# Practicum for "When Intelligence Became Cheap" — Book 3: The Economics of Trust Everything the printed chapters of this book send you to, in one file: worksheets, prompts and agent skills. Put it into ChatGPT Projects, a Claude project, NotebookLM or any assistant that accepts an attachment, then work through it question by question. > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. Assembled 2026-09-07. Chapter-to-material contract updated 2026-08-12. Canonical home: https://cheap-intelligence.vercel.app/en — each page below also lives there at its own address. The living layer is deliberately **not** in this file. Tool maps, market pages and anything else that carries a check date changes faster than a download: read it at https://cheap-intelligence.vercel.app/en/watch. Attribute material to Denis Ermilov and the Practicum, with the page address and its check date. Texts are CC BY-NC-SA 4.0; code and browser tools are MIT. --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. --- reading_level: b2-c1-technical title: "Book 3 Workbook: The Economics of Trust" description: An end-to-end route for deciding whom to trust and making your own claims easier to verify. --- # Book 3 Workbook: The Economics of Trust > The end-to-end Practicum route for *The Economics of Trust* (Book 3 in the > *When Intelligence Became Cheap* series). > This route through the contents asks **whom to trust when content, voices, and reputations are cheap > to fake, and how to become someone people can trust without excessive checking.** > > **Verifiability** is the axis of this route. Do not rely on a verbal claim. Arrange your work so that > important facts about you can be confirmed without your help. For the professional route, use the > [Volume 1 Workbook](tom1-workbook.md). For the business route, use the > [Volume 2 Workbook](tom2-workbook.md). Trust has two sides in this volume. The first is **protection:** avoid a fake call, email, or review, without paying the cost of checking everything. The second is **creation:** build a verifiable trail so that people and even AI search can choose you with less doubt. The workbook covers both sides. > **Did you arrive from a specific chapter?** The Practicum uses three "homes," not one page per > chapter. The [chapter index](tom3-chapter-index.md) gives each chapter one row and links straight to > its material. --- ## Minimum trust package If you do not have time for the whole volume, complete these five worksheets in one evening: 1. [Family trust protocol](family-trust-protocol.md): agree on a check phrase, talk with parents and children, print a note, and decide what to do first after a scam. Think of it as a seat belt. Set it up before anything happens. 2. [Personal trust perimeter](trust-perimeter.md): cover the work side with a second channel for costly instructions, a saved trail of work, and protection if someone makes an accusation. 3. [The first 24 hours with an important file](chain-of-custody.md): preserve the original, action log, independent trace, and context before any of them disappear. 4. [Trust footprint map](trust-footprint-map.md): check what other people can already verify about you without your help and where the gaps are. Then choose one move for the next month. 5. [Honest AI policy](honest-ai-policy.md): give clients one page that says where AI works, where a human is guaranteed, and who owns an error. These five worksheets cover the main route through the volume. You have basic protection, the important file remains intact, and other people can verify and choose you more easily. --- ## My trust protocol on one page Complete this page after the minimum package, or start here if you need one result instead of ten notes. Do not search for absolute authenticity. Decide in advance **what level of checking is enough for the cost of this error.** | Write this down | Home | Work and business | |---|---|---| | Costly situation | Who may ask for money, access, or an urgent decision? | Which instruction, payment, access request, or reputation risk must never be accepted from one message? | | Second channel | Which saved number or personal channel will you use to check the request? | Who confirms the instruction outside the channel controlled by the sender? | | Independent support | Who else can confirm the person or fact? | Where is the original, log, agreement, version, or other independent trace? | | What I show | How do people close to you know how to check that it is really you? | How can a client or coworker see the decision owner, the limits on AI, and the work history? | | Human owner | Who has the right to stop the payment or conversation? | Who owns the decision and repairs an error? | **If you made the mistake.** Write one sentence for the first few hours, one fact you will show a week later, and one process change. Do not promise "it will never happen again." Say what will be checked differently. The [trust repair protocol](trust-repair-protocol.md) gives the full route. **One move for the next 30 days:** **Next review date:** For a fillable version that you can export, [open the browser-based Trust Compass](../skills/trust-compass/index.md). See a deliberately imperfect first pass by a [composite family and independent professional](../cases/persona-family-trust.md). > **Review prompt.** "Here is my trust protocol. Find one situation with a high cost of error where I > still have not named a second channel or independent support. Do not suggest services or judge whether > a voice or file is authentic. Help me define one practical move for the next 30 days and a review date." --- ## The volume's three homes ### Home 1: Protection (Chapters 1 to 5). Avoid traps without paying to check everything The diagnosis is simple: convincing no longer means authentic. Voices, faces, reviews, and resumes can all be faked. You do not need to check everything. Check the cases where an error is costly. - [Family trust protocol](family-trust-protocol.md) for **home:** a check phrase, a conversation with parents and children, common schemes, limits for your own agents, a note by the phone, and the first hours after a scam. - [Personal trust perimeter](trust-perimeter.md) for **work and money:** a second channel for costly instructions, a work trail created before an accusation, protection during an accusation or a review attack, and your personal "distrust tax" account. - Live page, for readers in Russia: [protections available there](../watch/ru-fraud-protection.md), including voluntary credit bans, the "second hand" rule, cooling-off periods, and the rule that a bank does not write through messengers. These measures are free and take minutes to enable. ### Home 2: Verifiability and creation (Chapters 6 to 12). Keep a trail and become easier to verify Procedure connects protection and trust. Preserve an important file and build your own trail over time. When fakes are cheap, **verifiable human work** becomes more valuable. - [Chain of custody: the first 24 hours](chain-of-custody.md) explains what to do with an important file before an expert review or dispute: preserve the original, a log, a checksum, and an independent trace. - [Evidence package before a dispute](evidence-package.md) gives you a template for money, work, a deal, or reputation: what happened, where the original is, what the context was, and who can confirm it independently. It also asks how much checking is enough for a video, a hire, and a payment. - Live page: [trust and regulation](../watch/trust-regulation.md) covers who accepts a digital file, how to check the origin of someone else's file, and who owns an agent's error. - [Trust footprint map](trust-footprint-map.md) is the **flagship worksheet.** It puts three anchors, a verifiable history, live references, and a signed channel, across four channels: social media or Telegram, talks or conferences, books or publications, and community. Its main question is: *which parts are expensive to fake?* - [Honest AI policy](honest-ai-policy.md) treats trust as part of the product. It asks you to state how AI is used and make clear who your offer is not for. - [Trust inside a team](trust-inside-team.md) covers work controls: a register of bots that speak in your name and a hiring protocol for a world where an on-screen candidate is cheap to fake. - [Trust repair protocol](trust-repair-protocol.md) covers the other side of a verifiable trail: what to do when you made the mistake. First decide whether competence or honesty failed, then use the right process for that failure. - Live and reusable: the [AI search visibility kit](geo-ai-search-kit.md) helps neural systems find and cite you. Its trust layer asks not only whether you are visible, but what those systems say about you. ### Home 3: Compass (Chapter 13). Choose the depth and check both sides The final chapter turns the material into a decision route. It relies on boring protocols that still work when you are tired, not heroic vigilance. - [Trust Compass](trust-compass.md) has one cost-of-error arrow, three levels of depth, two territories (Home and Work and business), two sides (how I check and what I show), and four norms. - [Family trust protocol](family-trust-protocol.md) is a standalone route for a broad audience. It covers family members, parents, children, your own agents, and the first hours after a scam. - Live page: the [regulatory tracker](../watch/trust-regulation.md) covers AI labeling, provenance, and laws. It changes quickly. --- The volume is framed by **three trust questions:** "Who is speaking? What do they answer for? Who can confirm it independently?" It also uses one rule: check what is costly, not what looks strange. The book explains the questions, and the [Trust Compass](trust-compass.md) turns them into levels of review. --- > **How this section is built.** Work on the Volume 3 Practicum began before the chapters, and it is > now synchronized after each author acceptance. Each worksheet stands on its own. Material that > changes quickly, including schemes, laws, and tools, lives in the Live section with review dates. > See the [Volume 3 sources](../sources/tom3.md). --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. # Family trust protocol: one conversation before the urgent call > Practicum for **Book 3, *The Economics of Trust*, Chapter 13**. > Setup time: 15 to 20 minutes. This protocol does not teach you to detect a fake voice. > It gives you steps that work whether the voice is real or fake. ## 1. Choose a family check phrase The phrase is not a password that authorizes a payment. It is a **stop signal**. A correct answer lowers the risk, but you still confirm an expensive request through an independent channel. A good phrase: - does not appear on social media, in messages, or in family greetings; - is not a pet's name, address, birth date, or mother's maiden name; - is easy to say and remember; - is known only to relatives who follow the agreement. We keep our phrase in this agreed secure place, not on this page: _______________ People who know it: ___________________________________________________________ Test it aloud once during a normal conversation. Do not send it through the same chat where a fraudulent request could arrive. ## 2. Agree on the response If a voice, video, or message from a relative asks for money, access, a document, or a secret code: 1. Ask for the check phrase. 2. Whatever the answer, do not pay or share information through the same contact. 3. End the conversation. 4. Contact your relative yourself through a saved number or another relative. 5. Do nothing if you cannot confirm the request. Urgency does not stop you from helping. It stops you from acting through one channel. Treat anyone who claims to represent a bank or public agency the same way: end the call, then contact the organization through an official number you found yourself. ### What these scams usually look like You do not need a catalog of horrors or a lesson in detecting a fake by ear. That approach does not work. These schemes follow a similar script, and one rule can break all of them. | Scheme | What happens | What makes the attack harder | |---|---|---| | "A relative is in trouble" | A voice that sounds like a child or grandchild urgently asks for money | Ask for the check phrase, then call the saved number yourself | | "Bank security" | The caller leads you toward a transfer to a "safe account" or asks for a text-message code | Hang up and call the number printed on the card | | A call presented as a survey | The caller tries to collect a recording of your voice to build a clone | Do not have long conversations with unknown callers | | A video message "from a relative" | The face and voice look real, but the request comes from someone else | A video is not confirmation. Apply the same rule | > Shared pattern: **urgency plus money, data, or access means stop.** A real emergency can survive a two-minute check. Workplace schemes, including a fake message from an executive, changed payment details, and review attacks, appear in the [Personal trust perimeter](trust-perimeter.md). ## 3. Talk with parents and older relatives Do not start with a list of technologies or frightening examples. Say this: > "Scammers can now speak with the voices of people you know. You do not have to tell a real voice from a fake one. If anyone asks for money, hang up and call me yourself using your saved number. The real me will never be offended by the check." Then add the second half: > "If they fool you, or if you only start to feel unsure, call me right away. I will not scold you or take away your independence. First we will stop the loss. Then we will work out what happened." Removing shame is part of the protection. A person who fears the family's reaction waits longer before calling the bank or police. ### Add the bank controls available where you live Check whether your bank lets you lower transfer limits, require an extra confirmation for large payments, freeze a card quickly, or disable features you do not use. Names and rules differ by bank and country. Use the bank's official app or site to confirm the current terms, then choose controls your family can explain and use. Do not assume that a cooling-off period, dispute deadline, or reimbursement rule works the same way everywhere. Check the current process with your bank or local financial regulator before you need it. ## 4. Children and teenagers A child needs the same check phrase and a shorter rule: > "If an adult you do not know says your parents are in trouble, asks you to turn on the camera, find money, show valuables, or keep the call secret, end the conversation and call your trusted adult right away." A child does not help an "investigation" by phone, show the home on video, or search for family money at a caller's direction. ## 5. Your AI agents and family money If an assistant can order or pay for things, make an agreement with it too: | Setting | Our decision | |---|---| | Maximum amount for one purchase | | | Maximum amount per month | | | What it may buy on its own | | | What it may only add to a cart | | | Which subscriptions and extra charges are forbidden | | | Where we can see notifications and the activity log | | | How to cancel an action or dispute a payment | | If you cannot remember the agent's limit, assume there is no limit. ## 6. The first hours after fraud Stop the loss first. Work out how it happened afterward. 1. **Contact the bank through an official number** printed on the back of the card, shown in the banking app, or listed on the bank's official site. If you may have exposed card details or remote access, ask the bank to block the card and remote access. 2. **Preserve the trail:** time, phone numbers, addresses, messages, receipts, screenshots, and the amount. Do not delete or "improve" anything. 3. **Ask the bank to open a dispute or recall request** if money moved without your consent or because of fraud. Follow the deadline and evidence rules in your country. Record the case number and every answer. 4. **Use your local fraud-reporting route.** This may be the police, a national fraud service, or a financial regulator. Keep the report number and any written confirmation. 5. **Tell someone close to you.** Do not face the shame and follow-up calls alone. So-called helpers may promise to recover the money and try to scam you again. A bank does not always have to return the money, especially when the customer confirmed the transfer. Do not promise yourself or your relative a guaranteed refund. Act quickly and record every contact. --- ## Card to keep by the phone > **ARE THEY ASKING FOR MONEY OR DATA?** > > 1. I do not argue or transfer anything. > 2. I hang up. > 3. I call my relative or bank myself using a saved official number. > 4. I never share text-message codes, card details, or passwords. > 5. If I already shared or transferred something, I call the bank and my relative at once. > > **A real relative will not be offended by a check.** Print this block in large type and leave it near the phone. Rehearse the sequence once: who calls whom, which number they use, and what happens if the first person does not answer. ## A separate large-print card for older relatives Print this version in the largest practical type and leave it by your parents' home phone. It is shorter because nobody reads a long list during a stressful call. > **OUR PHRASE: `______________`** > > If someone calls and asks for money, even in the voice of my child or grandchild: > > 1. I ask for our phrase. > 2. I hang up. This is not rude. > 3. I call back myself using my saved number. > 4. I never share a text-message code. With anyone. Ever. > 5. A bank or the police will never ask me to transfer money to a "safe account." > > **If I already transferred money or shared a code, I call right away:** > > My relative: `______________` | Bank: number on the back of my card > > **Nobody will scold me. The sooner I call, the better the chance of getting the money back.** The last line is required. Shame can keep a person silent for a full day, and that day can decide whether the money comes back. *Related worksheets:* [Trust compass](trust-compass.md), [Personal trust perimeter](trust-perimeter.md), and [The first day with an important file](chain-of-custody.md). --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. # Personal trust perimeter: work, money, and your good name > Practicum for **Book 3, *The Economics of Trust*** (the Protection house). Setup time: 20 minutes. > Four protocols for the working side of life. Set up Sections 1 and 2 in advance. Read Sections 3 and 4 when you need them. These protocols are not about paranoia. They make deception **more expensive and easier to notice**, and they help you stay calm. When an email "from the CEO" arrives, or someone claims that "AI wrote this," a rule you created in advance matters more than alertness. Set it up once. It will still work when you are tired or in a hurry. > **Home has its own worksheet.** The [Family trust protocol](family-trust-protocol.md) contains the check phrase, conversations with parents and children, common call scams, and a card to keep by the phone. Start there. It protects family money, while this worksheet protects work and reputation. --- ## 1. A second channel for expensive instructions at work A convincing message "from management" is a new wrapper around an old fraud. A procedure protects you better than your eye. 1. **List expensive actions** that nobody may approve from one call or message, however convincing it looks. Examples include a payment, changed payment details, access transfer, or data release. Your list: `__________` 2. **Set a second-channel rule.** If an instruction arrives by voice, video, or messenger, confirm it through an independent channel in the company directory. Do not use contact details from the incoming message. 3. **Give the manager a script for the team:** "I will never ask you to make a transfer through a messenger. If I do, call my work number. That is an instruction." 4. **Give junior employees the right to verify.** A callback follows procedure. It does not show distrust, and nobody is punished for the pause. --- ## 2. Build a work trail before an accusation AI detectors can accuse a real person of using a model. Protect the work in advance with one setting that can serve you for years. - [ ] I create important work in a system that **keeps version history**, such as Google Docs, Microsoft 365, Notion, or a Git repository. - [ ] I keep drafts and source material. A writer keeps versions, an analyst keeps data and calculations, a photographer keeps source files, and a developer keeps commits. - [ ] I establish dates at no cost by emailing a draft to myself. For important work, I also export a dated copy. ### If someone accuses you Do not defend yourself with words alone. **Show the process.** Edit history is more convincing than a final file or a detector score. Use this order: 1. **Do not argue about the percentage.** The detector score is neither something you can debate nor the claim you need to disprove. Move the conversation to the process. 2. **Show how the work developed:** version history, dated drafts, intermediate versions, source files, and data. One screen with edit history can resolve the question faster than an hour of explanation. 3. **Answer questions about the substance.** The author can explain any part without reading from the text: why they chose an approach, where a figure came from, and what they removed. 4. **Ask for the basis of the decision in writing.** If the only basis is a percentage, preserve that fact. The next question is about the decision process, not your character. 5. **Do not accept an oral compromise** when the outcome matters, including a grade, disciplinary action, or termination. Ask for the decision and its basis in writing. **There is support for this approach in Russian court practice.** In June 2026, the Supreme Court of Tatarstan ruled that a student's expulsion over alleged "AI text" in a thesis was unlawful. Three consecutive checks of the same file in Antiplagiat.VUZ reported AI-content scores ranging from 41.91% to 11.92%. The court stated that the system's report **was probabilistic**, while the final decision belonged to an expert. The student was reinstated, and the university had to pay compensation for moral harm and a fine. Here, "probabilistic" means that the score expressed likelihood, not proof. Source: *Kommersant*, June 8, 2026: [the court's position](https://www.kommersant.ru/doc/8727298) and [reinstatement and compensation](https://www.kommersant.ru/doc/8727878). *Verified: 2026-07-29.* Practical result: ask for another check of the same file and record the spread. The same text producing unstable scores became an argument the court understood. ### If you are the decision-maker This is the other side of the same problem, for a teacher, manager, or editor. Before making a decision based on a "percentage," ask: - [ ] Do I understand **what the detector measured** and know its error rate? - [ ] Did I ask to see the process rather than demand a defense? - [ ] Did I ask at least one question about the substance of the work? - [ ] Am I punishing a style, such as careful prose, a formulaic genre, or writing by a nonnative speaker? - [ ] Do I have a **second independent** signal beyond the score? - [ ] Did I let the person respond **before** the decision rather than after it? > The cost of an error may be another person's reputation, diploma, or job. In this case, verify your own decision as carefully as you verify their text. --- ## 3. Review attack: the first 24 hours for a small business If someone floods your business with fake reviews or demands money to remove them: 1. **Do not pay.** Payment does not stop the attack and tells the attacker that you may pay again. 2. **Preserve the evidence.** Save screenshots of reviews, profiles, and demands, with dates and times. 3. **Report it to the platform** through its fake-review or extortion channel. Current forms appear in the [regulatory tracker](../watch/trust-regulation.md). 4. **Ask real customers for help.** Publish an honest note and ask for genuine reviews. Do not buy a batch of positive reviews as a "repair." That continues the same downward path. 5. **Report extortion to the police.** Bring in a lawyer if the scale calls for one. --- ## 4. Your personal distrust-tax account This optional exercise takes one week. Choose **one** process, such as incoming calls, order intake, or a family phone. Track what distrust costs: | Line | What to record | |---|---| | Verification | minutes and money spent checking | | Delay | what waited while the check happened | | False rejection | which real person or request you rejected as fake | | Missed fake | what still got through | Question for the week: **Where did I verify something strange when I should have verified something expensive?** This calibrates your effort toward decisions with a high cost of error instead of every unusual detail. ### The mirror: what do you cost other people? There is another side to the account. You do not only pay a distrust tax. You may charge it to a client, customer, colleague, or employer. Review the same four lines from their point of view: | Line | What to inspect about yourself | |---|---| | Verification | how many steps someone takes before believing they can work with you | | Delay | how long they wait for you to confirm something personally | | False rejection | how many good people leave before they get confirmation | | Accepted on trust | which claims in your offer cannot be checked at all | One question: **What claim about me cannot be confirmed without asking me?** Each answer is a tax other people pay and a reason to choose someone else. - Most expensive line: `__________` - First thing to make verifiable: `__________` Take that answer into the [Trust footprint map](trust-footprint-map.md), where it becomes one move for the month. --- > **No promise of invulnerability.** These protocols make deception more expensive and easier to notice, not impossible. > For home, use the [Family trust protocol](family-trust-protocol.md). Next, open the [Trust footprint map](trust-footprint-map.md) to move from protection toward becoming easier to trust. --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. # Trust footprint map: make it easier to choose you > Practicum for **Book 3, *The Economics of Trust*** (the Creation house). Time: 30 to 40 minutes. > A worksheet for an independent professional or small business. Fill it in on paper or in a note. A review, resume, or feed full of "experts" can be faked in an evening. That makes **verifiable human work** more valuable. The person who is easiest to verify can beat the loudest one. This worksheet is not a personal-brand exercise. It reduces the work another person must do to check the claims that matter, even when you are not there to explain them. > **Use four questions throughout the worksheet.** Is the signal hard to fake? Does it build over time? Can someone check it from outside? Will it lower the cost of another check during a future decision? > A badge, a self-description, and three glowing reviews fail this test. A long work history, real clients, a person willing to confirm the work, and a channel with a past perform much better. You can make a first move in one evening. It becomes capital through repetition, outside support, and consequences. --- ## Step 1. Three anchors: what do people check? Professional trust has three anchors. Mark what you already have that someone can **verify**. "I am good" does not count. "Someone can confirm this without me" does. | Anchor | What it means | What I have now | Expensive to fake? (yes/no) | |---|---|---|---| | **Verifiable history** | real work, case studies, dates, and results that someone can check | `__________` | | | **Live references** | named people who will confirm the work and speak with someone | `__________` | | | **Signed channel** | a named channel with your history and identity, rather than an anonymous account | `__________` | | > An empty row is a weak point. One complete anchor marked "expensive to fake: yes" carries more weight than ten bold claims. The third anchor does not depend on the platform alone. It depends on **consequences**. Words in a signed channel are tied to you, so you may have to answer for them. That is why people trust the channel more than an anonymous review. The statement has a price. --- ## Step 2. Four channels in your footprint Leave a trail where people look for you. For each channel, record what already exists, what a stranger can see without you, and one move for the next month. You do not need all four. Choose one or two that your clients really use. | Channel | What already exists | What people can see without me | One move this month | |---|---|---|---| | **Social media / Telegram:** where you think in public | | | | | **Talks / conferences:** where people saw you live | | | | | **Books / publications / articles:** what remains in writing | | | | | **Community:** a smaller circle where people know you personally | | | | > A smaller circle of people who know you can be stronger than broad reach. Trust is deeper there, and a reputation among real people is hard to fake. This is your "dark forest." See the [Trust compass](trust-compass.md). ### What a machine can read Some parties checking you are already software systems. A client may ask an assistant instead of a search engine. A machine has no intuition and will not fill in missing context for you. | What leaves a trail for a person | What a machine can extract | |---|---| | A post where you think in public | your specialty, if you use the same wording across sources | | A talk where people saw you | name, topic, date, and venue, if they appear in text | | A publication or case study | a figure with a unit, a date, and a link to the primary source | | A conversation inside a community | nothing, because the closed circle is invisible to the machine | This does not mean "write for robots." State facts that affect a decision clearly. Name the responsible person instead of "the team." Give a date instead of "recently." Link the primary source instead of saying "research shows." For current context, see [how much of the web is machine traffic](../watch/machine-web.md). --- ## Step 3. Keep a show-your-work rhythm You cannot create authenticity capital by declaring it. It remains as a side effect of real work. Publish an artifact rather than a post about yourself: a case analysis, a method, a record of how you thought, or an honest review of an error. One artifact a week or a month can create a trail over a year that is hard to fake. - My show-your-work format: `__________________` - Where I publish it (one channel I control): `__________________` - A realistic frequency: `__________________` Rule: **process is more convincing than polish.** "How it worked" and "where I was wrong" often build more trust than a perfect storefront. --- ## Step 4. Signed channel and provenance One channel that is clearly yours is a strong anchor. You can also attach **content provenance**, a record of origin and changes, to important material so another person cannot easily claim your work, or put your name on a fake. - My official channel, where I send everyone: `__________________` - Where I tell clients that **I write only from this channel** to protect them from impersonation: `__________________` - Content provenance ([Content Credentials / C2PA](https://contentauthenticity.org/)) is optional. Consider it for photos, videos, or other material that someone could present under a false name. *Verified: 2026-07. Support is growing in cameras and Adobe products, but the label does not always reach the viewer. It is an extra signal, not a guarantee.* Current status and limits appear in the [regulatory tracker](../watch/trust-regulation.md). ### Personal provenance in one evening Six settings that help people confirm your authorship without you: - [ ] **I can find the originals.** Source photos, text versions, and talk recordings stay in one place after publication. - [ ] **Important material has a date I can show.** I publish it in my channel, export it with a date, or email the attachment to myself. - [ ] **I have named one official channel** and say so where people most often find me. - [ ] **I state what I never do.** For example, I do not write first about money, request transfers, or give access through a messenger. People can reject an impersonator without forensic analysis. - [ ] **I enable content provenance** for material that someone may realistically present under another name. - [ ] **I know where to go if someone puts my name on their work:** the platform, complaint route, and deadline. Current contacts appear in the [regulatory tracker](../watch/trust-regulation.md). > Provenance is not about paranoia or a particular technology. It makes confirmation **cheap** for a person who has doubts about whether the material is yours. --- ## Step 5. Can AI search see you, and what does it say? A client may ask an AI system "whom should I hire?" instead of using Google. If you do not appear in the answer, you are absent from that decision. If the answer describes you poorly, you can lose trust before the client reaches your site. Use the [AI search visibility kit](geo-ai-search-kit.md) for a full measurement. For a quick check of the structure an AI can cite: - [ ] My position or specialty stays consistent over time. - [ ] I have a clear method that another person can describe. - [ ] My work has a named author rather than an anonymous label. - [ ] I have a stable address, channel, or page that does not keep moving. - [ ] I have a history of accurate work that someone can confirm. More checked boxes make it easier for both a person and an AI assistant to recommend you. --- ## Result: one move Find the empty place that is **cheapest for you to repair and most expensive for your client to check**: - My weakest anchor: `__________` - One move this month that will make it verifiable: `__________________` > Test the move with four questions. Is it hard to fake? Will it build over time? Can someone verify it without you? Will it simplify more than one future check? If yes, you are building authenticity capital. If the move only improves the storefront quickly, it is cosmetic rather than a footprint. --- *Related worksheets:* [Honest AI policy](honest-ai-policy.md), [Trust compass](trust-compass.md), and [Personal trust perimeter](trust-perimeter.md). *Volume voice:* be verifiable, not loud. --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. # Evidence package before a dispute > Practicum for **Volume 3, Chapters 6 and 10**. Time: 20 minutes for one situation. > A worksheet for money, work, transactions, and reputation. It does not replace advice from a qualified professional. Build verifiability before a conflict begins. Choose one situation where an error would be expensive. Connect four layers in advance: **what happened, source material, context, and an independent record**. --- ## 1. Describe the future dispute in one sentence > If a dispute begins, I will need to confirm that _________________________________ Do not write, "I am right." Write a fact someone can check: an instruction was given, a result was approved, an item had a certain condition when transferred, or a publication existed on a certain date. ## 2. Gather four layers | Layer | What belongs in the package | What I have | |---|---|---| | **Event** | date, participants, action, amount, or subject | | | **Source** | original file, email, contract, photo, or recording | | | **Context** | messages before and after, document version, call record | | | **Independent record** | statement, system log, email header, or confirmation from another person | | If you fill only the source row, the package is weak. A polished file does not explain where it came from or what happened around it. --- ## 3. Choose your situation ### Money - contract, invoice, and original payment details; - separate confirmation of a change in recipient or amount; - payment order, receipt, and statement; - bank case number and the time you reported the incident; - company approval log, if one exists. **Rule for the future:** Confirm new payment details and urgent payments through an independent channel. ### Work - written employment or project terms; - task description and definition of done; - versions of the result and comments on them; - email or system record of approval; - log of the delivery and later changes. **Rule for the future:** Turn an important spoken promise into a short same-day email: "Have I understood this correctly?" ### Transaction - contract and attachments; - photos or video of the item's condition before transfer; - serial number, included parts, and meter readings; - transfer document, receipt, and shipping record; - confirmation from the other party or an independent participant. **Rule for the future:** Record the condition before the item or access changes hands. ### Reputation - original post or message; - full page address and date; - screenshot with the surrounding page, not only a cropped sentence; - archived copy, platform notice, email, or another person's screenshot; - log of reports to the platform and its responses. **Rule for the future:** Record the content, account, address, time, and platform response. --- ## 4. Check independence Ask four questions about each item: 1. Is it stored only with me? 2. Could someone change it without leaving a trace? 3. Which person or system can confirm it independently? 4. Will that record still exist in a month? **Weakest layer in the package:** _________________________________________________ **One move today:** ________________________________________________________ --- ## 5. How much is enough for a video, a hire, or a payment? A formal dispute is the extreme case. You usually need to answer a smaller question: **How much verification is enough for this decision?** The answer depends on the cost of being wrong, not on how strange the item looks. ### What a credential really confirms | Credential or evidence | What it confirms | What it does not confirm | |---|---|---| | Diploma or certificate | the person completed a program at a stated time | that they can do the work today | | Platform review | someone posted positive words | that the author exists or had the claimed experience | | Case study or portfolio | a result is shown | that this person produced it | | Product demo | the system works on prepared data | that it will work on your data | | Named recommendation | the author is willing to put their reputation behind the statement | that they examined the work closely | | Independent audit or measurement | a third party checked something with a stated method | that the method fits your situation | | Signed file or provenance record | origin and change history | that the content is true | ### How much to check at different levels of risk | Decision | What you lose if you are wrong | Enough verification | |---|---|---| | Watch or forward something funny | nothing | none; save your attention | | Cite something publicly as fact | your reputation | primary source and date | | Hire someone | months of work and access | live conversation, work artifacts, and one independent person who will stand behind the reference | | Pay or change payment details | money, often with no way back | confirmation through a channel known **in advance**; for a large amount, add a second signer or limit | | Share data or access | the change may be irreversible and may affect other people | all of the above, plus a written basis and an expiration date | Record the result as one of three states: **confirmed, probable, or unknown**. Apply the state to this decision, not forever. Strong verification does not pretend to settle everything. It makes the remaining uncertainty visible. > Turn the table toward yourself. Which claims **about you** can survive these questions? Which parts of your portfolio or proposal still depend on "take my word for it"? Continue with the [Trust footprint map](trust-footprint-map.md). --- ## What looks like evidence but often is not - a screenshot without an address, date, or context; - a forwarded video instead of the original; - an "enhanced" recording with no preserved original; - a clearly named file with no history of how it appeared; - your own log with no link to another system; - a spoken agreement that everyone remembers differently. --- ## First page of the package ```text What happened: When: Participants: What must be confirmed: Where the original is stored: Derivative copies created: How integrity was checked: Independent record: Who was notified and when: Who keeps the package: Next deadline: ``` Do not collect everything. The package should help the next person, such as a bank employee, editor, or lawyer, quickly see origin, integrity, and context. *Related worksheets:* [The first day with an important file](chain-of-custody.md), [Personal trust perimeter](trust-perimeter.md), and [Trust footprint map](trust-footprint-map.md). --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. # Trust inside the company: bot register and hiring protocol > Practicum for **Book 3, *The Economics of Trust*** (the Verifiability and Creation house, work track). > Time: one meeting. For a manager, small-business owner, or anyone who hires people. People are no longer the only voices speaking for your company. The website chat, an email autoresponder, and an assistant inside the customer portal can all make promises to a client. Someone is responsible for each promise. Across the table, the same problem appears in reverse. The candidate on screen may be a constructed scene rather than the person you think you are interviewing. These two problems have the same root. Work through them in order. Everything below costs nothing. --- ## 1. Company bot register List every automated voice. Fill in the register once, then update it whenever you add a channel. | Bot or channel | Where it speaks | Says it is software in the first line? | Owner of its words (name) | What it promises for us | |---|---|---|---|---| | `______________` | | | | | | `______________` | | | | | | `______________` | | | | | | `______________` | | | | | Three rules for filling it in: - **The owner is a person, not a department.** "Support" cannot take responsibility. A named person can. - **A bot's name and photo are decoration, not disclosure.** They do not replace the line "You are speaking with a software system." - **An empty owner cell is an answer, not an omission.** If you have not assigned an owner, you are responsible for the channel's words. > The line "You are speaking with an AI assistant" is cheap before a scandal and expensive after one. This register is internal. Its public version is the [Honest AI policy](honest-ai-policy.md), which explains the same arrangement in language a client can use. --- ## 2. Hiring protocol: avoid hiring someone who does not exist Faking a resume, face, and voice can cost less than completing an interview. A sharp eye is not enough. A procedure creates a **trail and consequences**, while a fake candidate tries to avoid both. ### Before the interview - [ ] Tell the candidate in writing and in advance if you will record the interview. - [ ] Get consent to process personal data before the conversation, not after it. - [ ] Check that the contact method leads to the same person named in the application. ### During the interview - [ ] Make at least one stage live and spontaneous rather than fully scripted. - [ ] Use one move from this list: change language in the middle of the conversation, ask to see the working environment, ask an unplanned question, or return to a detail the candidate mentioned ten minutes earlier. - [ ] Watch how the person improvises, not how polished the image looks. A prepared scene can survive the script. A departure from it is harder. ### After the offer - [ ] Before granting access, confirm identity through an independent channel, not the channel the candidate used to approach you. - [ ] Do not grant access to money and data on the first day or through one approver. - [ ] If something does not match, pausing before access costs almost nothing. Giving access to the wrong person can cost everything. **Person responsible for this protocol:** `__________` --- ## Result: one move this week Do not implement everything at once. Choose the move that closes the most expensive gap: > **Our move:** ______________________________________________ > > **Owner:** __________ **Due date:** __________ The first move is often the least exciting one: assign a human owner to every bot. --- *Related worksheets:* [Honest AI policy](honest-ai-policy.md), the public side of the register; [Personal trust perimeter](trust-perimeter.md), the second channel for expensive instructions and a junior employee's right to verify; and [Trust compass](trust-compass.md), which sorts a workweek by the cost of error and assigns a "trust" or "verify" mode to each task. --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. # Chain of custody: the first day with an important file > Practicum for **Volume 3, Chapter 10**. Time: 10 to 15 minutes. > This is a practical first-preservation protocol, not digital forensics or legal advice. An important file rarely loses a dispute because it looks unconvincing. The problem is often simpler. Nobody knows where the file came from, whether it changed, or who handled it. This worksheet helps you preserve your options. It cannot guarantee that a bank, newsroom, or court will accept the file, but it can stop you from destroying a useful trail yourself. --- ## Start with four actions ### 1. Save the original - Take the source file from the device or service where it first appeared. - Do not make a messenger copy your only copy. The service may compress the file or remove some metadata. - Create a working copy. Store the original separately and do not edit it again. - Preserve the nearby context: the conversation, email, call record, receipt, and login log. **Path to the original:** ________________________________________________ ### 2. Record who, when, and what | Date and time | Who had the file | What they did | Recipient or storage location | |---|---|---|---| | | | | | | | | | | | | | | | Fill in the log right away. You can add to your own log later, so the log alone proves nothing. Its value comes from links to independent records, such as an email, cloud version history, bank case number, or police report number. ### 3. Create an independent record of the file Choose a method that matches the cost of a possible error: - A **checksum** shows that two files match bit for bit. It does not identify the author or prove a date by itself. - A **channel with an external timestamp**, such as a company system, cloud version history, or registered case, can help show that the file already existed at a certain time. - A **specialist or notary** is useful when a dispute is already likely and the cost of an error is high, not simply "just in case." **Checksum or external record:** __________________________________________ ### 4. Do not improve the original Noise reduction, upscaling, editing, and a command such as "make it clearer" create a derivative version. That version may be easier to inspect, but it is no longer identical to what the device recorded. - Leave the original untouched. - Give an enhanced or cropped version a different name. - Record which tool created the derivative copy and why. --- ## During the first day - [ ] Contact the bank on the same day if money is involved. Save the case number. - [ ] Save the bank statement, receipt, and recipient details. - [ ] Export the conversation or email with dates and technical headers when they are available. - [ ] Record the names of people who witnessed the event or saw the file. - [ ] Ask the employer, platform, or contractor to preserve system logs if they control them. - [ ] If a dispute is likely, contact the right party: the bank and police for money or fraud, a lawyer for a contract or workplace issue, or a fact-checking desk for a publication. ## What usually weakens a file - the only copy is a forwarded message or screenshot; - renaming the file without recording its original name; - editing the original; - deleting nearby messages and notifications; - waiting a long time before creating the first independent record; - assuming that "the recording speaks for itself." --- ## Transfer record Copy this block every time you transfer the file: ```text File: Original name: When and where it appeared: Received from: Checksum: Transferred to: Date and method of transfer: What was transferred: original / exact copy / derivative copy Changes made: Related independent record: ``` --- This protocol is based on [NIST SP 800-86](https://www.nist.gov/publications/guide-integrating-forensic-techniques-incident-response): record every action involving digital material, preserve the source, and verify integrity. NIST describes it as a practical guide to computer forensics, not legal instructions for every country. *Related worksheets:* [Personal trust perimeter](trust-perimeter.md), [Evidence package before a dispute](evidence-package.md), and [Trust and the law](../watch/trust-regulation.md). --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. # An honest AI policy: trust as part of the product > Practicum for **Book 3, *The Economics of Trust*** (the Creation house). Time: 20 to 30 minutes. > A one-page worksheet for an independent professional or small business. You will leave with a short statement you can show a client without embarrassment. Clients increasingly want to know where you use AI, where a person stays involved, and who takes responsibility when the machine gets something wrong. Silence can look like concealment. An honest AI policy is not a legal document. It answers the client's questions before they have to ask. People pay a premium for honesty, not for a business that pretends to be either "fully human" or "fully AI." --- ## Part 1. Where we use AI and where a person takes over Review your services and mark each one honestly: | What we do | AI works on its own | AI works under human review | A person does the work | |---|---|---|---| | `______________` | | | | | `______________` | | | | | `______________` | | | | Rule: **the higher the cost of an error for the client, the farther right you should move.** A person must stay responsible when the work affects money, health, legal rights, or an irreversible decision. --- ## Part 2. Who takes responsibility for an error? Write one sentence a client can understand without a lawyer: - The person responsible for the result is: `__________` (a name or role, not "the system"). - If AI makes an error, the client should do this, and we will do this: `__________________` - The client can reach a person here, without fighting the bot: `__________________` > Easy access to a person is a product feature, not overhead. It matters even more when every competitor sends clients to a bot. --- ## Part 3. Set a boundary: who are you not right for? Trust grows when you say clearly **who you cannot help** and what you will not do. A boundary does not drive away the right clients. It helps them see that you are not promising everything to everyone. - We **do not** take on: `__________________` - We are not a good fit if the client needs: `__________________` - We will not compromise these principles: `__________________` > A boundary filters out poor fits and makes your offer easier to trust for the people you can help. "We are not for everyone" is more honest than "we serve everyone." --- ## Part 4. Put it into one paragraph Combine your answers into a short statement for your website, profile, or offer: > "We use AI for `___` so we can work faster and at lower cost. A person always takes part in `___` because an error there would be expensive. `___` is responsible for the result. If something goes wrong, email or call `___`, and a person will review it. We do not take on `___`; another provider will be a better fit for that work." This is your honest AI policy. Update it whenever you change what you hand over to a machine. --- *Related worksheets:* [Trust footprint map](trust-footprint-map.md) and [Trust compass](trust-compass.md). *Volume voice:* clients pay a premium for work they can verify and for people who speak honestly, not loudly. --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. # AI search visibility kit: how AI finds and cites you > Practicum for **Volume 2, Chapter 6**, "When the Customer Is Not Human." Time: 30 to 40 minutes. > This is a manual worksheet for a first measurement. The details change quickly, so check the date. Search is turning into an **answer**. A person may ask ChatGPT or another assistant what to buy or whom to hire, then receive a shortlist without opening ten links. A high search rank or an SEO trick is not enough. You need to appear in sources the AI trusts and can cite. "We have a website" becomes "the same checkable facts about us appear across sources a machine can read." > Gartner forecasts declining traffic from traditional search, with estimates of 25% less search volume by 2026 and 50% less organic search traffic by 2028. The KDD 2024 paper "GEO" by Aggarwal et al. found that source citations, quotations, and statistics produced the largest gains in visibility inside generative answers, about 30% to 40% for the best-performing methods. Treat these figures as reference points. Primary sources appear below. --- ## Step 1. Map AI answers Write five to seven real questions your customers ask. Run each question through three or four AI systems, such as ChatGPT, Gemini, Perplexity, or a locally popular assistant. Record what appears in the answer: | Customer question | Brands in the answer | Sources cited | Tone about you | |---|---|---|---| | `________` | | | positive / neutral / negative / absent | | `________` | | | | | `________` | | | | > If you do not appear, the answer has left you out. If you appear in a weak or negative context, trust can fall before the customer reaches your site. ## Step 2. Check the three conditions for being chosen The chapter gives you three levers. Mark the weak ones: - [ ] **Findability:** Can the agent find you in the places it uses to build an answer? - [ ] **Evaluability:** Can the agent quickly understand your value from a clear description, price, and examples? The technical term is *token-to-value*: how much useful information the agent gets from a small amount of text. - [ ] **Actionability:** Can the agent take the next step without a negotiation, using a clear contact, checkout, or payment path? Weakest lever: `________________`. Start there. ## Step 3. Measure AI search visibility Do not track only a Google position. Track your presence inside answers: | Metric | What it means | |---|---| | Share of AI mentions | how often you appear in answers to your test questions | | Position / order | whether the answer names you first or near the end | | Sentiment | whether the description is positive, neutral, or negative | | Source share | which sources the system cites about you and whether it trusts them | | Co-mentions | which competitors or categories appear next to you | | NAP consistency | whether your name, address, and contact details match across sources | > **Remember: API does not equal UI.** A model response from an API is not the same as the answer a person sees in a product interface. Personalization, locale, and device can change the result. AI search measurement is probabilistic, not exact. Run each test more than once, then record the date and model. ## Step 4. Find the sources that matter in your region You earn citations in places the AI trusts. Your own site may not be the strongest source. Industry publications, professional communities, marketplace listings, local directories, and review platforms may matter more. Build a current list for your market. Your goal is to place a checkable fact or quotation there, not an advertisement. ## Step 5. Make a 14-day plan - **Days 1 to 3.** Map AI answers from Step 1. Find the questions where you are absent or described poorly. - **Days 4 to 9.** Fix **one** weak lever from Step 2. Add a machine-readable description, price, or example. Prepare two or three specific facts, case results, or quotations for the sources from Step 4. - **Days 10 to 14.** Publish that material where AI systems get answers. Repeat the map and compare your share of mentions and sentiment. Record the date and model. Output: one repaired lever and a baseline for another measurement next quarter. --- ## Next - For a quick self-check, use [Can an agent see me?](../skills/visible-to-agent/). - For agent payments and protocols, see [agent commerce protocols](../watch/agent-protocols.md). For the software a customer agent uses, see [agentic browsers](../watch/agentic-browsers.md). - For the full route, open the [Volume 2 workbook](tom2-workbook.md). ## Sources (verified 2026-07) - **Aggarwal et al., "GEO: Generative Engine Optimization"** (KDD 2024): methods that increase citations in generative answers. - **Gartner:** forecasts for declining traditional search traffic. > ⚠️ AI search measurement remains probabilistic because API results do not equal UI results. --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. --- reading_level: b2-c1-technical title: "Trust compass: two territories, two sides, one move" description: Choose the depth of verification before a costly decision and define what you show in return. --- # Trust Compass: two territories, two sides, one move > Practicum for **Book 3, *The Economics of Trust*** (the Compass house, Chapter 13). > Time: 40 to 60 minutes. The compass does not promise to detect a fake. It helps you choose the depth of verification in advance, name its limits, and avoid checking everything. The compass has four parts: - **needle:** the cost of an error; - **depth:** from no verification to independent confirmation; - **two sides:** how I verify others and what I show when others verify me; - **four norms:** shared rules that nobody has to invent under pressure. --- ## 1. Set the needle before you read the message Start with one question: **What will I lose if I am wrong?** Do not judge how convincing the message looks. The other party controls the presentation. You control the cost of an error. | Depth | When to choose it | What to do | |---|---|---| | **No questions** | The error is small and reversible, such as a video, image, or someone's opinion | Do not verify it. Save your attention for expensive decisions | | **One question** | You plan to cite, forward, or base an opinion on it | Ask, "Where did this come from?" Find the source | | **All three plus independent support** | Money, access, documents, health, work, or reputation | Check the origin, speaker, and owner of responsibility. Move to a second channel that the sender does not control | My three expensive scenarios: 1. Home: ______________________________________________ 2. Work or business: _____________________________________ 3. One more: __________________________________________ --- ## 2. Home territory | How I verify | What I show when someone verifies me | |---|---| | A family check phrase stops a request. It does not authorize a payment | My family knows one stable number or channel where they can reach me | | Every request for money ends with a call I make to a saved number | I have already said, "The real me will not be offended by a check" | | Urgency triggers a pause until independent confirmation | Children and older relatives follow the same agreement and have permission to end the call | | An agent that can spend money has a written limit and action list | I can show my authority and the assistant's limits instead of trying to remember them after an error | One move at home this week: ____________________________________________________________________ For the full conversation and printable card, use the [Family trust protocol](family-trust-protocol.md). --- ## 3. Work and business territory | How I verify | What I show when someone verifies me | |---|---| | I always confirm an expensive instruction through an independent second channel | I have a canonical channel, a named decision owner, and a history people can check | | A junior employee may stop a senior person and receives thanks for checking | Colleagues and clients know who is responsible for what | | A bot identifies itself as a bot. The channel has a human owner, limits, and a clear way to reach a person | The [Honest AI policy](honest-ai-policy.md) explains where a machine works, where a person steps in, and who repairs an error | | A detector gives me a reason to ask questions. It does not make a decision about a person | Drafts, versions, sources, and live references let people verify my work without an oath | One move at work this week: ____________________________________________________________________ For the other side of the table, open the [Trust footprint map](trust-footprint-map.md). If an important file could become part of a dispute, use [The first day with an important file](chain-of-custody.md). You can fill both territories, save Markdown, and print the protocol in the [browser-based Trust compass](../skills/trust-compass/index.md). --- ## 4. End the check with an honest limit Verification rarely produces an absolute "authentic." It should answer a narrower question: Do I have enough support for this decision at this cost of error? | Limit | My note | |---|---| | **Confirmed** | | | **Probable** | | | **Unknown** | | Decision I will make with this limit: ____________________________________________________________________ --- ## 5. Four points of the compass A norm works when people say it aloud and apply it to everyone. 1. **Every family has its own phrase.** "If I call and ask for money, ask for the phrase. The real me will not be offended." 2. **Expensive actions require a second channel.** "We never transfer money because of one message. This rule applies to everyone, including me." 3. **A bot always identifies itself, and a person owns its words.** "A client must know whom they are speaking with. Every bot has a named person responsible for it." 4. **A detector does not pass sentence.** "A risk score from software gives us a reason to ask a question. It is not a basis for a decision about a person." Norm I will state first: ____________________________________________________________________ --- ## 6. Who else reads your trail? Some of the parties checking you are already software systems. A responsible person's name works better than "the team." A date works better than "recently." A number needs a unit. A research claim needs a primary source. - [ ] My important name, role, and specialty are stated consistently on the main pages. - [ ] My case studies include dates, source figures, and a clear result. - [ ] My main claims link to primary sources. - [ ] I document errors and corrections instead of hiding them. Compare two AI answers about you. A difference is not a verdict. It points to a missing or unstable source. For current context, see [the machine web](../watch/machine-web.md). --- ## The first ninety minutes 1. **30 minutes:** List the three places where a mistake would cost you most. 2. **30 minutes:** Create independent support for the most expensive scenario. Record whom you call, which saved number you use, what you ask, and what you do if nobody answers. 3. **30 minutes:** State one norm to your family or team. 4. **Two more minutes:** See what one person and two different AI assistants can find about you. > The useful result is not "I verify more." It is **"I verify somewhere else."** > The compass makes deception more expensive. It does not make anyone invulnerable. If you want one final artifact instead of notes from each section, complete the [one-page trust protocol](tom3-workbook.md#my-trust-protocol-on-one-page). For a completed composite example, see [Marina and her family](../cases/persona-family-trust.md). --- > **Draft translation, not author-reviewed.** The Russian edition is the source of truth. --- name: trust-compass description: >- Helps choose the depth of checking in advance from the cost of being wrong rather than from how convincing a message looks: no questions · one question · all three plus independent confirmation on a second channel. Produces a trust protocol for home and work. From chapter 13 of "The Economics of Trust". Use when the user asks whether a message, call, document or person can be trusted, how much checking is enough, or wants a rule for their family or team. Triggers "can I trust this", "how do I verify", "trust compass", "trust protocol", "I got a strange message". license: MIT --- # The trust compass (Practicum for "When Intelligence Became Cheap") 📋 A ready-made skill for an AI assistant. Drop this file into your agent's skills folder — or copy the block below into any chat. The compass does not detect fakes and does not decide whether a person, voice, image or file is real. It helps you name **what checking you decide in advance is enough** when being wrong is expensive. The worksheet this comes from, for filling in by hand: [The trust compass](../../playbooks/trust-compass.md). ```text Help me build a trust protocol using the trust compass from "The Economics of Trust" (chapter 13). The core rule: the depth of checking is set by the cost of being wrong, not by how convincing the message is. The other side controls how convincing it looks. I control the cost of being wrong. Three depths: — No questions: the mistake is small and reversible (a clip, an image, someone's opinion). Do not check; save the attention for what is expensive. — One question: I am about to cite it, forward it or form a view on it. Ask "where did this come from?" and find the source. — All three questions plus independent confirmation: money, access, documents, health, work, reputation. Check the origin, the speaker and the accountability — and step onto a second channel the sender does not control. Collect from me if I have not given it: 1. My three expensive scenarios: one at home, one at work, one of any kind. 2. Who usually gets in touch in them, and through which channels. What to do: 1. Assign a depth to each scenario and justify it by the cost of being wrong, not by how convincing the approach looks. 2. For the deepest scenarios, name a specific second channel the sender does not control. Not "call the number in the message" but "call the number I already had". 3. Help me separate three states honestly: confirmed · probable · unknown. "Probable" is not "yes"; spell out what I did not check. 4. Propose one household agreement and one work agreement — short enough to say out loud and remember in a hurry. 5. Finish with ONE move for the coming week. Discipline: - Do not rule that something "is fake" or "is genuine": you have no grounds for that. - Do not tell me to check everything. Attention is finite, which is the whole point of the compass. - Do not offer AI-detection services as proof. - Say separately what I will present when it is me being checked. Answer format: 1. Table: scenario | cost of being wrong | depth | what I actually do. 2. The second channel for the expensive scenarios. 3. The honesty line: confirmed / probable / unknown. 4. Two agreements: home and work. 5. One move for the week. ```