Gnosis, Belief and Paradigm Shift

Belief work goes wrong in a predictable way: the aim is stated so loosely that nothing can count as failure, and the record is kept in a form that records nothing. Both problems are procedural rather than metaphysical, and both can be fixed by running the exercise as a sequence instead of as an attitude. The theory underneath this is the early chaos position that a spell has no reality of its own and serves whatever end it is applied to, which is a claim about the absence of an intrinsic mechanism rather than about the presence of one. The same theory carries a well-documented weakness, because practices which reward noticing successes and ignoring failures appear effective regardless of the mechanism involved. Expectancy effects have a measured influence on reported pain and on some physiological outcomes, which is a real effect of limited size and a different thing from curing a disease. A Bayesian approach, which means keeping a record and updating it when the evidence disagrees, is entirely compatible with wholehearted occult commitment and is the only part of the exercise that can fail. This page sets out the sequence in four steps and spends most of its length on step four, which is the step that decides whether the exercise told you anything.

The exercise, stated as a sequence

The exercise runs over a fixed period and ends in a decision. The period is four weeks, which is long enough to produce a usable number of entries and short enough that the record is still honest at the end. The decision at the end is written down in advance, in one sentence, before the aim is chosen. Without the pre-written decision the exercise is a mood that lasted a month.

The four steps are: write a sentence you can be wrong about, open a record that starts before that sentence, add nothing for a fortnight, and then read the record against the aim. The order matters and the second step matters most, because a record opened after the aim has begun is already being shaped by the aim.

Step one: the sentence you can be wrong about

A statement you can be wrong about has a condition attached to it that some future observation would settle. I will not be healthier by the end of the month is falsifiable. I will be more confident is not, because confidence is a self-report and self-reports move with attention. My judgement about this situation will be clearer is not, because clearer is not a measurable property without a criterion.

The most common version of this mistake is the vague positive. Aiming at wellbeing, at peace, at being the person I want to be, or at having the situation under control, produces aims that no observation can contradict. This is not a metaphysical problem. It is the reason the record has nothing to compare itself against, and the fix is entirely in the choice of sentence.

Step two: the record that starts before the aim

Open the record first, and leave it blank for a day or two. This sounds like a small procedural detail and it is the single most important one, because it establishes a baseline that was not chosen with the aim in mind. A record opened on the day the aim starts contains only data that the aim has already coloured.

What goes in the baseline period is ordinary: how many hours of sleep, what the mood was like on a scale, whether anything unusual happened, what was done and not done. Nothing exotic and nothing interpretative. The purpose is to have a few days of unselected material so that the later entries can be compared against something other than your expectations.

Step three: the fortnight with nothing added

For the next fortnight the record gets one line a day and nothing else. The line is a number, a short description, or both. No interpretation goes in. The temptation to write what the entry means is the reason the record stops being evidence, because interpretation is where the aim re-enters.

People find this fortnight frustrating and the frustration is informative. It is the fortnight in which the record shows you very little, and the instinct is to add commentary that makes the silence mean something. Resist it. A record of fourteen single lines with no commentary is a genuinely useful artefact and it is much rarer than it should be.

Step four: reading the record against the aim

Now put the aim sentence next to fourteen lines and a baseline. Look for one thing only: an entry in the fortnight that would have been surprising if the aim were false. Not a feeling that things improved. Not a general impression of the fortnight. A specific entry, in the record, that the aim's being false would not easily explain.

If you find one, you have a datum. You do not have a mechanism, you do not have a causal claim, and you do not have anything you can repeat. You have one entry that was not what the null hypothesis would easily produce. That is worth more than a month of confident impressions and worth less than most people will claim for it, and both errors are common.

The update, and the part beginners skip

The Bayesian move is the least intuitive part of the whole exercise and the part that almost nobody performs. Having read the record, you update: you adjust how much weight the aim deserves. If the fortnight contained nothing surprising, the weight goes down. If it contained one surprising entry, the weight goes up a little. Neither outcome is dramatic and both are honest.

The version of this that beginners skip is the downward update. Raising the weight when something works is easy and requires no decision. Lowering it when nothing works requires you to act on the result, and it is the only part of the exercise where the aim can actually be abandoned. An exercise in which the weight never goes down is not an experiment.

Running it a second time with a different aim

The second run is where the exercise becomes useful, because the first run always produces one memorable entry and the second produces a comparison. Run the same four steps with a different aim, keeping the record format identical. The two records can then be read against each other, and that comparison is much harder to explain with a single favourable anecdote.

The second run should use an aim of similar specificity and a comparable likelihood of being checked. A run with a vivid aim followed by a run with a very safe aim will produce a difference you will be tempted to read as a result. Change one variable at a time, which here means change the aim and leave the format alone.

Where the sequence fails at step four

Step four fails when the reader is looking for support. The fourteen lines are read and the eye finds the one that fits, and the thirteen that do not are dismissed as noise. This is a property of human inference and not a flaw in your sincerity, and the countermeasure is structural rather than moral.

The countermeasure is to decide the decision rule before reading. Write down, at the start, what count of surprising entries would be enough to increase the weight and what count would be enough to decrease it. Two surprising entries increases it. Zero decreases it. Deciding this in advance removes the discretion that step four otherwise exercises, and the discretion is where the failure lives.

The aim that is a question instead of a command

Most aims in this material are commands, because the sigil encodes an imperative statement of intent. A question cannot be encoded that way, and a question-shaped aim is genuinely useful for a different purpose. What do I actually want here, and what would I do if I had to choose this week, are questions that produce information rather than a directive to be rationalised around.

The useful sequence is to run the four steps with a command and then run them once more with a question. The command exercise tells you what happened. The question exercise tells you what you were avoiding, which is usually the more useful of the two and much harder to obtain any other way.

Keeping the record when the answer is no

When the fortnight produces nothing, the record is more valuable than when it produces something, and this is the point at which most people stop keeping records. A null result recorded in a format you will still understand in a year is one of the few things in this material that transfers directly to anyone who wants to know whether the practice works.

So keep the file. Do not delete the aim, do not reframe the fortnight as preparation for the next one, and do not quietly extend the period. Close the record, write the date on it, and leave it. Then decide the next aim independently, rather than choosing something more likely to succeed because the last one failed.

What the sequence is for

The sequence produces a small amount of information at a cost of about an hour a day for a month. It is not going to establish that magic works, and nothing at this scale could. What it can do is tell you whether a particular aim, in a particular form, produced anything that a null hypothesis would not easily explain, and it can do that without requiring you to believe anything in advance.

That is a modest result and it is worth having, because the alternative is a practice in which the belief is total and the evidence is never collected. The sequence was not designed to persuade sceptics. It was designed so that the person running it has something to say other than yes.

The person you would show it to

There is a person you would show this to, and writing their name down at the start is a more effective commitment device than any sigil in this material. Not to persuade them. To make sure the record is written in a form that would survive their reading, which means no interpretation in the fortnight, no commentary in the baseline, and a decision rule fixed in advance.

In practice this one change does more for the quality of the evidence than any other. A record kept for an imaginary reader is a record with a consistent format, and a consistent format is what makes the numbers mean something. The imaginary reader is a fiction, and it is the cheapest methodological control available.

When to run it again and when to stop

Run it again when the first run was inconclusive, which is most of the time, and when the aim is one where a result would actually change what you do. Stop when you have run the same aim three times and produced nothing worth acting on, because at that point the honest description is that this aim does not work for you, and re-running it is a way of avoiding that sentence.

The stopping rule is worth writing in advance for the same reason as the decision rule. Deciding afterwards that you have run enough is a different decision from deciding at the outset, and only one of them is honest. Three runs, then choose a different aim or a different practice, and let the record say what it says.

Frequently Asked Questions

Does belief need to be present for an expectancy effect?

Not always, which is one reason the simple explanation of these practices as pure belief does not hold. Open-label studies suggest belief in a treatment is not always necessary for a measurable physiological effect, and expectancy has a documented influence on reported pain and on some physiological outcomes. The effects are real and limited, and a limited effect on reported pain is a different claim from an effect on disease.

How do I write a goal that can actually be proven wrong?

Attach an observation to it that some future event would settle, and specify what that observation is in advance. I will be healthier by the end of the month is checkable. I will be more confident is not, because confidence is a self-report. I will be happier is not, for the same reason. The test is whether you can describe the evidence of failure before you start.

What is a Bayesian approach to magic?

Keeping a record and updating how much weight you give your belief when the evidence disagrees, which in practice means lowering the weight when nothing happens, not just raising it when something does. An experiment in which the weight never goes down is not an experiment. The difficult part is the downward update, and it is the only part of the process that can actually abandon an aim.

Why does my confirmation bias matter if I am being sincere?

Because it operates on selection rather than on dishonesty. Practices that reward noticing successes and ignoring failures appear effective regardless of the mechanism involved, and this is one of the most thoroughly documented findings in the whole area. Sincerity does not help, because the bias is in which entries you attend to. The fix is structural, which is why writing the decision rule before reading matters more than trying to be careful.

How long should one of these exercises run?

Four weeks, with a baseline period opened before the aim starts, is short enough to stay honest and long enough to produce a usable number of entries. Shorter than two weeks produces too few entries to compare. Longer than six weeks usually produces drift in what you are recording, and the record stops being comparable to its own baseline.