Haute Lumière · The reading

Every program you run should feel like the same house, and none of them should feel like the same program

A template makes your programs identical; a canon makes them recognisable, which is the thing you actually wanted.

Four programs to reconcile, and the work begins where it always begins: with the list of things none of them may do.

THE SAME HOUSE

The symptom shows up first in a small thing. A participant finishes one program, starts another, and asks a question they should never have needed to ask: is this the same kind of place. In the first program, missing a week was absorbed without comment. In the second, it produced three automated messages and a tone. Nobody designed that difference. It is what four competent people produce when they build four good things at four different times under four different pressures.

The usual diagnosis is a branding problem, and the usual remedy is a shared palette, one typeface, a deck everybody works from. That treats the surface and leaves the body alone. The incoherence a participant feels is not in the colour of a heading; it is in what was promised at the door, what was refused in the middle, and how they were addressed the week they fell behind. Those are decisions, and a palette has no opinion about decisions.

The second diagnosis is more expensive: standardise everything. One shape, one length, one arc, one set of materials, applied across the catalogue. It holds for about a quarter. Then the introductory program is being stretched to fill a frame it does not need, the advanced one is being amputated to fit the same frame, and the practitioners who care most are the first to start quietly working around it. Sameness is not coherence. It is coherence's cheapest imitation, and people can feel the difference within one session.

What the house actually wants is a third thing, and it is worth stating precisely, because most of the argument follows from it. A person who has been through one of your programs should be able to walk into another and already know how they will be treated — what will be asked of them, what will never be asked of them, what happens if they stop. And the second program should still be genuinely, visibly different from the first, because it is for a different person at a different point with a different amount of time.

Coherence is not the absence of difference. It is difference a person can predict.

That is a testable standard rather than an aspiration, which is what makes it useful on a Tuesday. Take two of your programs and write down, for each, the answer to four questions: what is promised, what is refused, how a struggling participant is addressed, and what happens when someone goes quiet. If the answers differ in ways nobody chose, you do not have a branding problem. You have programs that believe different things about people, and the participant is the one carrying the discrepancy.

WHAT A CANON IS

A canon is a short, ordered list of principles, each of which is able to refuse a specific thing, held above every program and subordinate to none of them. Every word in that sentence is doing work, and the rest of this argument is mostly an unpacking of it. Short, because a list nobody can recite from memory is not operating. Ordered, because the principles will collide. Able to refuse, because a principle that forbids nothing is a sentiment with a serial number.

It helps to define it against the three instruments it is most often confused with, since each of those three is real and useful and doing a different job. A template constrains form: the arc, the number of sessions, the shape of a module. A style guide constrains surface: the words, the marks, the type. A manual constrains sequence: do this, then this. A canon constrains none of those. It constrains decisions, and leaves form, surface and sequence open to whatever the particular program needs.

This is why a canon reads thinner than the documents it replaces and does more than all of them together. A template can tell you that a program has six modules. It cannot tell you whether to add a seventh that would improve completion by making the material easier than the truth warrants. That is the question that actually arrives, in a meeting, on a Thursday, against a deadline, and it is the only kind of question worth writing a document about.

Short is load-bearing, and it is the part most houses get wrong on the first attempt. Seven principles can be held in a head and used in a room. Twenty is a filing cabinet, and a filing cabinet is consulted after the decision, to justify it. The discipline is subtraction: write thirty, then find the three or four that the other twenty-six are consequences of. Most of what appears on a first draft is a special case of something else on the same page.

There is a further reason to keep it small, and it only becomes visible after a year of use. A canon is learned by repetition in live decisions, not by study, and the number of decisions available in a year is finite. Each principle has to come up often enough to be absorbed. A list long enough that most entries never surface is a list where most entries are never learned, which means the effective canon is the four lines people happen to remember — selected by accident rather than by you.

THE REFUSAL TEST

One test determines whether a canon will be used or filed, and it can be run over a draft in twenty minutes. Every entry must name something it forbids, specifically enough that you can point at a real thing you would otherwise have shipped. 'We value clarity' forbids nothing; nobody has ever proposed being unclear. 'No program asks a participant to perform progress they have not made' forbids a specific, common and thoroughly tempting thing, and you can probably name three places the house currently does it.

Run the test line by line. For each principle, complete the sentence: because of this, we will not do X. If X does not arrive within a minute, the line is decoration, and decoration is not free. It dilutes the entries around it, and it teaches every reader that items on this list do not have to bite. A canon of nine lines where two bite is a canon of two lines with seven pieces of camouflage.

A principle that has never cost you anything is a decoration you are paying to store.

This makes the canon uncomfortable to write, and the discomfort is the evidence that it is real. Each line you keep takes away something the house currently does, or something it was looking forward to doing next year. The meeting in which you write it is not a values exercise. It is a negotiation in which the house trades options for the ability to decide quickly for the next five years, and it should feel like trading something away.

The compensation is larger than the cost, and it arrives from a direction nobody expects. A constraint that removes the obvious move forces the next move, and the next move is usually better, because the obvious move was obvious to everyone including the people you are compared against. A program that may not use a deadline to manufacture urgency has to find another reason for a person to begin tonight, and the reason it finds is a better program than the deadline would ever have produced.

The entries themselves are found rather than invented, and they are not found on a whiteboard. Go through the last two years of decisions the house regrets, and ask what each one had in common. A regretted decision is a principle that was not written down, discovered late and paid for at full price. Most houses have four or five of these and have been telling the story of each over dinner for years, without noticing that the stories are the document.

One caution, because this is precisely where canons turn poisonous. A refusal is a refusal of a move, never of a person. 'This crosses four, here is the part that crosses it' is usable, correctable, and can be answered. 'You do not understand the canon' ends the conversation and converts the list into an instrument of rank. A canon used to win arguments stops being used to decide them, usually within a season, and the recovery is slow because what was lost was trust rather than a rule.

The second list — what each program is free to change — takes an afternoon and prevents a year of quiet disobedience.

THE ORDER

Two principles will collide in a real room, over a real proposal, and this is not a flaw in the drafting. It is what happens when a list of true things meets a specific case. Warmth collides with honesty the week a participant is doing badly and asks how they are doing. Access collides with depth the moment the price comes up. Speed collides with care on the last afternoon before a launch. Every canon worth having contains these collisions, because a list that contains none is a list of things nobody disagrees about.

In an unordered canon, every one of those collisions travels upward. The person in the room cannot resolve it, since both lines are equally authoritative, so the decision goes to whoever is most senior and available. The cost is paid twice: once in the hours, and once in the slow lesson everybody learns that the canon does not actually decide anything. Within a year the list is what gets cited after the senior person has already decided.

An ordered canon hands the person in the room a tiebreak they can apply in ten seconds and be right about. It does not make the decision comfortable; nothing makes that decision comfortable. It makes it available, at the level where the information is, which is nearly always lower than the level where the authority sits.

An unordered list of principles is a machine for sending decisions upstairs.

Ordering is the hardest hour of the whole exercise, and it is the hour that pays for the rest. You cannot do it in the abstract, because the question of whether honesty outranks warmth has no answer in the abstract. You order on cases. Take a real decision the house made badly, set the two colliding principles beside it, and ask which one should have yielded. Do that six times and the order writes itself, and you will have learned something about the house that nobody could have told you directly.

Expect one or two of the orderings to be genuinely contested, and resist the urge to paper over them. A contested ordering that gets settled and written down is worth more than eight comfortable ones, because it marks the exact place where the house had been deciding at random. The argument that settled it belongs in the document, one line, beside the entry: this yields to that, and here is the case that settled it. In five years that line will be the most useful sentence on the page.

WHAT MAY VARY

A canon on its own produces a predictable failure, and it takes about six months to surface. Practitioners become unsure what they are permitted to change. The safest reading of any authoritative document is that nothing may be altered, so the careful ones copy the last program exactly and the confident ones adapt in private and stop showing their work. Both outcomes are bad. The second is worse, because it takes the best thinking in the house off the record.

The remedy is a second list, written at the same time and given the same weight: the axes along which programs are expected to differ. Length. Register. Depth. Sequence. Medium. Price. Pace. How much prior knowledge is assumed. How much of the work happens between sessions. How much of it happens alone. The list is specific to the house and writing it takes an afternoon, which is a poor excuse for the number of houses that never do.

Then the axes that may not move, which are usually about four and which are the canon in its applied form: what is promised at the door, what is refused inside, how a person is addressed when they are struggling, and what happens when they stop. Those four are what a returning participant is actually recognising. Everything else is free, and saying so out loud converts adaptation from a small act of disobedience into a duty that can be checked.

Consider how a single principle expresses itself across three programs in a house that has done this properly. Take the line forbidding a program from asking for performed progress. In an introductory program for people with no background, it shows up as a first session with nothing to present and no round of introductions in which everyone reports a credential. In an advanced program it shows up as explicit permission to bring the work that is going badly, and a facilitator who brings theirs first. In a self-paced program it shows up in what the software declines to do: no streak, no comparison, no message implying somebody else is ahead.

That is the whole method in miniature, and it explains why the canon is written as decisions rather than as forms. A template could not have produced those three, because they share no form. A style guide could not have produced them, because they share no surface. What they share is a refusal, and a refusal travels across media, price points, durations and audiences without needing a translation. The participant moving between the three cannot say what the through-line is, and can feel it before the first session ends.

THE FIRST HOUR

A canon is not transmitted by its document. The document is the record; the transmission happens when a person watches a decision get made and hears the reason stated out loud in the canon's own words. This is the part most houses skip, because from the outside it looks as though nothing is being taught, and it is the only part that reliably works.

The practical form is tedious and effective. For the first several months, every decision that touches the canon is announced with its principle attached. Not 'let us not do that' but 'we are not doing that, because of three, and here is what three protects.' It feels heavy-handed for about eight weeks. Then it stops being said aloud, because everyone has it, and that silence is the moment the canon is actually installed.

People learn a principle from the first decision it costs them, not from the page it is written on.

A second practice is worth more than any workshop. Put a new practitioner on a refusal before you put them on a build. Hand them a real proposal the house declined, without telling them why, and ask which line it crossed. They will get it wrong once and right thereafter, and they will have learned the edges of the canon, which is the only part that is difficult. The middle of a principle is obvious to anybody who reads it. The edge is where the work is, and the edge cannot be written down without becoming a manual.

Something will happen along the way that looks like failure and is not. A new practitioner holding the canon will find a place where the house does not follow it, and they are usually right. Two responses are available: change the practice, or amend the canon and say why. The third response — explaining that this particular situation is special — is available exactly once, after which nobody believes the list again, and the cost of that one use is years.

A cheap check will tell you whether transmission has happened at all. Ask three practitioners, separately, to write the canon down from memory and to put it in order. The lists will not match exactly and they do not need to. What matters is whether the top two entries agree, because the top two are what settle the contested cases. If the top two differ across three people, the house has three canons and has been running them in parallel without noticing.

None of this requires a training program, a certification or a slide deck, and their absence is not a shortcut. A canon learned from a deck is recognised rather than held: the person can quote it and cannot apply it under pressure, which is the only condition that ever matters. An hour spent watching a real decision get refused for a stated reason does more than a day of study, and it is an hour the house was going to spend on the decision anyway.

The argument that settles an ordering is the one worth keeping. It outlasts everyone who was in the room for it.

THE INSTRUMENT

Measuring a canon is where most of these efforts quietly become theatre, because the obvious thing to measure is adherence to a document and the obvious thing is useless. An adherence score can be high in a house where the canon has no teeth, and low in a house using it well and arguing about it productively. Three other measurements are worth the trouble, and all three are cheap enough that cost is not the reason they go unmade.

The first is the one that finally matters: outcomes for the person inside the program. Did they finish, did it do what it said it would do, would they enter another one. A canon exists to make those numbers better across a whole catalogue rather than in any single program, and if it does not, it is an aesthetic preference wearing a governance costume. This measurement is slow and it is the only one that can settle the argument.

The second is transfer, and almost nobody measures it. Take someone who has completed program A, put them at the start of program B, and listen to what they ask. If they are asking about the manners of the house — how much work, what happens if I miss a week, is this going to expose me in front of people — then the canon has not transferred, whatever the document says. If they are asking only about the subject matter, it has. The whole claim of coherence rests on that difference, and it costs one conversation to check.

The third is the sharp one: count refusals. Keep a single log with three columns — the date, the thing that was proposed, the principle that stopped it. A quarter with no entries does not mean the house has been behaving. It means either the canon contains nothing that bites, or people have learned to route around it and have stopped bringing it the proposals it would refuse. Both are serious, and neither is visible in any other measurement you keep.

One discipline governs all three, and it is the discipline that keeps any instrument honest. When a measurement cannot be taken — the log was not kept, the cohort was too small, the record is missing — the honest output is the name of the missing read, never a zero. A blank and a zero are different facts. A system that reports the first as the second will eventually tell you the house is clean on a morning when it is not, and it will do so in a confident voice.

An instrument that has never disagreed with you is an instrument you have not checked.

THE AMENDMENT

A canon that cannot change becomes scripture, and people are unfailingly polite to scripture. They nod at it in the meeting and design around it in the afternoon, and the gap between the stated house and the working house widens until the document describes an organisation that no longer exists. A canon that changes whenever somebody finds it frustrating is not a canon at all; it is a record of the last conversation, and nobody can build against it.

The workable middle is a rule about what is allowed to cause a change. A principle is amended when a real case defeats it — when applying it produced a worse outcome for an actual person in an actual program, and the house can point at that case. Not when it is inconvenient, not when it costs money, not when a large client dislikes it. Inconvenience and cost are precisely what the line was written to impose, and treating them as grounds for amendment empties the list within a year.

The procedure is three steps and takes ten minutes. Write the case. Date the change. Keep the old wording beside the new, permanently, never silently replaced. Everything valuable about a mature canon depends on that third step, which is the one people skip because it makes the document untidy and because the untidiness has no benefit for about eighteen months.

The history turns out to be the most useful part, and this is not obvious until you have some. A new practitioner reading 'no program manufactures urgency' learns a rule they can follow. A new practitioner reading the same line with the three cases that shaped it — the launch where it was tried, what it did to that cohort, the year it was narrowed to admit a genuine external deadline — learns why the line sits where it sits. Only the second one can hold it against a senior person with a good argument, which is the only circumstance in which it will ever be tested.

Keeping the history has a second effect, and it acts on the people doing the amending. A change you must write down beside its predecessor is a change you think about for an extra ten minutes. It is the same mechanism as a visible edit history anywhere else: the record does not prevent bad revisions, it makes casual ones rare. Casual revision is how a canon dies without anybody deciding to kill it, and it is the most common ending by a wide margin.

THE SECOND PROGRAM

The case for any of this is a compounding one, which is why it is so often argued badly. The first program built under a canon is slower than it would otherwise have been, because the canon is being written inside it and every principle is being argued for the first time against a live deadline. Anybody measuring at that moment concludes the exercise was a mistake, and their measurement is accurate and their conclusion is wrong.

The second program is noticeably faster. A large share of the design decisions that consumed the first one are already made, and the remaining work is fit — this audience, this depth, this length, this amount of time between sessions — which is the work practitioners are good at and enjoy. By the fourth, design meetings are short, and what gets discussed is the particular thing rather than the same eight questions about posture and promise.

The second effect lands on the catalogue rather than on any program in it. A participant who trusts the manners of one program will enter a second without re-evaluating the house from the beginning, and that first evaluation is the most expensive moment in the entire relationship. Each program that behaves like the others makes every other program cheaper to enter. That is not a marketing claim; it is an arithmetic consequence of not asking people to decide about you twice.

The third is delegation, and for a small house it is the one that changes what is possible. You can hand a program to someone new when the things they must not do are written down and ordered, because you are no longer relying on their taste matching yours. You are relying on a list they can read in an afternoon and a tiebreak they can apply without you in the room. A house that cannot delegate a program has a catalogue limited by one person's calendar, and no amount of demand will move that limit.

There is a fourth effect, harder to put a number against, and it is mentioned last because practitioners raise it unprompted. Designing inside a canon is better work. The eight questions that used to be relitigated in every planning meeting are settled, so the meeting is about this audience and this material, which is the part of the job people took the job for. A house that keeps losing its best designers to fatigue is often losing them to one argument held eleven times.

None of the three arrives in the first quarter, and it is better to say so than to promise otherwise. The honest shape is a slow start, a noticeably faster second build, and a catalogue that gets cheaper to extend every year it stays coherent. For a house that will only ever run one program, a canon is overhead and should be skipped. It begins paying at two, and it does not stop.

WHAT IT COSTS

Three failure modes account for nearly every canon that stops working, and all three are visible early enough to correct. The first is the canon as weapon: the list gets cited to win arguments rather than to decide them, usually by whoever is most fluent in it. The fluent person wins, the room learns that citation is a rhetorical move, and honest citation stops. The correction is a standing rule that a refusal must name the entry and the specific thing, and must be answerable by the person it was aimed at.

The second is the canon as theatre. It is recited at the opening of things, printed nicely, and refuses nothing at all in a full year. The refusal log catches this one, which is most of the reason to keep the log. The correction is not more enforcement; it is going back through the entries and asking which of them has ever cost the house anything, then deleting the ones that have not and being honest about what is left.

The third is growth. A canon accumulates entries the way every document accumulates entries, and each addition looks reasonable on the day it is proposed. At twenty entries nobody can hold the list, the ordering stops meaning anything, and the canon has quietly become a manual. The correction is a scheduled subtraction: once a year, ask of every entry whether it is a consequence of another entry, and remove it if it is. Expect to remove more than you added.

The real costs deserve naming too, since they are paid whether or not they were anticipated. An argument at the start, in the ordering, which will be the most uncomfortable meeting of the year and should be scheduled as such. Autonomy at the edges, which the most experienced practitioners feel most sharply, and which is discussed with them rather than announced to them. And maintenance of perhaps an hour a quarter, in the amendments and the log — small, and the first thing a busy house drops.

What to do with this on an ordinary Tuesday is short enough to begin today. Write the list of things your programs must never do, in plain sentences, each one naming a real move it forbids. Cut it until you can recite it from memory. Order it against six decisions the house has already made badly. Then write the second list — what each program is free to change — and hand both to the person building the next one, before they start rather than after.

This volume is one chapter, on the shelf called The Developmental Canon, and the reading costs nothing: no account, no card, nothing to sign. Buying it is for keeping the file. The argument is the one set out above, and the whole of it is here.

Free to read

Free to read, and free to hear. Every chapter of every book in this house, and every narration of it, is open to anybody. No account, no card, nothing to cancel.

What is in it

• Program Design: How to Use the Canon Across All Luminous Programsopen this in the house search

A canon does not tell a program what to be. It tells it what it may not become.

A template is what a house reaches for when it wants coherence and has only an afternoon.

The list of what may vary is half the canon. Skip it and nothing adapts.

Write the refusal down or the most senior voice in the room becomes the principle.

Keep looking

Every phrase on this page opens into the house search. The shelf holds The Developmental Canon and six other shelves, and the reading is free.