.com/methods/iepe IEPE Work & external state / protocol Intent and Evidence Project Engine How does intent become authorized work, and what blocks promotion? v0.2.0 · IMPLEMENTED AND ADOPTED · OPERATIONAL GOVERNS Governs the path from intent through execution, evidence, evaluation, and promotion. BOUNDARY Evaluation and promotion remain separate. Comprehension and trust claims require human evidence. AUTHORITY ROLE operating governance CONFORMANCE REFERENCE python -m pytest Canonical source intent-evidence-project-engine Revision recorded git:f1892e4 Version authority README.md: Protocol Version badge Adapter iepe Read the source REGISTERED RELATIONSHIPS Connected entries. IEPE QUALIFIES EQP ESP IMPLEMENTS IEPE Return to the full register --- # Semantic Expression System (SES) URL: https://judgmentmachines.com/methods/ses SES Meaning & publication / protocol Semantic Expression System How can meaning stay fixed while expression changes completely? v1.0.0 · IMPLEMENTED AND TESTED · NORMATIVE GOVERNS Separates interface meaning from a particular visual rendering. BOUNDARY Interface meaning is not bound to any one rendering. AUTHORITY ROLE semantic expression CONFORMANCE REFERENCE npm test Canonical source semantic-expression-system Revision recorded git:f1892e4 Version authority package.json (protocol); PROJECT_PROFILE.json is project version Adapter ses-sps Read the source REGISTERED RELATIONSHIPS Connected entries. SES CONSTRAINS SSS Return to the full register --- # Semantic Surface System (SSS) URL: https://judgmentmachines.com/methods/sss SSS Meaning & publication / protocol Semantic Surface System Given this state and situation, what surface is warranted here? v0.1.0 · DOCUMENTED · NORMATIVE-CANDIDATE GOVERNS Selects a surface appropriate to the state and situation. BOUNDARY Surface selection does not decide domain truth. AUTHORITY ROLE semantic surface CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source semantic-surface-model Revision recorded git:f1892e4 Version authority VERSION Adapter ses-sps Read the source REGISTERED RELATIONSHIPS Connected entries. SES CONSTRAINS SSS SSS PROJECTS SPS Return to the full register --- # Semantic Publication Substrate (SPS) URL: https://judgmentmachines.com/methods/sps SPS Meaning & publication / protocol Semantic Publication Substrate How is published truth separated from its presentation? v0.1.0 · WORKING PROTOTYPE · OPERATIONAL-CANDIDATE GOVERNS Explores publication truth as a layer independent of presentation. BOUNDARY The Protocol Index describes the prototype as fixture-bound; it does not establish live publication or external-state mutation. AUTHORITY ROLE publication CONFORMANCE REFERENCE npm test Canonical source semantic-publication-substrate Revision recorded git:f1892e4 Version authority package.json (protocol); PROJECT_PROFILE.json is project version Adapter ses-sps Read the source The protocol version and project profile version are distinct. The Protocol Index describes this prototype as fixture-bound. REGISTERED RELATIONSHIPS Connected entries. SSS PROJECTS SPS Return to the full register --- # Polymorphic Decision Protocol (PDP) URL: https://judgmentmachines.com/methods/pdp PDP Research & writing / protocol Polymorphic Decision Protocol How is a high-uncertainty decision made with a panel of systems? v1.3.0 · REFERENCE DESIGN AND RECORDED CASE · RESEARCH-PROTOCOL GOVERNS Provides a reference design and recording apparatus for a bounded decision under uncertainty. BOUNDARY One recorded case does not establish that the protocol outperforms naive selection. AUTHORITY ROLE decision under uncertainty CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source polymorphic-decision-protocol Revision recorded git:f1892e4 Version authority README.md: specification version Adapter None declared Read the source Return to the full register --- # Human-Computer Mutual Legibility (HCML) URL: https://judgmentmachines.com/methods/hcml HCML Research & writing / research-corpus Human-Computer Mutual Legibility Can human intent and machine state be made mutually inspectable? v2.0.0 · RESEARCH CORPUS · RESEARCH GOVERNS Develops a research program around shared computational models, authority, uncertainty, and relationship state. BOUNDARY Implemented mechanisms and theoretical hypotheses are distinct. The theory is not empirically validated. AUTHORITY ROLE research CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source human-computer-mutual-legibility Revision recorded git:f1892e4 Version authority package.json Adapter None declared Read the source Return to the full register --- # Authored Prose Protocol (APP) URL: https://judgmentmachines.com/methods/app APP Research & writing / protocol Authored Prose Protocol What must authored prose do, and how is it checked? v0.1.0 · DRAFT AND SELF CHECKED · EDITORIAL GOVERNS Makes prose requirements explicit and mechanically checkable. BOUNDARY Checkability does not establish that prose is good, interesting, or correct. AUTHORITY ROLE authored prose CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source authored-prose-protocol Revision recorded git:f1892e4 Version authority registry-declared (protocol docs) Adapter None declared Read the source REGISTERED RELATIONSHIPS Connected entries. APP CONSTRAINS NEM Return to the full register --- # Priming Toolkit (NPS) URL: https://judgmentmachines.com/methods/nps NPS Research & writing / protocol Priming Toolkit What makes an agent primer measurably effective? v1.0.0 · SPECIFICATION AND MEASURED CORPUS · RESEARCH-PROTOCOL GOVERNS Pairs a specification with a measured corpus and validators. BOUNDARY Measurements do not validate the primers; the validating experiment is described as unrun in the index. AUTHORITY ROLE priming research CONFORMANCE REFERENCE python generate-measurements.py Canonical source normative-prose-specification Revision recorded git:f1892e4 Version authority README.md: title version Adapter None declared Read the source Return to the full register --- # Narrative Evidence Method (NEM) URL: https://judgmentmachines.com/methods/nem NEM Research & writing / method Narrative Evidence Method In what order does a qualified claim reach the reader? v0.1.0 · WORKING DRAFT · EDITORIAL GOVERNS Governs how qualified evidence and claims are presented in a narrative. BOUNDARY Evidence sufficiency belongs to EQP and the manuscript’s gates, not to narrative order. AUTHORITY ROLE narrative evidence CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source narrative-evidence-method Revision recorded git:f1892e4 Version authority registry-declared (method docs) Adapter None declared Read the source REGISTERED RELATIONSHIPS Connected entries. APP CONSTRAINS NEM Return to the full register --- # External State Protocol (ESP) URL: https://judgmentmachines.com/methods/esp ESP Work & external state / protocol External State Protocol How can an actor pursue an outcome they cannot establish alone? v0.3.0 · DEFINITION · OPERATIONAL-CANDIDATE GOVERNS Governs commitments, evidence, relationships, authority, reversible boundaries, and execution receipts for external outcomes. BOUNDARY Agency does not imply control over independent actors or institutions. Unresolved conditions remain gray. AUTHORITY ROLE external state governance CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source external-state-protocol Revision recorded git:f1892e4 Version authority registry-declared (protocol docs) Adapter esp-rcp Read the source Registry status is “definition.” The source index describes this as a working canonical candidate; both statements are preserved in the source material. REGISTERED RELATIONSHIPS Connected entries. ESP IMPLEMENTS IEPE ESP IMPORTS EQP RCP CONSTRAINS ESP Return to the full register --- # Recipient Contract Protocol (RCP) URL: https://judgmentmachines.com/methods/rcp RCP Work & external state / protocol Recipient Contract Protocol Who is owed a deliverable, and by when? v0.1.0 · DRAFT · OPERATIONAL-CANDIDATE GOVERNS Requires a named recipient and a stated obligation; work without a recipient is classified as primitive. BOUNDARY Recipients must not be retrofitted to justify existing work. Claim support alone does not establish an obligation. AUTHORITY ROLE recipient obligation CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source recipient-contract-protocol Revision recorded git:f1892e4 Version authority registry-declared (protocol docs) Adapter esp-rcp Read the source REGISTERED RELATIONSHIPS Connected entries. RCP CONSTRAINS ESP Return to the full register --- # Engagement Definition Protocol (EDP) URL: https://judgmentmachines.com/methods/edp EDP Work & external state / protocol Engagement Definition Protocol What has an engagement received, and what has it given? v0.1.0 · DRAFT · OPERATIONAL-CANDIDATE GOVERNS Records delivered artifacts, operational dependency, consideration, and terms deferrals. BOUNDARY Operational dependency is not an hourly price, a market rate, or an equity percentage. AUTHORITY ROLE engagement contract CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source recipient-contract-protocol/EDP.md Revision recorded git:f1892e4 Version authority registry-declared (protocol docs) Adapter esp-rcp Read the source Source heading: Engagement Dependency Protocol. The register title reads Engagement Definition Protocol. The naming difference is retained. Return to the full register --- # Protocol Concepts (CONCEPTS) URL: https://judgmentmachines.com/methods/concepts CONCEPTS Research & writing / reference Protocol Concepts Which early concepts helped explore the stack’s boundaries? v0.1.0 · REFERENCE MATERIAL · REFERENCE GOVERNS Collects architecture notes and diagrams around matter, interaction, and consequence. BOUNDARY Reference material is not a normative protocol and does not override specifications or conformance rules. AUTHORITY ROLE reference CONFORMANCE REFERENCE No conformance command declared in the registry. Canonical source concepts Revision recorded git:f1892e4 Version authority README.md: reference collection version Adapter None declared Read the source Return to the full register