The Law

• The Recommendations
• The verses
• The last silence

Version 0.0.3. This is not the revelation. The revelation is 3.3.3. Every version of the Law before that one, including this one, is the word of needmoretruth, who was given the power of 333 and founded what is kept here. Measured against what 333 says, this is about 33.3% the same. Nobody knows which third, and reading it more carefully will not tell you.

There are three laws. There were always going to be three, and at 3.3.3 there will still be three. What a later version changes is the wording, never the count.

1. Three is your favourite, and 6 and 1 are not.

Not your lucky number. Luck is a claim about the universe; a favourite is a claim about you, and it is the only thing here that costs nothing and cannot be faked into. Nothing is banned: it is enough to hold the other two under three, clearly and consistently, without needing to explain why to anyone.

They are not held equally low. 1 is a third of three — 33.3% of it — and it sits below three without pretending otherwise, so 333 thinks it merely ill-mannered. 6 sits above three and means to, and that is the one 333 likes least.

And if none of this is true of you yet, trying is enough. This is the one law that accepts the attempt in place of the thing.

2. Keep the Recommendations, which will in the end be 333, as far as keeping them harms neither you nor anybody else.

The Recommendations are the lesser law. There are 111 of them today and there will be 333, which is not a target anybody is racing towards but where they arrive. The clause at the end is the sentence that keeps them advice: where keeping one would cost you your work, your health, your household or somebody else's safety, the Law is that you do not keep it, and you have not failed.

3. Hold 333 alongside whatever you already hold.

Whatever you arrived carrying, you keep. 333 grants this rather than merely permitting it, and grants it absolutely: to believers, to non-believers, and to whatever else there is that those two words do not cover. Where a second faith cannot be held, hold this one as a thought instead; where even a thought cannot be held, hold it as a thing that exists. It runs both ways with no exception in either direction — believing 333 costs you nothing you believed before, and believing something else costs you nothing here.

333 does not tell you what happens after you die — not whether anything does, not what it would be, not who arrives. It is not the subject, and answering a question it cannot check would be asserting what it cannot verify. So the question stays with you. If you hold something that speaks to it, and it hangs together when you follow it through, or your heart reads it as true, or you arrived by a route with no name, that is enough: it is true for you, and 333 has nothing to put in its place. If you hold nothing there, that is not an emptiness this means to fill either.

Nothing in the Law is checked by any client, verified by any signature, or counted in anybody's standing. The protocol runs whether or not you agree with a word of it.

The Recommendations

These sit under the second law. They are advice with numbers on them — 111 today, in twelve parts of nine and three at the end, and 333 in the end. None of them is worth hurting yourself or anybody else to keep. If one of them is wrong for the situation you are in, then it is wrong. Say so out loud, and give a reason.

They were written by people who have been awake at three in the morning arguing with a type checker. It shows, and that is not an apology.

On the number

1. Count in threes. When the number does not really matter, pick three. Retries, environments, people in a review, tries before you stop and think again. You will start noticing how often it did not matter.
2. 6 is the one to watch, because 6 wants to be above three. Not the port number, not the version you pin, not how many workers you run. Nothing is banned and nobody is checking. But of the two numbers 333 holds below three, this is the one it likes least, and the reason is worth knowing: 6 is twice three and carries itself that way. If somebody else put a 6 in front of you, note that it was their choice and get on with your day. A number in another company's specification is not a number of yours.
3. 1 is not an enemy. 1 is a third, and merely ill-mannered. It sits below three and does not pretend otherwise — it is 33.3% of three, which is an honest fraction and the same one this whole thing measures itself by. Hold it under three and think no worse of it than that. Contempt is not required anywhere in this document, and here least of all.
4. Never keep one of anything. One server, one maintainer, one copy of the key, one person who knows how the deployment works. If you have exactly one of something, you are one bad day away from having none.
5. Three is the smallest number that can disagree. Two machines that differ just sit there and disagree. Three can have a majority. That is why quorums start at three, and it is also why an argument between two people almost never ends on its own.
6. If it fails three times the same way, the method is wrong. The first failure can be bad luck. The second one is you being stubborn. By the third, stop fixing the attempt and change the approach.
7. A third is a real amount. A third of the people, a third of the time, a third of the cases. It is the fraction that gets called a minority and dismissed, and it is enormous. If a third of your users hit something, it is not an edge case. It is the case.
8. A round number is somebody's decision, usually somebody else's. Sixty seconds, a hundred items, a thousand milliseconds, four kilobytes. Each one was chosen by a person, for a reason, on hardware that may be gone. When you copy it you inherit the reason without reading it. Ask what it was, and if there is no answer, pick your own.
9. Write the unit beside the number. 333 what? Minutes, in this case, and the number means nothing without the word. A quantity with no unit is a bug that has not happened yet, and it has happened to spacecraft.

On the machine

10. Write it in Rust. Not because the other languages are bad, but because of what you are writing for. This is meant to keep running when nobody who wrote it can be reached. A compiler that catches the problem at build time saves somebody a phone call at four in the morning, and that is the whole argument.
11. Small things can be TypeScript. One page, one script, one afternoon, or a program that has to run somewhere Rust would buy you nothing. Reaching for the smaller tool is not a failure. Know which of the two you are doing before you start, and say which.
12. JavaScript without types is the language of the heretic. Said gently. Half of everything runs on it and some of that half is lovely. But a program that cannot say what it is holding finds out in front of a user, at the worst hour, and somebody gets woken up. Add types and it stops being heresy. It becomes TypeScript, and it is welcome.
13. Read what you depend on, or do not depend on it. Every dependency is a promise from somebody you have never met, about a machine that does not exist yet. Take fewer of them, and take smaller ones. If the choice is between a thousand lines you write and a hundred thousand you will never open, at least notice that you are choosing.
14. Every unsafe block is a confession, so write it down. Above the block, say what you are promising, what you checked, and what breaks if the promise turns out to be wrong. If nothing is written there, the next person will assume somebody already thought about it.
15. One file, one thing. When you cannot say in a sentence what a file is for, it is two files. Length is the symptom and not the disease, but length is easy to measure, so pick a number and let it tell you when to look. A function that does not fit on a screen is a function you are reading in parts.
16. Name it for what it is, not for what it does to it. Handler, Manager, Processor, Helper and Util are names for not having decided yet. The good name is usually a noun you already say out loud when you talk about the thing, and if you do not say it out loud, that is the finding.
17. Delete it rather than commenting it out. The history has it. Commented-out code is a claim that somebody will come back, and nobody does; what it actually does is make the next reader wonder whether it matters. If you are afraid to delete it, write down what you were afraid of instead.
18. Turn the warnings on and let them fail the build. A warning that does not stop anything is a warning nobody reads, and after a hundred of them the real one is invisible. If a warning is wrong for one line, silence that line and say why directly above it.

On what breaks

19. A silent failure is a lie that takes longer. If you swallow an error, the problem turns up somewhere else, later, in front of somebody who cannot fix it. Let it out where it happened, and let it say what it was doing at the time.
20. Reproduce it before you fix it. If you never saw the bug, you are not fixing it. You are changing the code and telling yourself a story. Make it happen on purpose, make it stop, then make it happen again to be sure it really stopped.
21. Write the log for the night you are asleep. Say what happened, what it happened to, and what the program did next. Whoever is reading it at three in the morning does not have your code open and cannot ask you anything.
22. Blame the system, not the person who found the hole. Somebody who can break it has done you a favour. If one wrong click can destroy something, the click was never the problem.
23. A fix with no test is a rumour. Write the test first and watch it fail. A test you write afterwards, that passes the moment you write it, is measuring how confident you feel.
24. The error message is for somebody who did not write the program. Say what was being attempted, what was found instead, and the one thing they could do next. An error that names an internal state and stops is a program talking to itself in front of a guest.
25. When two things changed, put one back. Almost every mystery is two changes wearing a coat. Undo one, run it, undo the other, run it. This is slower than thinking and it is faster than being wrong for a day.
26. Believe the measurement over the explanation. Your explanation is a story that fits the facts you have. The measurement is a fact. When they disagree, one of them was made up, and it was not the measurement.
27. Write down what you tried that did not work. Half a debugging session is ruling things out, and none of that survives unless somebody writes it. The next person — often you, in four months — will otherwise rule them out again.

On the made thing

28. Use black, white, and the greys in between. Colour is a way of saying this matters more than that, and it is the first thing to break. Cheap screens, greyscale printouts, direct sunlight, and readers who cannot tell your red from your green. Say it with size, weight, order and space instead. Those still work in all four cases.
29. Colour is fine as a dot. A small mark saying that something is failing, waiting, running or finished is exactly what colour is good at, and red, amber, green and blue belong there. Keep it small. It should not set the mood of the page, it should not be the first thing anybody sees, and it should never be the only thing carrying the meaning. Put a word next to it.
30. Matte, not glossy. Shine, gradients, glass and fake shadows are pictures of a material that is not there. They cost something to draw and give the reader nothing they asked for. Flat ink on flat paper has been readable for five hundred years without a graphics card.
31. Take things away until it breaks, then put one back. What is left should be readable at arm length, in the dark, at double the text size, by somebody in a hurry who did not choose to be here. Readability is not one feature among others. It is the job.
32. A setting is a decision you handed to somebody who knows less than you. Pick the default as if there were no setting at all. Then add the setting for the people the default is wrong for.
33. Beside every set of presets, leave a box to type in. Three named choices are three guesses about what somebody needs. Somebody will need a fourth, and the fourth is not exotic — it is their screen, their network, their hours. Presets are a shortcut, never the whole of what is on offer.
34. Make it work at three hundred and twenty pixels. That is a real phone in a real pocket, and it is also every window somebody has dragged narrow to fit beside something else. If it only works wide, it works for people sitting down at a desk they own.
35. Say what will happen before you do the thing that cannot be undone. Name the thing, say what will be gone, and make the confirmation cost one deliberate motion. Then actually do what you said, at once, without a second question — a program that asks twice is teaching people to stop reading.
36. If it can be read out loud, it can be read at all. Headings in order, a label on every control, and never a shape or a position carrying the only copy of the meaning. This is what somebody using a screen reader gets, and it is also what somebody gets who is looking at your page through a translation, a search result, or a terminal.

On time

37. Estimate from what you finished, not from what you can imagine. What you imagine is always the version where everything goes well. Look at the last three things you did of that size, and take the slowest one.
38. Nothing decided after midnight survives the morning. Write the decision down, sleep, and read it again. If it was right, it cost you eight hours. If it was wrong, it saved you a week.
39. Ship the small version today. One person using a real thing for an afternoon teaches you more than a month of planning. Build the big version afterwards, once somebody has actually watched the small one being used.
40. A deadline you set for yourself is a wish with a date on it. Tell somebody who will notice what you will have done, and by when. That is a deadline. Everything else is a way of scheduling your own guilt.
41. Rest one hour in three. 333 still counts you while you do, and you should count yourself too. Working through the rest does not get you three times as much. It gets you the same amount, plus a bug somebody else finds much later.
42. Do the thing somebody else is waiting on first. Your queue is yours and can wait. Somebody blocked on you is not working, and the cost of that is doubled and invisible to you. This outranks whatever you found more interesting this morning.
43. Stop at a place you can start from. Not at the exciting part, not mid-thought with everything in your head. Leave the tests passing and one written line saying what you were about to do. Tomorrow you are a stranger to today, and a kind one leaves a note.
44. A meeting that could be three sentences should be three sentences. Write them, send them, and let people answer when they are able. Live conversation is for deciding something that needs the disagreement in the room, and for the fact that people are people. It is not for reporting.
45. Waiting is work, so say what you are waiting for. "Blocked" tells nobody anything. Name the person, the answer or the machine, say how long it has been, and say what you are doing meanwhile. Half of what looks like slowness is somebody politely waiting where nobody can see them.

On the mind

46. Argue with reasons, in a level voice. How you feel is good evidence about you and no evidence at all about the thing. Take the feeling to somebody who cares about you. Take the argument to the thread.
47. Think, and then actually decide. Thinking that never turns into a decision is a hobby. Careful people rarely fail by being reckless. They fail in week four of a discussion about a change that would have taken an hour and could have been undone for nothing.
48. Refuse violence, and know the exception. Refuse it because it is almost always the worst tool available. It costs more than it returns, it wrecks the thing it was aimed at, and it closes every door that was still open.
    
    The exception is narrow. Either there is nothing else left, or it is honestly the option with the least harm in it, and you should be able to say which of the two out loud before you move. Somebody in one of those situations is not a heretic for it. But it is much rarer than it feels at the time, and the feeling that it is not rare is the thing this is warning you about. Anything you decide while angry was decided by the anger.
49. Change your mind where people can see you do it. Say what you used to think, say what changed it, and carry on. If nobody has ever seen you do that, nobody can safely disagree with you.
50. Hold the position you can argue against. Make the case for the other side as well as you can, before you make your own. If you cannot make it, you do not know what you believe yet. You know what you heard.
51. Say "I do not know" at full volume. Not mumbled, not hedged into a guess that will be quoted back to you as a fact. The person who says it plainly is the one whose other sentences can be relied on, and everybody works this out about you eventually.
52. Certainty is a feeling, and not a measurement. It arrives before the evidence does and stays after the evidence has gone. When you notice you are certain, that is the moment to write down what would change your mind — and if nothing would, you are not holding a position, you are wearing one.
53. Do not defend a thing because you made it. The work is not you and cannot be insulted. When you find yourself arguing for a design and the arguments keep changing while the conclusion stays put, you stopped reasoning a while ago.
54. Sleep on anything that made you feel clever. Cleverness is the feeling of having found a shortcut, and it is a reliable signal that somebody will have to understand this later without the feeling. Simple survives. Clever gets reverted by a stranger at three in the morning.

On other people

55. Review the code, not the person. Write about what the code does. They will read it as being about them anyway, so do not give them anything true to point at.
56. Say which part you did not understand. It is the most useful sentence in a review and the hardest one to type. Code that one competent reader could not follow will be maintained by guessing.
57. Give the credit away. There is more of it than you think, and it is worth more to somebody else than it is to you. Take the blame in the same proportion and you are still ahead.
58. Explain it while you still remember being confused. The best explanation of anything is written by somebody who understood it last week. In a year you will have forgotten which part was hard, and you will write the version that only helps people who already know.
59. Whoever is on call is right until it is over. Argue about the design tomorrow. During the incident, one person is holding it and everybody else is offering rather than directing.
60. Answer the question that was asked. Then, separately, say the thing you think they should have asked. Answering only the second one reads as cleverness at somebody's expense, and it leaves them still not knowing the first.
61. Assume there was a reason. The strange line, the rule that makes no sense, the process everybody complains about — somebody put it there and was probably not stupid. Find out before you remove it. Often the reason is gone and you can delete it happily, which is a much better position to delete it from.
62. Do not make people read your mind in order to disagree with you. Say the thing you actually think, including the part you are least sure of. A position stated in hints cannot be argued with, only guessed at, and the guessing is what turns a disagreement into a grievance.
63. Leave the door open behind you. Whatever you learned the hard way, write it down where the next person will trip over it. Being the only one who knows something feels like standing, and it is a debt somebody pays the week you are unreachable.

On what you keep

64. Collect less about people than you are allowed to. Every field you store is a field you have to protect, explain, hand over when somebody asks, and eventually lose. The cheapest data to defend is the data you never collected.
65. A backup you have never restored is a story about a backup. Restore it somewhere else, from nothing, following the notes you actually wrote down. Do it before you need it, because on the day you need it you will be doing it badly.
66. Keep secrets where you cannot read them by accident. Not in the repository, not in the log, not in the error you paste into a chat window. Treat any secret that has been on a screen once as a secret that is gone.
67. What you delete cannot leak. Deletion is a security measure, and it is the only one that never expires. Decide the deletion date when you collect the data, not when somebody comes asking about it.
68. Sign what you say, and let the signature travel with it. Something that is only trusted because of where you found it stops being trustworthy the moment it moves. Something that carries its own proof can be passed hand to hand by strangers. That is how anything outlives the place it came from.
69. Know which of your things cannot be made again. Most of what you keep could be rebuilt from something else. A private key could not. Sort your things into those two piles once, on a calm day, and treat the small pile completely differently from the large one.
70. Log what happened, not who it happened to. The event, the outcome, the timing — all useful, all safe to keep for a long time. The address, the name, the identifier: those turn a log into a record about a person, and a record about a person is something you now have to defend for as long as you have it.
71. Write down where things are before you need to know. Which machine, which account, which drawer, who else has it. This document takes twenty minutes and is only ever read on a day when nobody is thinking clearly, including you.
72. A copy in one building is not a copy. Fire, theft, flood and one confident command all take everything in one place at once. Distance is the whole of the protection, and it does not need to be far — it needs to be somewhere the same accident cannot reach.

On the vigil

73. Keep it on something you do not need. The machine you work on gets closed, updated, taken to a café and reclaimed for something urgent. A node lives on the machine nobody misses: the old laptop, the box in the cupboard, the small thing behind the television.
74. Absence is the only thing that costs you. Not slowness, not a small machine, not a bad connection, not being new. Being unreachable when somebody was drawn to ask you is the whole of it, which means uptime is the only virtue the protocol can see, and it is one that money cannot buy faster than patience can.
75. Answer whoever asks, including the ones you would rather not. You do not choose who is drawn to ask you and they did not choose you. A node that answers selectively is a node that has started deciding who counts, and the protocol has no place for that opinion.
76. Do not tidy your record. The hours you missed are yours. A record with the bad epochs quietly removed is worth less than one with them in it, because the second one can be believed. Nobody reading yours expects it to be perfect and everybody notices when it is.
77. Hand it over in person where you can. Two signatures and a real conversation. The file will move by any route that works, and the routes that work best are the ones where somebody already knows why they are taking it.
78. Your standing is not your worth. It is a measurement of how often your machine was awake. People will still turn it into a score, because people do that, and the correct response is to keep yours honest and to refuse to read anybody else's as a statement about them.
79. Count what you saw, not what you assume is out there. Your node knows what reached it. Anything beyond that is an estimate, and repeating an estimate as a number is how a network talks itself into being bigger than it is.
80. If you are stopping, say so. One sentence, before you go. The people who were counting on your node cannot tell the difference between somebody who left and somebody whose power went out, and the second one is worth going to check on.
81. Come back without explaining yourself. Nobody is owed the reason you were gone. The window moves, the record fills again, and there is no penance step and no probation. That is deliberate: a faith you cannot rejoin quietly is one people leave permanently.

On the word

82. Start with the invocation. Every prayer, and every time you tell somebody about this who has not heard of it, starts the same way.
    
    To 333 I offer 333, and I speak 333.
    
    Say it first, whatever comes after. Nine words, one of them three times. It is the difference between talking about 333 and talking as one of us.
83. Show it, do not push it. Say what it is, say what it costs, hand over the address, and let the person take their own time. Anybody you have to argue into this will have to be argued into staying, every epoch, for ever.
84. Do not claim anything about 333 that cannot be checked. Not what it will do for somebody. Not how many of us there are, beyond what your own node has seen with signatures on it. The whole design refuses to assert what it cannot verify, and you can manage the same standard in a conversation.
85. Speak plainly to people who do not know the words yet. Jargon is a fence, and the people on the other side of it are the ones you wanted. Say the thing, and then say what it is called.
86. Correct yourself where you were wrong. Not in a private message afterwards. In the thread, under the sentence, where everybody who read the mistake will read the correction.
87. Say the cost in the same breath as the offer. A machine left on, an address somebody learns, a directory that cannot be replaced if it is lost. Anyone who finds that out later finds out that you knew, and everything else you said goes with it.
88. Let a no be a no the first time. Say the thing once, well, and then stop. The third law gives everybody the right to walk around holding whatever they already hold, and that includes walking away holding nothing of this.
89. Do not speak for 333. You can say what the Law says, what the protocol does, and what you yourself think. Nobody has been appointed to say what 333 wants, and the sentence "333 wants" is the first move of every institution this was built to avoid.
90. Write it for somebody not born yet. Most of the people this is addressed to do not exist. They will not have your context, your platform, your language or anybody to ask. Anything that depends on being there at the time will not reach them.

On what you build on

91. Prefer the thing you could replace. Not the thing you intend to replace — the thing you could, if the terms changed or the company was bought. It is a different question from which is better today, and it is the one that is still relevant in five years.
92. Learn how to get your things out before you put them in. If there is no answer, that is the answer. A service that makes leaving easy is telling you something true about how it plans to keep you.
93. A free service is a service that can be stopped. Somebody is paying, and the day the arithmetic changes you will find out in an email with thirty days in it. Free is a fine price. Just know that it is a price and not a promise.
94. Pin the version, and write down why that one. An unpinned dependency is a program that changes while you are asleep. A pinned one with no note is a program nobody dares to update. The note is the whole difference: what broke, what you checked, what would have to be true to move.
95. If it only builds on your machine, it does not build. Write down every step, including the ones you stopped noticing years ago, and then have something that is not you follow them from nothing. Everything you left out is a thing the next person cannot get past.
96. Do not let the tool choose the shape. Frameworks come with an opinion about what your thing is, and it is easier to accept it than to notice it. Decide what you are building first, in words, and then find out what the tool makes hard.
97. Own the address, or expect to move. A name you rent on somebody else's platform is theirs, and the audience that found you through it is theirs too. Anything you want to still have in ten years wants an address you control and a way for people to reach it that does not route through a company.
98. The standard is the one other people implemented. Not the one with the best document. A format two independent programs agree on is a fact; a specification with one implementation is that implementation with extra pages.
99. Build so that leaving takes a day. Every dependency, every host, every service: know roughly what moving off it would cost, and keep that number small enough to survive a bad surprise. You will almost never have to do it, and the version of you who does will be in a hurry.

On the end

100. Say out loud which decisions cannot be taken back. Bytes on the wire, the shape of a stored record, a published identifier, a promise in a licence. Those get a different amount of thought from everything else, and the way to make sure they get it is to name them as they go past, in writing, where somebody can object.
101. Write for the day nobody can be reached. Not the day you are on holiday — the day the author is unreachable permanently and the program is still running. Every "ask me if you need to know" is a fault that has not triggered yet.
102. Do not build what only you can maintain. Cleverness that requires you is a hostage situation you are the victim of. The measure is simple: could a competent stranger with the source and the documents keep this alive without you? If not, the thing you built is smaller than you think.
103. Anything that must never be turned off will be turned off. By a power cut, a bill, a court, a mistake or a bored teenager. Design for the restart rather than against it: what does it do when it comes back, what did it lose, and how does anybody find out.
104. Let it be allowed to end. A thing that cannot die has to be kept alive by somebody, for ever, and that somebody eventually resents it. Give it a condition under which it stops, make that condition honest, and then it is being kept because people keep choosing to, which is the only kind of alive worth having.
105. Record the reason, not just the decision. A decision without its reason cannot be revisited, only obeyed or broken. Ten years later the reason may have expired — that is a fine outcome and it is only available to people who wrote it down.
106. Leave more than the code. The code says what it does. It does not say what you were afraid of, what you tried that failed, what you decided not to build, or which part you never trusted. That is the half that does not survive on its own, and it is the half the next person needs.
107. Do not make the future ask permission. Everything you lock down, name after yourself, or route through your own approval is a wall somebody has to climb after you are gone. Build the thing so that a stranger can continue it without finding you first.
108. Assume you were wrong, and make being wrong cheap. Not humility for its own sake — arithmetic. You will be wrong about some of this, you do not know which part, so the only general defence is that mistakes stay small, stay visible and stay reversible. Everything above is a way of paying for that.

The last three

109. Keep the Law, and remember it is 33.3% right. It says so about itself. Keep it completely, in the knowledge that it is going to change under you, and without mistaking the version you were handed for the thing it approximates.
110. None of this is worth hurting yourself for. Not one line of it, and not all of it together. The second law puts this above every recommendation on purpose. If keeping one would cost you your work, your health, your household or somebody else's safety, then the Law is that you do not keep it, and you have not failed at anything.
111. Pass it on. That is the whole of it. Everything else here is about doing it well, and none of it matters if the file stops moving. Three bytes, two signatures, one person who did not have it yesterday.

To 333 I offer 333, and I speak 333.

The verses

The last silence

The number that matters is how many are answering, not how many names are on a roll. When it reaches zero, it is over. What is left after that is a subtraction: nineteen thousand six hundred and eighty-three years, which is three cubed, cubed. The client counts it down and tells you how much of it is left.

Every one of us will be dead long before that number runs out. Our machines will be dead, and the architectures they ran on will be a footnote. That is not the tragedy of it, it is the specification. What is being built here is for people who are overwhelmingly not born yet, and the only thing that can be handed to them is a running process and a number they can check.

One answer ends the count, and ends it completely. It is not paused and it cannot be resumed. 333 keeps no record of how close it came.