THE GOVERNED OUTCOME RECORD From "AI for Your Business — Advanced", chapter 2 ================================================================ Your archive is a record of what HAPPENED. What you need is a record of what WORKED. They look almost identical and they are not the same thing at all. Every business records what happened -- the law insists on it. Almost nobody records the judgment that sat on top: did that turn out right? A rival can buy your tools this afternoon and, given a year, accumulate the same volume of what-happened. What they cannot get at any price is your what-worked, because the only way to produce it is to have done the work and formed the judgment. Set this up in under an hour. If it takes longer, you have built the wrong thing. There is no database here, nothing to migrate, no archive to clean. ⚠️ DO NOT BACKFILL. Reconstructed judgments are guesses wearing the clothes of evidence. Start Monday with entry one. 1 - NAME THE ASSET ---------------------------------------------------------------- When your selected edge works, what is it drawing on that a rival hasn't got? Not the tool. Not the people exactly -- people leave. The thing underneath that accumulates. The thing underneath my edge that a rival hasn't got is ______________________________________________________________ ⚠️ If that sentence contains the name of a piece of software, it is wrong. Go one level down and ask "and what's underneath that?" until you stop reaching for a tool. If you're not sure, find your candidate in this table: YOU SELECTED THE ASSET IS USUALLY --------------------------- ----------------------------------------- A record only you have which of your judgments proved right -- not the files themselves Decisions that compound the allocation or call itself, and what followed it Switching costs you've which commitments you actually met, and earned where you failed An offering only you can the method -- which approach settled which make kind of problem 2 - DEFINE WHAT "WORKED" MEANS (before you capture anything) ---------------------------------------------------------------- If you don't write this, five people apply five private standards and after three months you have a record you can't trust. It must be OBSERVABLE -- checkable by a second person, from records that already exist. "Worked" means: ______________________________________________ ______________________________________________________________ Good: "the client acted on it without needing a second explanation" "the job closed on the first visit, inside the window we promised" Bad: "the client was happy" (nobody can check it) Read it aloud to whoever will be recording. If they say "well, what about when...", that's your first exception, not an objection. Log it and move on. 3 - THE TWO MOMENTS (both must already exist) ---------------------------------------------------------------- What you judged and what actually happened almost never occur at the same time. So capture at two points -- both of them moments the work already pauses. THE DECISION MOMENT (fields 1-3, typed by the person who just made the call, while they still hold it in their head. A judgment reconstructed a week later is a memory of one.) Ours is: ______________________________________________________ THE OUTCOME MOMENT (field 4 -- bolt it onto a pause that already exists: the job marked closed, the file archived, the invoice going out.) Ours is: ______________________________________________________ ⚠️ The commonest way this dies is a NEW ritual -- a Friday review, a weekly log, a form somebody has to remember. That survives about four weeks. If you can't find an existing moment for either half, that's a signal about the edge, not about the record. 4 - FOUR FIELDS -- AND NO FIFTH ADDED IN THE MOMENT ---------------------------------------------------------------- 1. The situation, in a few words 2. What we did 3. What we judged at the time 4. What actually happened <-- the outcome link. THIS is the asset. Every field past four cuts compliance, and a record kept 40% of the time is worse than useless -- you can't tell whether the missing entries are the failures. Worked example, two firms: SITUATION first-year company, | boiler swap, tight loft director's loan | access, Tuesday WHAT WE DID sent the standard | sent Danny, booked 2 hrs, explainer with figures | took the long ladder WHAT WE "clear enough -- they've | "he's done three like it, JUDGED had one before" | should be one visit" WHAT ACTUALLY came back with two | closed first visit, HAPPENED questions on the balance | out by 3.40 Two sentences at the decision, one at the close. That is the whole instrument. The fields are not frozen for ever -- you will find something missing, and you should. But they change ONLY AT A REVISION, deliberately, with the authority of the rule from chapter 3. Never mid-quarter because somebody had a good idea on a Tuesday: change the instrument halfway through and you cannot compare anything either side of the change. 5 - FOUR ROLES, FOUR PEOPLE ---------------------------------------------------------------- This is where firms quietly cheat, and the roles are what turn a personal habit into something the business owns. CANDIDATE OWNER ______________________ owns whether the edge works. Does NOT record and does NOT check -- nobody marks their own homework. RECORDER ______________________ does the work, makes the call, logs at the decision moment, and CLOSES OUT field 4 at the outcome moment. CHECKER ______________________ no stake in the answer. Samples sampling ____ per ____ against the written definition, and first sampling date: looks for entries that never got ______________________ closed. An orphaned row -- a judgment with no outcome -- is what kills this quietly. YOU owns none of it, reads the exception log. ⚠️ Four roles is why a sole trader cannot do this. That isn't a boast -- it is the reason a small team can build something a solo competitor can't. 6 - EXCEPTIONS: LOG THEM, DON'T ARGUE THEM ---------------------------------------------------------------- When a case doesn't fit the definition, log it as an exception rather than forcing it into a box. Nobody rewrites the definition on the spot -- changing the yardstick mid-quarter destroys the comparison. Log it, leave it, move on. The authority to revise belongs to the rule you write in chapter 3. 7 - THE CAPTURE BOUNDARY (close chapter 1's permissions question) ---------------------------------------------------------------- Chapter 1 gave your candidate a status. Now write the boundary -- the specific fields you may record -- in one line the recorder can actually follow. Status: [ ] clear [ ] irrelevant [ ] restricted pending advice WE MAY RECORD: ______________________________________________ WE MAY NOT RECORD: __________________________________________ Routed to (if restricted): __________ Expected back: __________ Lawful basis for holding it: _________________________________ How long we keep it: __________ Who may see it: _____________ Responsible for deleting it on request: ______________________ Three routes exist where client material is involved: ASK properly (not boilerplate buried in the small print); AMEND at renewal; or RING-FENCE -- record only what is unambiguously yours and name what's excluded. ⚠️ Where the status is restricted, capture starts ONLY on the fields you have explicitly listed as approved above -- never on "everything except" the excluded list. If a field isn't on the approved list, it isn't captured. Name the excluded material explicitly too, so nobody has to guess on a Tuesday afternoon. ⚠️ Not legal advice. Your professional body, your contracts and your own adviser decide what binds you. This sheet only asks whether you could SHOW someone, in writing, what you're allowed to reuse. 8 - START DATE ---------------------------------------------------------------- We start (forward-only, no backfill): ________________________ By your go/no-go date you'll have somewhere around a hundred judged outcomes. That is not a training set and was never meant to be. It is enough for one specific job: enough for a person to see one repeating pattern clearly enough to write a rule from it. A hundred cases somebody actually judged will show you more than ten thousand nobody did -- because the ten thousand don't say which ones went well. ePORTS — your ecom source. eports.global