V.0 · The Methods

Taken here: the catalogue. Given back: the expiry date. Reads as: pedantry.

There is a canonical list of one hundred and ninety-eight methods of nonviolent action, compiled from the historical record and sorted into three families: protest and persuasion, non-cooperation, and intervention. It is the most useful single document in this field and it is sixty years old, which is the problem. It is sorted by what a method is.

Sorting by what a method is does not tell you whether it will work on Tuesday. This inventory tracks remaining deniability: how far the institution has got in naming the method and preparing a response. Once it has a category and a procedure, your action may arrive as a case it already knows how to handle. That changes the contest; it does not end it. The form may be ready. The people who have to make the arrangement work may not be.

So the inventory below is banded, not familied. Bands are a snapshot and they decay. This one was taken on the date in the header, and anyone reading it later should assume every entry has moved down at least one band.

BandConditionWhat the institution does
ANot yet namedimprovises, badly, and often makes it worse for itself
BNamed, no procedureescalates to someone senior, who invents a response
CNamed, procedure existsapplies the procedure. The method still works; the cost is now known and priced
residueNamed, procedure is automatic and cheaplogs it

Band C is not failure. Most of what has ever worked lives in Band C, and lives there because it is worth the priced cost.

The fourth row is not a band and is deliberately not lettered. A, B and C are positions on one clock — remaining deniability. The residue is what the clock leaves behind, sorted by whether anything has to be decided when the method arrives, which is a different measurement. V.4 sets out why the distinction is worth the untidiness.


V.1 · Band A — not yet named

These are cheap, lawful, individually deniable, and mostly consist of doing your job with a thoroughness nobody has ever complained about in writing. They work because the vocabulary needed to prohibit them would also prohibit diligence.

#MethodFamily
1Exhaustive consultation — route every decision past every stakeholder entitled to see itnon-cooperation
2Complete documentation — record every decision, its basis, and who made itnon-cooperation
3Literal compliance — deliver the letter, withhold the inferencenon-cooperation
4Full-form completion — every optional field, every attachment, every timenon-cooperation
5Escalation by policy — refer upward exactly as the policy providesnon-cooperation
6Written-basis requests — ask for the authority under which an instruction is givenprotest
7Attributed minutes — every decision carries a name and a dateintervention
8Accessibility requirements invoked in fullintervention
9Records requests, correctly filed, repeatedlyintervention
10Withdrawal of discretionary unpaid labour — work to contractnon-cooperation
11Declining to attend meetings for which no agenda was circulatednon-cooperation
12Insisting on the written procedure for a thing that has noneintervention

Method 12 asks a small, awkward question: where is the procedure? An answer may identify a rule, expose a discretionary choice, or send you to an office that sends you back. Record which happened. The useful moment is when this is how we do it has to explain who we are and where it was agreed.


V.2 · Band B — named, no procedure

Recognized when they happen. No settled response. The institution goes and finds somebody senior, and what that person invents under time pressure is usually worse for them than what you did.

#MethodFamily
13Work-to-rule, declarednon-cooperation
14Slow-downnon-cooperation
15Refusal of overtime, collectivenon-cooperation
16Coordinated absencenon-cooperation
17Mutual aid provision — supply the thing the institution failed tointervention
18Alternative communication systemsintervention
19Parallel record-keeping — keep the account the institution will notintervention
20Narrow targeted boycott, with a stated condition for ending itnon-cooperation
21Selective patronage — move custom rather than withhold itnon-cooperation
22Withdrawal from advisory bodies, publicly and with reasonsprotest
23Non-cooperation with discretionary data collectionnon-cooperation
24Lawful disclosure of an internal record through the provided channelprotest
25Resignation of an honour, title, or membershipprotest

Method 20 carries a condition for ending it because a boycott without one is not a demand, it is a mood, and moods cannot be conceded to.


V.3 · Band C — named, procedure exists, still worth the price

The classical repertoire. Everything here has a rulebook written against it, and everything here has also brought down governments, which is the point of the band. Band C methods are not stale. They are expensive and effective, and the institution has a price in mind, which is more than can be said for Band A.

But the cost is priced, not known. What is priced is the institution’s ordinary response: the procedure it applies, the charge it usually brings, the disciplinary step it usually takes. That is not the same as knowing what happens to a particular person on a particular night. The ordinary response is a floor, states exceed it, and nobody who exceeds it is held to the price list afterwards.

So the decision is not only the group’s. Whoever will carry the risk decides for themselves whether to carry it, is told which parts are known and which parts are guessed, and can withdraw later without arguing for it. Treating a person’s exposure as a line in the group’s budget is how a group talks itself into spending somebody else’s week in a cell.

#MethodFamily
26Strikenon-cooperation
27General strikenon-cooperation
28Rent strikenon-cooperation
29Consumer boycott, broadnon-cooperation
30March or demonstrationprotest
31Picketingprotest
32Vigilprotest
33Occupation or sit-inintervention
34Blockade of a process — never of peopleintervention
35Deputation to a public bodyprotest
36Teach-inprotest
37Symbolic display — banners, colours, lights, empty chairsprotest
38Fastintervention

Method 34 states its own limit. A blockade of a process is an argument about the process. A blockade of people is a different act with a different name, and it performs the conversion described in Part I §I.4 on your behalf, for free.


V.4 · The residue — logged

A note on the ordering, because it changes here. Bands A, B and C are sorted by one thing: remaining deniability — how long until the method is named, until a procedure exists, until the procedure is circulated to peers. That is a clock, and V.5 describes how it runs.

What follows is not the bottom of that clock. It is what is left when the clock has finished running: methods that completed the journey so long ago that nothing on the other side has to decide anything when one arrives. That is a different measurement — decision-forcing, not deniability — so this is not a fourth band of the same series. It is the residue, and calling it Band D implied a rung where there is a bin.

Named, procedured, automatic, and cheap. Not wrong. Not useless as expression.

#MethodWhat now happens
39Online petition to a body under no obligation to read ita number goes up
40Hashtag campaign as the terminal actreach is reported
41Statement of solidarity by an organization with no exposureit is filed
42Awareness dayit recurs annually

The honest use of the residue is as a recruiting surface — the low-cost first step that puts a person’s name next to a position and makes the second step thinkable. Used that way it is valuable. Used as the act itself it is a way of feeling finished.

And the reason it can only ever be the first step is mechanical. Awareness travels on a single exposure and through weak ties: one sighting, from a stranger, is enough to transmit that a thing exists. Doing something does not travel that way. Changing what a person actually does takes several exposures, reinforcing each other, from people they already have some reason to trust — which is a property of dense connection and not of reach. The residue is extremely good at the first and structurally incapable of the second, and no quantity of the first converts into the second. That is why the number goes up and nothing else does.


V.5 · How a method moves

Methods can lose novelty as institutions recognize and standardize responses to them. One possible sequence is: a method attracts attention, acquires a name, reaches those responsible for responding, and prompts a procedure that other institutions may adopt. Check whether those steps occurred here; they are not a universal timetable.

Two consequences follow.

Publicity is a cost paid in deniability. Coverage may make a tactic recognizable; it does not establish that a response procedure exists. Sometimes the trade is correct—reach bought at the price of surprise. It should be a decision and it is almost never treated as one.

The inventory needs a date and a review. Give the inventory a date and an owner. Re-band methods from evidence of recognition, an existing response procedure, and its actual use—not response speed alone. Record acknowledgement, decision, and implementation delays separately. A slow answer may follow a settled procedure; a quick answer may be improvised. Check the Process and Delay Map before declaring a method unrecognized or an intervention unsuccessful.


V.6 · What is not on this list

Three exclusions. They have been stated in one breath until now, which implied a single foundation they do not have.

No method here is a method for harming a person. Part I §I.4 gives the grounds. Two of the three are not strategic at all.

No method here damages property. This is a boundary rather than a principle — a decision about what this repertoire contains, resting on strategic arguments that §V.7 sets out and does not settle. It is held. It is not claimed to be self-evident, and a reader who wants to argue with it should argue with §V.7.

No method here deceives the people it is meant to persuade. Different ground again: a decision extracted by misdirection is not the agreement it resembles, and a movement that lies to the people whose cooperation it needs has conceded the argument it says it is making.

Four measurements are involved and the handbook has been sloppy about keeping them apart: armed against unarmed, violent against nonviolent, harm to people against harm to property, and risk imposed on others against risk accepted for oneself. They do not collapse into one scale, and a plan that treats them as one will get at least one of them wrong.

What this section does not claim. It does not claim that nothing here endangers anybody. It plainly does. Band C contains occupation, blockade and fast; Part IV rule 4 refers to the person taking the arrestable risk; and until this revision this section said otherwise, which was false against the inventory three pages above it. What the handbook asks is narrower and is worth stating exactly: it asks nobody to endanger anyone else, and it asks that the risk a person accepts for themselves be accepted knowingly, in advance, and revisably.

There is also nothing here about evading a lawful investigation into harm. This handbook has no use for it and no view to offer.


V.7 · The flank

§V.6 states an exclusion and Part I §I.4 gives the mechanics behind it. This section states the cost, because the exclusion has one, and a document that listed only the benefits of its own commitments would be the manifesto Part I §I.5 declines to write.

The effect. Where a movement contains both a moderate wing and a more confrontational one, the moderate wing’s position at the table is not set by the moderate wing alone. The presence of the second changes what the first looks like from the other side. This is documented, it has a name — the radical flank effect — and the crucial fact about it is that it runs in both directions.

The positive flank. The confrontational wing makes the moderate wing the reasonable option, and reasonable options get met. An institution facing a demand it does not want to concede discovers that conceding is cheaper than the alternative on offer, and the alternative on offer was not made by you. The moderate wing acquires leverage it did not generate, from people it may not like, frequently while publicly repudiating them.

The negative flank. The confrontational wing supplies the institution with the thing it most needs, which is a category. The whole movement is then described by its furthest edge; the moderate wing spends its capacity on disavowal instead of demands; sympathetic third parties withdraw; and the repression that follows is popular. Nothing is conceded to anybody, and the moderate wing has been made answerable for actions it did not take and could not have prevented.

Which one arrives is not random, and it is not determined either. The evidence is mixed and reading it as a rule produces confident predictions that fail. What follows are tendencies, drawn from cases and experiments that do not all point the same way.

ConditionPushes toward
The two wings are visibly distinct — separate names, separate spokespeople, separate actionspositive
The demand is already gaining ground; the movement has momentumpositive
The institution is internally divided, uncertain, or weakpositive
The wings are not distinguishable to an ordinary observernegative
The wider public is already polarized about the movementnegative
The institution is strong, cohesive, and confident of its supportnegative

And a warning about how that table was assembled. The strongest study of the negative case measured votes in one country’s elections. The strongest study of the positive case measured liberalization in countries that were not democracies. A third, running experiments on a public instead of reading history, found a radical flank raising stated support for a separate moderate group. Votes, liberalization and stated support are three different things, and stacking them into one column is doing something the evidence does not support.

Which effect you get depends on who is watching, which institution is deciding, what the campaign is asking for, how the two wings are related — and, before any of those, which outcome you were asking about.

Read the table honestly and the uncomfortable part is still unavoidable. The conditions producing a positive flank are mostly conditions you do not control. The one you partly control is differentiation, and it is the only lever in the section.

The numbers do not say what this handbook would prefer, and they do not say the opposite either. There are two bodies of evidence and they are measuring different things.

The orthodox result comes from a record of 323 maximalist campaigns between 1900 and 2006 — campaigns to remove a government, expel an occupation, or secede. Nonviolent campaigns in it succeeded around 53 per cent of the time against about 26 per cent for armed insurgency. The proposed mechanism is participation: methods needing no weapon, no training and no youth admit the old, the encumbered and the professional, a movement that admits them gets numbers, and numbers reach the people who supply the six inputs.

The revisionist result comes from recoding that record into four categories instead of two and extending it back into the nineteenth century. Under that coding, strictly nonviolent campaigns are reported at 48 per cent, campaigns including unarmed violence at 61, reactive violence at 60, and fully armed insurgency at 30.

Notice the last figure before any of the others. Arming is what collapses a campaign’s prospects, on both accounts. Whatever else is in dispute here, nothing in this literature recommends a militia.

The recoding is disputed, and the disputes are specific rather than atmospheric:

The objectionThe specific
The threshold for “unarmed violent” is doing enormous workMadagascar, 1991 to 1993, is coded unarmed violent. Of 147 recorded events, around 2 per cent involved anyone throwing a stone
“Unarmed” is defined to include weaponsMolotov cocktails and homemade grenades are counted as unarmed, having been used to lethal effect
The nineteenth century is doing work nobody has measuredThe recoding finds no nonviolent campaigns at all in that period, and how much of the result rests on those cases has not been shown
The gap runs both waysA failed armed campaign leaves records. A failed nonviolent one often leaves nothing, and a replication shows the original finding is sensitive to how many are missing

Both sides have published since and neither has conceded. That is the accurate state of it, and it is where §V.6 sends anyone who wants to argue with the property exclusion.

What this handbook may take from any of it, and what it may not. It may take mechanisms. It may not take odds. Every figure above is a figure for a campaign to bring down a government. Not one of them is a success rate for a records request, a work-to-rule, or a refusal to sign something — which is the limit Part I §I.5 states, arriving here where it does the most damage to the temptation.

The handbook’s own exclusion also sits outside the winning category. The 53 per cent describes campaigns that were predominantly nonviolent; a separate count of primarily nonviolent uprisings found roughly two in five contained some violence, kept in the category as incidental. §V.6 is stricter than that. So the orthodox figure is comparative evidence this handbook can learn from and not a result it can claim, and Part I §I.4 no longer rests on either number.

The longer-run finding is the one that bears on what this handbook is for. Transitions that came out of civil resistance produced democratic outcomes at around 57 per cent, against about 6 per cent for transitions out of armed insurgency — and that pattern is reported for transitions inside autocracies as well, which is worth saying because it cuts against the idea that any of this only applies where the courts already work. A campaign is a year. What it leaves behind is the rest of it.

So the exclusion in §V.6 is a choice and not a finding. It rests on the conversion argument — that spending the asymmetry is irreversible, and is done in an afternoon by the angriest person present on behalf of everyone else — and that argument survives the numbers, because it is about direction rather than odds. What does not survive is any implication that the excluded methods lose. Keep the exclusion. Stop pretending it is free.

Redundancy with a maintenance bill

Part III rule 6 says two roads, or do not go. The useful extension is meaner: name the shared failure. Two logins to the same account are not two roads. Two people who both ask the same unofficial expert are not two roads. The same file saved under two names is not independent evidence. Redundancy begins when the second route fails for a different reason from the first.

For each necessary method, name the shared-failure point and the cheapest manual alternative. At the repair desk: if the ticket system is down, use a paper intake sheet; if the parts cupboard key is missing, keep a second key with the shift minder; if only one volunteer knows the intake rule, train a backup before the next intake day; if the donor spreadsheet doubles as the calendar, print the next two weeks before the annual database ceremony eats the afternoon.

Then price the second road. Who checks it still works? When was it last used? What makes people use it before panic? Redundancy without maintenance is folklore. It is the spare tire under the van, flat since 2019 and still technically present.

Keep the spare route boring enough that someone will actually rehearse it.

What follows practically

Differentiate, and do it before you need to. The one lever you hold is whether an ordinary observer can tell your action from somebody else’s. Separate names, separate calls, separate visual identity, and a spokesperson you chose — Part VI §VI.3 item 6 — all settled in advance rather than in a statement issued at eleven at night. Differentiation established beforehand reads as identity. Differentiation attempted afterwards reads as disavowal, and disavowal is an admission that the two were the same thing until you said otherwise.

Do not manufacture a flank. A moderate wing that quietly welcomes an escalation it can profit from has performed the §I.4 conversion by proxy and kept the deniability, which is worse than performing it directly, because it is also dishonest. It is unreliable as well: a flank you arranged is a flank that can be attributed to you, and it will be.

Do not claim the credit either. Where a concession arrives after an escalation you did not run, the honest account is that you do not know what produced it. A movement that awards itself leverage generated by people it disowned learns the wrong lesson twice — once about what works, and once about who it is.

And do not expect to be thanked. This is the one case in the handbook where a group may gain from something it opposes, cannot ask for, must not arrange, and will be blamed for regardless. No clean position is available. Only an accurate one.

FALSIFIER — the flank. The obvious test is the wrong one, and it was the test this section first published. Comparing campaigns with and without a flank and finding no average difference proves nothing: two real effects running in opposite directions produce exactly that average. So the test has to be conditional. This section predicts a flank helps where the wings are visibly distinct and the institution is divided, and hurts where they are neither. If those two predictions fail to separate — if outcomes under the stated conditions cannot be told from outcomes outside them — the conditions are not doing the work claimed for them and this section is noise with a table around it.


FALSIFIER — the residue. If methods in the residue can be shown to produce concessions at rates comparable to Band B and C methods over a full campaign, V.4 is wrong and those methods belong in the bands.

FALSIFIER — the bands. The one above tests V.4 and nothing else. The bands themselves rest on a different claim, and it needs its own: if independent checks cannot distinguish the bands by recognition and the existence and use of a response procedure, the ordering should be revised or abandoned. Equal response times alone do not settle that question.

This is also the more useful of the two, because it is the one somebody could run. Response time is observable, but it tests delay. Record evidence for recognition and procedure separately; do not make the clock answer a different question.