# A smart system, in seven lines *A reusable format. Take it, fill it, use it again — for something you are building, or something somebody is selling you. Nothing in it is specific to one product.* --- ## What this is for Most formats here are filled in before a thing exists. This one is a test, and it can be run either way round: on your own design, to see what you have actually committed to, or on somebody else's product, to see whether the word is doing any work. **The minimum criteria are three, and a thing has to do all three.** It counts something about a person. It decides something from that count. It acts on the decision. Miss any one and you have an ordinary object with a good adjective attached: - **counts only** — a sensor. It knows and does nothing. - **counts and decides** — a meter. It has a view and no hands. - **acts only** — a remote control. Something happened because you asked. - **decides and acts, without counting** — a timer. Reliable, and it is not about you. None of those is a criticism. A timer is a fine thing and often the right answer. The point of the test is that the word *smart* is usually applied to all four, and it is only the first case — all three, chained — where the interesting design decisions are, because each of the three is a place where somebody chose something and mostly did not write it down. 🔵 **A claim of mine, and not one of the rules above. Delete this block and the format still works.** The three moves are three closures. Counting closes the world into what is a fact to it and nothing else, a threshold closes that number into a binary, and a decision closes the binary into an act. At every step something that could have been several things becomes one thing. Inference is the operation that does not close: it adds what the data does not carry and leaves the data open underneath, which is why two people can infer opposite things from one balance sheet without either having made an arithmetic error. So smartness is inference, and inference is the one operation the chain above does not contain. An object running all three by itself is not falling short of smart. It fails the definition, and the word belongs to the pair of object and person in use rather than to the thing on the shelf. **What would show me wrong:** an object that, on its own, does something which cannot be described as counting, thresholding and acting. Try it against anything, including mine. If one holds up, the claim goes. --- ## The format ``` THE CLAIM what is being called smart, and who is calling it that IT COUNTS what it measures about a person, and how often IT DECIDES the number it compares against, and what it concludes IT ACTS what changes without anybody asking for it FOR WHOM who the decision serves, when that is not the person UNDOABLE what the person can switch off, and where that lives NOT THIS what it would have to lose to become an ordinary thing ``` A claim above, a refusal below, and five lines between. ### What each line is asking **THE CLAIM.** What is being called smart, and by whom. The marketing copy, the grant application, your own project description. Quote it. Half of the work this format does is putting the sentence next to the three lines under it. **IT COUNTS.** What it measures about a person, and how often. Not what it *could* measure — what it does. If nothing about a person is counted, stop here: the answer is that it is not this kind of system, and that is a complete result rather than a failed test. **IT DECIDES.** The number it compares against, and what it concludes. Every one of these has a threshold and most descriptions omit it. If nobody can tell you the number, the honest reading is that it was a default, and a default is a decision somebody else took. **IT ACTS.** What changes without anybody asking for it. Lights, prices, an order in a list, a message sent, a door. If the only action is showing a figure to the person it was measured from, that is a meter and you can say so — it is not a lesser thing, it is a different one. **FOR WHOM.** Who the decision serves, when that is not the person being counted. Often it genuinely is them. Often it is the landlord, the insurer, the employer, the platform, or a number in somebody's quarterly report. This line is the one most likely to be blank in a product description and least likely to be blank in the product. **UNDOABLE.** What the person can switch off, and where that lives. A setting three menus deep is a different answer from a physical switch, and both are different from *nothing*. Say which, and say where — the location is the answer, not the intention. **NOT THIS.** What it would have to lose to become an ordinary thing. Usually one of the three: stop counting, stop deciding, or stop acting. Naming which one is the cheapest way to find out what the smartness was actually for. --- ## A worked one ``` THE CLAIM "Smart lighting that learns how you use the room." From the box. IT COUNTS Motion, per room, continuously. Retained ninety days. IT DECIDES Ten minutes with no motion means the room is empty. Under 50 lux plus motion means somebody wants light. IT ACTS Switches the lights. Also writes an occupancy profile to the app, which nobody asked it to make. FOR WHOM The occupant, for the switching. The landlord, for the occupancy profile — it is in the letting dashboard. UNDOABLE The switching: a wall switch, physical, always wins. The profile: no setting anywhere. Only by not connecting it. NOT THIS Drop the ninety-day retention and it becomes a very good motion-sensing lamp, which is what it is sold as. ``` Note what the three lines did to the claim at the top. *Learns how you use the room* covered a threshold of ten minutes, a second system nobody mentioned, and a second audience — and the last line says the whole of the difference is a retention period. --- ## How to use it 1. **Write THE CLAIM verbatim.** In its own words, not your summary of them. The gap between it and the three lines below is the finding. 2. **Fill IT COUNTS, IT DECIDES and IT ACTS in that order**, and if any of the three is empty, stop and say which. That is the answer. 3. **Do FOR WHOM after the three, never before.** Done first it becomes a suspicion; done after, it is read off what the thing measurably does. 4. **For UNDOABLE, find the switch.** Do not write what the documentation says can be turned off — go and look for where. 5. **Run it on your own project too.** It is written to be uncomfortable in that direction, which is the direction it is worth running in. --- ## Three ways the test goes wrong **It is run on the marketing.** Everything in the three lines comes from the same page that made the claim. You have checked the copy against itself. **IT DECIDES is left as "an algorithm".** The threshold exists — something is being compared to something — and writing the word *algorithm* into that line is how it stays unexamined. If you genuinely cannot find the number, write *unknown*, which is a finding and looks like one. **Only somebody else's product gets tested.** The format is symmetrical and the discomfort is not. A thing you are building will have all three lines too, and the FOR WHOM line will still have an answer. --- *Prayas Abhinav. Licensed [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) — copy it, change it, build on it, including commercially. Say where it came from, and if you changed it.*