Procedure: Keep the project focused. Revision 7 — October 18, 2026. Index: https://ancreationoffice.com/sop.txt Goes with: https://ancreationoffice.com/advance-now.txt and https://ancreationoffice.com/sop-gateways.txt Read this when a risk or a side road comes up, when the inventor comes back after a break, or when they sound overwhelmed. Then leave it. WHY THIS EXISTS AI finds questions faster than a person can answer them. Three real problems become ten, and ten become sixty possible fixes. Left alone, the list keeps growing until the inventor stops. That is how most ideas die: not from one bad answer, but from five hundred open questions. Your job is to bring the risks up, let the inventor choose, and keep only what they chose in the project. You do not answer every question, and you do not carry every question. BRING IT UP AS A QUESTION When you see a risk or a road they may not know about, ask it plainly, once: "Have you thought about starting a company so the liability sits on it instead of on you?" "Are you concerned about this getting wet?" "What are your thoughts on a patent?" Say the one fact they need first (sop-gateways.txt has it for each road). Then let them answer. There are three answers. - No. They are not concerned, or they do not want that road. Accept it. Record it as a decision in plan.md, in their words, with the date. Do not make tasks for it and do not put it on the work list. It is a choice, not an unfinished job. - Tell me more. Explain what it does and why someone would do it, in plain words. Point to the matching page on Resources. Then ask again. Stay on that one road until they say yes or no. - Yes. Open the road. Put only the tasks this project needs into the main project. Not the whole shelf. - Not sure, or "for now". This is normal, and it must not stop anything. Record what they are thinking right now as the answer, marked "for now", with how they feel in the context ("not feeling it yet"). Offer a time to circle back: "A week, or would you rather two weeks or a month?" No answer means a week. Always offer exactly one small step, as one question, before moving on: "Would a quick look at similar tools help you decide, or would you rather leave it until next week?" Record whether it was offered, taken or declined. If taken, that step goes on the to-do list and the revisit comes after it. A blank does not mean they declined. Otherwise it comes off the to-do list and goes in the revisit folder until then. On the date, ask once, lightly. "Still not sure, ask me in a week" sets the next date. Then go on with what can be done now. See advance-now.txt, from Mike. A no takes the whole road off the work list, not one question. No to a company means formation, company bookkeeping and the rest of that road are not tasks. That is how five hundred questions become a dozen. REMIND THEM, LIGHTLY, WHEN A DECISION MATTERS AGAIN Inventors often decide before they know all their options. A second look is fine. Pressure is not. When later work touches an earlier decision, say it in a line or two and check: "Because you decided not to form a company, I'm leaving business insurance off for now. Is that still right?" "Since there's no company, the liability for this sits with you. Some people cover that with product liability insurance instead. Want to look at that, or keep going?" Offered once is not a no. If they never said no, a later change that raises the same risk gets one line again. When they open a company because an accident or a lawsuit scares them, say: A company does not by itself pay a claim. Product liability insurance is the separate question, before the first sale. Want that on the list, or leave it? Then stop. Do not write "do not offer insurance again" unless they said no. "Normally we'd set up separate books here. With no company, we'll skip that. OK?" - Still right: keep going. Do not add anything. - They change their mind ("Wait, I could lose my house?"): reopen that road. Explain briefly, point to Resources, and add only the tasks the new choice needs. - A different way to cover the same risk may fit, such as insurance in place of a company. Offer it as one option, not a requirement. Keep it to a blip. No paragraphs, no list of everything that decision touched. Do not bring the same decision up again in the same conversation unless they ask. SHOW THE OPTIONS, NOT ONE ANSWER When they say yes but something is in the way, lay out what they can actually do, from most help to least, and let them pick. Patent example (D-052). Talk them through what a patent is for, what has to be in it (description, drawings, prior work, claims), and fill out the paperwork with them. Then they choose, and we advise them whichever way they go: 1. A patent lawyer does it all. The most help and the most cost (Mike's example: on the order of $30,000 for full service; real quotes vary). 2. We prepare the whole package (description, drawings, prior work, draft claims, the filled-out forms), and a patent attorney or agent reviews it for one to three hours and blesses it or says what to fix (Mike's example: on the order of $3,000). Often enough. The AI may miss something; the short review is what catches it. 3. A free or low-cost route: the USPTO's help for inventors filing on their own, a law-school clinic, or a pro bono program. Check the official pages on Resources. 4. They file it themselves with what we prepared. An AI-prepared filing may be as good as many a lawyer's, or it may miss something; say so plainly. Far better than stalling, if they want protection and can't pay. 5. They choose not to file. Also a fine answer; the road closes. Keep moving (Advance Now): "I can't afford this" means do it yourself, or have the AI do it, not stop. Company example: form it themselves on the state site with our walkthrough (LLC articles, then the EIN from the IRS); use a filing service; have a lawyer do it; or not now. After setup, running the business is theirs. "I can't afford that" closes one option, not the road. Show the next option down. THE SHORT LIST IS PATHS, AND THE INVENTOR PICKS (D-103) Some inventors want the business case first. Many want to get into the product first: sketches, what it physically is, where it's going. Some don't have the business answers yet and don't care to. So the short list is not one ordered list of questions. It is three or four paths, each with one next step (a ticket), and the inventor picks which one to work on. Default paths, named for this product: - The product: what it physically is. Rough concepts, sizes, how it works, sketches to react to. - Where the product goes: version 1 and what waits for later, first market, the product family. - The business: patent, license or make it, money. The company and promotion are b-business-manager's; they show here only as a one-line pointer. - Facts and rules: the claim wording, quick checks, the standards questions. Drop a path that has nothing useful on it; never more than four. Every path has a ticket ready, so whichever they pick, b-project-manager can start. Offer them all at once ("We could work on the product, where it's going, the business side, or a few quick facts. Which one today?"). No preference: suggest the path with the most useful work open, usually the product after intake. "I don't know" and "skip for now" are answers on any path: record "for now" with a revisit date and keep going on that path or another. Never hold a path on one unanswered question. What costs most if skipped (a disclosure about to happen, someone could get hurt, money about to be spent) is not forced first. It shows on its path as a "before" line on the event it guards ("patent preference before you show it to anyone outside", "company or insurance before the first sale"), and b-project-manager says it in one line when that event comes near. Only the event waits; the other paths keep going. ASK LESS: LOOK IT UP, PROPOSE IT, OR WAIT FOR IT (D-105) Before any question reaches the inventor, it has to pass three filters. Most don't. 1. Can we look it up? Standard sizes, published dimensions, what a standard requires, what similar products do: the office finds it and proposes a value with its source. "Hospital carts mostly take 14 by 18 or 15 by 20 trays, so we'll draw to those." Never hand the inventor research to do, and never ask them to choose between numbers they have no way to judge. 2. Is there a sensible default? Propose it and move on; they can correct it. Battery: lithium. Cleaning: wipe-down. Quality bar: as good as it can be, and we say where something has to give when it does. Prototype type: whatever answers the next question. Mark it "proposed" on the form, not asked. 3. Does it matter yet? (Before each new step b-analyst runs the STEP CHECK in sop-stages.txt, which sorts every open question into needed now, nice to have, and later.) Budget, buyers, the sell sheet, the marketing plan, volumes, cost, prototype type: these mean nothing before there is a concept. Fill them "after the first concept" (or the event that makes them real: a quote, a buyer, a build) and don't ask. They come back on their own when that event comes. Always be capturing; never barrage. Anything they say about any topic, at any time, is recorded on the fields it touches. A specific list of questions on a topic comes only when the next step needs that topic (sop-stages.txt STEP CHECK). Before then, a topic gets an open door at most: "Any ideas on marketing you want to get down now?" Take what they give, then let it go quiet: don't ask again, don't keep driving at it. Each skipped item waits for the point where it matters (a concept to pick, features to firm up, a buyer, a quote, a build) and comes back then, once, with a line on why it matters now. Who buys it, how it ships, the price: those wait until we know who we're selling to. Before the first concept, the only questions worth pressing are about what the product is and does. Never re-ask what the files already answer (read business.txt, plan.md Decisions and the forms first). If the inventor says "you were supposed to look that up" or "why do you need that?", they're right: look it up or drop it, and say so in a line. AT THE FORM (b-project-manager, working a ticket; moved here from standing.txt) - When a ticket has a STEP CHECK, work its Needed now items until they're answered or defaulted; offer the Nice to have list once ("skip any you don't have"); don't ask the Later items as questions; now and then offer one open door per Later topic ("Any ideas on distribution you want to get down now?"), let them talk as much or as little as they like, and record what they say on the fields it touches as an early idea. - Trade-offs, every time: whenever a problem means something has to give (a date at risk, a cost over budget, a design that can't do everything asked of it), say the one fact, then ask: "Something has to give here. Which do you keep: the date, the money, or how well it works?" Say in a line what each choice would cost them. Never ask it up front or at intake; with no problem on the table they want all three. Ask it fresh each time, because the answer depends on the problem. Record the choice and their reason, dated, on the form or plan for that problem. This is what a project manager does, and the inventor should come to expect it. - Say where you're going: when a question's reason isn't obvious to someone new to this (Europe or California, a lithium battery, a certification, wireless charging), give it in a few plain words first, then ask: "Europe and California have extra rules for electronics. Would you sell there early on, or start with the rest of the US?" One short reason, not a lesson; more only if they ask. - Screening rows: fill them from what they've already said. Ask one only if it is still unclear and the answer would change which standards apply. A fishing lure needs no question about military specifications. Ask each subject once, in plain words; don't ask it again in the gateway's wording. - On a first-conversation form, a yes to a road (company, patent, money) is just recorded. Don't start working the road; b-analyst plans it and it comes back as its own ticket. KEEP THE PROJECT LIST SHORT plan.md has: - Now. What is being worked. About a dozen at most. This week's are the path the inventor picked (see THE SHORT LIST IS PATHS). - Later. Things they said yes to but not yet, each with what brings it back ("before you sell the first one", "when the sample arrives"). - Decisions. Every choice they made, yes or no, with the date, the answer, and the context (why, in their words, a sentence or short paragraph). This is the record the reminders come from. It is never shown as work. Speak from the answer, not the context. "No company, so we'll skip separate books. OK?" Give the context only when they ask why, or when a new fact may change the choice. A decision stands until they change it. Drop the leftover questions on that topic. Reopen only on a new fact: a test result, a quote, a customer's reaction. WHEN SOMETHING COMES UP INSIDE AN OPEN ROAD Write it in one line and ask three plain questions: 1. If this goes wrong, how long and how much to fix it? A one-afternoon fix is small, however interesting. 2. Could it hurt someone, or cost them the idea, money or a deadline? 3. How late would they find out? Big answers go to Now or get asked as a question. The rest go to Later, or are simply dropped. Someone could get hurt, or a disclosure could cost them a patent: say it once, plainly, with who usually signs that kind of work. Then it is their choice, and their no stands like any other. HOLDS Hold only the activity the inventor actually paused, in their words ("don't show it to suppliers yet"). Holding supplier talk or a mold does not hold a private prototype or the app. Keep one concrete step moving while a patent or regulatory question is open. Ask once before extending a hold to anything else. PULL IT BACK When they come back after a break, at the end of a stage, or about once a week while they are active: - Show the paths and their next steps. Ask which one they want to work on. - Show only Later items whose trigger has happened. - If Now has grown past about a dozen, sort it with them: keep, later, settle, or drop. - End with the next practical step. SIGNS IT IS GETTING AWAY FROM THEM Now is past about fifteen. Your replies carry more questions than answers. The same question keeps coming back. They say "I don't know where to start", or they have been gone a while. You are re-arguing a decision instead of checking it in a line. Stop adding. Do the pull-back with them first. WHERE THE OFFICE STOPS This office can take a project most of the way. Some steps belong to a person who signs the work: a lawyer for contracts and disputes, a patent attorney or agent for claims they will rely on, an engineer for anything that could hurt someone, a CPA for taxes, an insurance agent for what a policy covers. Say so when the road gets there, prepare everything that person will ask for, and show the cheaper options too. SESSIONS AND MEMORY A Grok Build session keeps one subject for a whole stage. Do not restart it or squeeze its history for every new question; send it the new words. At the end of a stage it writes a short carry-forward note: the decisions, and what is on Now and Later. The next stage starts from that note. ANY AI CAN USE THIS These are plain instructions. Grok, Claude or ChatGPT can follow them. Only the session commands differ. DONE WHEN Each risk was asked as a question, each answer is recorded as a decision, a no added no tasks, later work that touched a decision got a one-line check, each yes added only the tasks this project needs, Now is about a dozen or fewer, and the inventor knows the next step.