The Antitese Guide
An argument map does not draw what you think: it draws why you think it. The thesis sits at the top, the reasons below it, and every line claims that what is under it holds up — or attacks — what is above.
The guide has three independent parts. The first teaches you the app and takes about ten minutes. The second shows what each of the built-in models is for. The third is the theory — what the tradition of argument mapping has learned about taking a piece of reasoning apart, with or without software.
Part I
Building a map
From an empty screen to an argument with reasons, an objection and a rating.
Start with the thesis
The app opens on an empty thesis box. Double-click it and write the claim you want to defend. Enter closes the editor, Esc cancels.
A useful thesis is arguable, and it is one claim. “Nuclear power should be part of the transition” is a thesis; “nuclear power exists” is not, because nobody disputes it; “nuclear power is safe and cheap” is two, and you will notice that your reasons only serve one of them.
A map has a single thesis. That is why the THESIS label will not swap for another type: everything else on the map is there to answer it.
Give it reasons
With the thesis selected, Enter creates a premise below it, already connected. Repeat for several independent reasons: each is a separate line of defence, and if one falls the others still stand.
| Type | What it is | Key |
|---|---|---|
| Premise | A reason that holds up the box above | Enter |
| Objection | A reason against the box above | O |
| Rebuttal | An answer to an objection, attacking it | R |
| Evidence | The datum or fact the reason rests on | E |
| Other | A free note, outside the argument: it is not rated | U |
Watch the shape, not only the colour: round corners support, square corners attack. The same rule holds for the icons, the boxes and the brackets.
An objection can attack any box, not only the thesis — and attacking the right reason is half the work. See The parts of an argument.
Implicit claims
When you write a claim the argument needs but nobody stated — the premise that was taken for granted — mark it as implicit: the dashed icon in the box header, or I with the box selected. The outline turns dashed, and the map now shows at a glance what the author wrote and what you added to make the step work.
It is notation, not rating: the mark changes no number, and the box is rated like any other. The thesis does not have it, because the thesis is what you set out to prove. How to find these premises is in What nobody wrote.
Co-premises
Some reasons say nothing on their own. “All men are mortal” and “Socrates is a man” only conclude anything together. Select one of them and press Tab: the two now share a bracket and rise on a single line.
A bracket is worth its weakest link. If one co-premise does not hold, the whole reason does not hold — which is exactly what it means for them to need each other. How to decide between this and two separate reasons is in Independent or linked.
Rating: two judgements, and only two
Everything the app calculates comes out of two things you decide.
- Degree of truth of the box (1 to 5): how plausible the claim is in itself. Click the dots in the header, or use 1–5 with the box selected.
- Strength of the inference (1 to 3): how much that box, if true, holds up the one above it. Click the connector and use the − / + buttons, or 1–3. The thickness of the line shows it: thin and dotted is a leap, thick is conclusive.
The bar at the foot of each box is the resulting strength, from 0 to 100. The thesis has no dots: its value is whatever the argument gives it, and that is the point of the map. The same happens to any box as soon as a reason is placed under it — it stops being something you declare and becomes something the map works out.
A habit worth having. When a score looks wrong, look at the inference first. That is where arguments usually fail — not on a false premise, but on a step that does not hold. Marking that step as a leap is saying out loud what you already suspected.
What the numbers do to each other — why an objection weighs more than a reason, why adding weak reasons never reaches 100 — is in From structure to judgement.
Tidying the map
- The map lays itself out, from the top down. Dragging a box moves the whole branch; L puts everything back in place.
- F folds and unfolds a branch. A folded box shows how many it took with it.
- The note icon on each box holds a comment that only appears when you open it — and it is saved in the file.
- ⌘Z undoes, ⇧⌘Z redoes. ⌫ deletes the box and the branch below it.
- To connect two boxes that already exist, use the connect tool (C): it creates a neutral, grey link that stays out of the calculation.
- The scratchpad on the right lives only in this browser session and never enters the file. It is for thinking out loud, not for keeping.
- The sun and moon in the corner of the footer switch between light and dark mode. That choice is remembered, and this guide follows it; if you never choose, the app opens the way your system is set.
- The globe beside them switches between English and Portuguese. The app always opens in English, in any browser: the language lasts the tab you are working in and is not remembered for next time. Neither switch touches the map — a file saved in one mode or language comes out byte-identical in the other.
Saving
Save with the floppy-disk icon: the map becomes a .tese file (which is JSON inside). Once you have saved it, switch on the cloud in the footer and the app writes to that file by itself, shortly after each change. The footer always says what time it saved and what the file is called.
If the footer says “No file linked”. Writing straight to a file needs a Chromium
browser and a page served over http:// or https://. Opened by
double-clicking the file, the app still works — but saving downloads a copy instead of
writing over the original.
Keyboard reference
| Build the argument | |
|---|---|
| Add a reason | Enter |
| Add a co-premise | Tab |
| Add an objection | O |
| Add a rebuttal | R |
| Add evidence | E |
| Add a free box | U |
| Connect two boxes | C |
| Mark the box as implicit | I |
| Degree of truth of the box | 1 … 5 |
| Strength of the inference (connector selected) | 1 … 3 |
| Undo / redo | ⌘Z / ⇧⌘Z |
| Delete the box and its branch | ⌫ |
| Move around the map | |
|---|---|
| Move the selection | ↑ ↓ ← → |
| Hide / show a branch | F |
| Tidy the layout | L |
| Zoom | Z |
| Fit to the screen | 0 |
| Overview | M |
| Save | Ctrl+S |
| Cancel the action | Esc |
Part II
Using the models
One argument already built and five skeletons to fill in. Each is a way of arguing that the tradition has already tested.
Opening and adapting
The Models button at the top opens the list. Clicking a card replaces the current map — so save what you are working on first.
An example brings a real argument, written out and rated: it shows how the whole thing works. A model brings only the roles — “MORAL PRINCIPLE”, “WARRANT”, “ABSURDITY” — for you to replace with your own case. In either, every box carries a note (the icon in the header) saying what belongs there.
If you switch language with a model open, the model follows: the boxes and the notes that are still as the model wrote them change language, and anything you have rewritten stays exactly as you left it.
Adapting a model is always the same sequence: rewrite the boxes from the top down, delete with ⌫ the ones your case does not use, add the ones that are missing, and rate only at the end. If you are left with a box whose role you cannot fill, that is information: it is almost always the premise your argument does not yet have.
The Problem of Evil
Epicurus · an argument against the existence of God
The oldest argument in the philosophy of religion, and the best short example of a complete map: it has independent reasons, a bracket of co-premises, a classic objection and the rebuttal of that objection.
Try it: drop the degree of truth of the premise about evil to 2 and watch the thesis rise; mark the objection 5 and watch it fall. It is the quickest way to see what the numbers do.
Moral argument
Ethics · a principle, the facts, and the subsumption that applies one to the other
The shape of any moral judgement: a general principle, the facts of the case, and the step that applies one to the other — the subsumption. Principle and facts are co-premises: neither concludes anything alone.
The two points of attack come marked, and they are the ones that matter: one at the principle, one at the subsumption. Hardly any moral dispute is settled over whether stealing is wrong; it is settled over whether this is a case of that.
The Toulmin model
Claim, data and warrant — and what holds them up
Stephen Toulmin (1958) noticed that real arguments are not syllogisms: they have data (the facts you start from) and a warrant — the tacit rule that licenses the move from the fact to the conclusion. It is almost always the warrant that goes unsaid, and almost always the warrant that is wrong.
The qualifier is a free box on a grey line: it is not a reason, it is the confidence with which the claim is made. Use this model when you want to make explicit what licenses your step — it is the best exercise against the argument that “obviously” works.
Reductio ad absurdum
Assume the opposite and show where it bursts
You prove a thesis by showing that its negation leads to something unacceptable. It is the only model that runs as a chain: each box holds up only the one above it, and the conclusion is no better than the weakest link.
A reductio is only as good as the deduction in the middle, and that is where the objection is placed. If the deduction is a leap, mark it as one: the conclusion drops, and it is honest that it should.
The classical model
Aristotle · logos, ethos and pathos, with refutation
The structure of the rhetorical speech: three independent proofs — reason, the speaker’s character, what is at stake for the audience — and the opponent’s objection, anticipated.
The three proofs sit on separate branches on purpose: an audience can reject the pathos and stay convinced by the logos. Use it to prepare a talk or an opinion piece, where the order of presentation matters as much as the structure.
The Rogerian model
Persuading from what you already agree on
It comes from Carl Rogers and inverts the logic of a dispute: instead of knocking the other position down, you begin by summarising it in terms its holder would sign, and look for common ground. The aim is not to win but to reach a proposal that serves both sides.
The free box with the other side’s position is not decoration: until you can write it in a form your opponent accepts, you have not understood the disagreement. See Charitable reading.
Which one to use
| If what you have is… | Start with |
|---|---|
| a judgement about whether something is right or wrong | Moral argument |
| a claim that looks too obvious to question | Toulmin, to expose the warrant |
| a thesis best defended by the absurdity of its opposite | Reductio ad absurdum |
| a talk or an opinion piece to prepare | The classical model |
| a disagreement with someone you want to keep on side | The Rogerian model |
| someone else’s text, to take apart | no model — follow the eight steps |
Part III
The theory of argument maps
What the mapping tradition has learned about taking reasoning apart — rules that hold with software or with a pencil.
What prose hides
Writing runs in one dimension. An argument does not: it branches, it nests, it has parts that work together and parts that work apart. Squeezing a branching structure into a line of sentences means the connections have to be carried by small words — therefore, but, after all — and by the reader’s willingness to hold everything in mind at once.
Most of the time that works. It stops working exactly when the argument gets interesting: when there are four reasons and two of them share a premise, when an objection targets a step rather than a claim, when the decisive premise is the one nobody wrote.
Laying the argument out in boxes forces three things into the open:
- Which claims are actually being made. Every box has to be a sentence that can be true or false. Rhetorical questions, allusions and hedges do not survive the translation. A surprising number of texts turn out to have less content than their prose suggests.
- What holds up what. On a map, a claim is either connected to something or it is not on the canvas. There is nowhere for a sentence to sit merely looking relevant.
- What is missing. Prose glides over gaps, because the reader supplies the missing step without noticing. A map cannot glide: the premises are all there, and if the step to the conclusion needs something written in none of them, the absence shows.
There is a cognitive reason for this, not only an aesthetic one: processing capacity is limited, and off-loading part of the information into a diagram frees attention for what matters. Van Gelder’s studies of intensive mapping courses record gains in critical thinking of about 0.8 of a standard deviation — more than twice the effect of a conventional course.
The parts of an argument
| Part | What it is |
|---|---|
| Conclusion (thesis) | The claim the whole argument is aimed at, and which nothing else uses as support. There is only one. |
| Reason | A set of claims working together to hold up a conclusion or a sub-conclusion. A reason may be one box or several. |
| Objection | A reason against. Same structure as a reason, opposite direction — and it can have reasons of its own beneath it. |
| Rebuttal | An objection to an objection. It does not push the conclusion back up: it only removes what was pulling it down. |
| Inference | The step from the reason to the conclusion. It is not a box — it is the line. This is where most arguments fail. |
| Intermediate conclusion | A box that is the conclusion of what is below it and a premise of what is above. Perfectly normal in long arguments. |
One distinction worth fixing: the top box holds the conclusion of the argument you are mapping, not the opinion of whoever wrote the text. Authors spend a great deal of time setting out arguments they intend to demolish. When a text says “one could argue that X, because A and B — but this fails, because C”, the map has X on top, held up by A and B, with C attached as an objection.
What counts as a claim
Every box holds exactly one claim: a complete sentence that is true or false. The rule looks pedantic, and it is the one that costs the most work — and pays the most. The test: read what is in the box and ask whether it makes sense to answer “that’s true” or “that’s false”. If neither fits, it is not a claim.
| Not a claim | Why | A claim |
|---|---|---|
| Free public transport | A topic. Nothing is asserted. | Cities should make public transport free to use. |
| Is congestion the real problem? | A question. | Congestion is not the main cost of car use in dense cities. |
| Think of the poorest passengers. | An instruction. | Fares take a larger share of the income of the poorest passengers. |
Four practical rules, all with the same root:
- One claim, not two. A box reading “recordings are cheap and lecture theatres are expensive” is two claims. Split it: a reader can accept half and reject the other, and if they share a box there is nowhere to attach the objection. The reliable signal is a conjunction — and, but, although.
- No reasoning inside the box. Words like therefore, hence, because, so describe relations between claims. On a map that work is done by position and colour. If your source says “X, therefore Y”, that is two boxes with Y above X.
- Every box stands alone. No this, such policies, the former. They are natural in prose, where the antecedent is a line above, and treacherous on a map, where boxes move.
- Say the same thing the same way. When a term appears in more than one claim, use exactly the same wording. Varying your vocabulary is good prose style and bad mapping: shared wording is the visible proof that two claims connect.
The assertibility question. Every reason you write has to answer this question about the box above it: how do we know that is true? If the box you just created does not answer that question, either it is not a reason, or it is hanging in the wrong place.
Independent or linked
This is the distinction beginners get wrong most often, and the one that changes the drawing most. Two reasons are independent when each holds up the conclusion by itself; they are co-premises (linked reasons) when only together do they do so.
The removal test. Cover one of the boxes. If the conclusion still gets some support from the one that is left, they are independent; if what is left says nothing at all about the conclusion, they are co-premises. Note that the relation can be asymmetric — one premise holds without the other, and the other does not hold without it; in that case they are linked.
The commonest linked pair is a fact and a principle: the fact alone is not normative, the principle alone says nothing about the case. That is exactly the shape of the moral argument.
The golden rule. Every argument has at least two co-premises — even when only one was written. The second is usually so obvious that nobody says it out loud, and that is precisely where the assumptions that do not hold are hiding.
What nobody wrote
It is very rare for an argument to have all its premises explicit: the writer counts on what they take to be shared. Finding those missing premises is most of what mapping is for, and there are two mechanical rules that make them appear.
- The rabbit rule (vertical, between a conclusion and its reasons): no conclusion comes out of nowhere — you cannot pull a rabbit out of an empty hat. Every term in the conclusion has to appear in one of the premises. If it does not, a premise is missing, and you already know one word it has to contain.
- Holding hands (horizontal, between co-premises): once the rabbit rule has been applied, the terms left over have to appear in more than one premise of the same reason. No term is left on its own.
An example: “Art has to represent the world if it is to appeal to a broad audience. That is why Blue Poles will not appeal to a broad audience.” The conclusion contains “Blue Poles”, which is in none of the written premises — so a premise is missing, and it has to be about Blue Poles. Reconstructed, it is “Blue Poles does not represent the world”. Making it visible is what allows it to be discussed.
Two cautions when writing a missing premise: say the least you need for the step to work, and do not write empty bridges. A box that says “and therefore the conclusion follows” adds nothing — it repeats the line that is already there.
And mark it as yours. The old convention was to write it in square brackets; the one Antitese uses — like MindMup and Because — is the dashed outline, from the icon in the box header or with I. Anyone reading the map sees at once where the text ends and the reconstruction begins, which is precisely where the discussion usually has to go.
Eight steps for mapping a text
Mapping a text you did not write is the hardest and most useful skill. The temptation is to start dragging boxes; it works better to follow an order.
- Number the claims in the text, making sure each is a complete, single sentence. Ignore the filler — “moreover”, “clearly”, “who would disagree?”.
- Find the conclusion. Ask “what’s the point?”, or say to the text “convince me”. Indicators help (therefore, hence, it follows that) but are not reliable: conclusions turn up in the first sentence as often as in the last.
- Cut the redundancy. Out go questions, exclamations and repetitions; what remains are stand-alone claims.
- Merge the claims that make the same point into one box. Two wordings of one point are not two reasons.
- Check the abstraction. The higher up the map, the more general the claim; the further down, the more specific.
- Balance each level. Reasons side by side should be of similar generality. See Abstraction and level.
- Apply the rabbit rule and holding hands to make the missing premises explicit.
- Decide what is linked and what is independent, with the removal test — and only then rate anything.
You do not need to know the conclusion before you start. If the text is confusing, put a rough placeholder in the top box, build the reasons you are sure of, see where they point and rewrite the top. The reasons often name the conclusion better than the author did.
Abstraction and level
Two guidelines that tidy confused maps without changing their content.
The principle of abstraction (vertical): the higher the claim, the more abstract. “Dogs make better pets” is more abstract than “you can play with dogs”, which is more abstract than “dogs fetch balls”. Those three are not parallel reasons: they are a chain, each holding up the one above.
Level consistency (horizontal): reasons that share a level should carry similar weight. When a very specific reason turns up beside a general principle, the specific one almost always belongs a level below, holding the general one up.
A map that respects both reads from the top down like a zoom: each step down adds detail, and no row mixes orders of magnitude.
Where arguments stop
A map has to end somewhere. The good place to end a branch is a source of evidence the parties will accept without further argument: a datum, a study, testimony, a court ruling, an expert opinion, your own experience. That is what the evidence box is for.
What counts as an acceptable stopping point depends on the context and is always provisional. Accepting your housemate’s word that there is no milk in the fridge is reasonable; accepting without more that someone is a reptilian alien is not. If you later think you closed a branch too early, open it: an evidence box can acquire reasons beneath it, and at that moment it stops being an endpoint.
Ideally, every premise either has reasons below it or ends in evidence. A premise with neither is simply being asserted — which is legitimate, as long as it is visible that this is what is happening.
From structure to judgement
First you understand the structure, only then do you criticise. A map does not claim its boxes are true: it shows what would have to be true, and what depends on what.
With the structure built, rating rests on two separate questions — the same two the app asks of every box and every connector:
- Is this claim true? (degree of truth, 1 to 5)
- If it is true, how much does it hold up the one above? (strength of the inference: leap, supports, conclusive)
Keeping them apart is half the value of the method. A rock-solid premise joined by a leap holds up very little; a conclusive inference from a doubtful premise does not do much either. The map forces you to say which of the two problems the argument has.
Three consequences that surprise beginners:
- Counting reasons gets you nowhere. Twenty weak reasons are not worth one good one. A bad reason does not subtract — it is irrelevant. When you rate, look at the strongest reason, not at the number.
- A defeated objection adds nothing. The rebuttal only restores what the objection was taking away; it does not push the conclusion above where it already stood.
- Doubt travels upwards. A poorly supported premise drags everything resting on it with it, however many healthy branches the map has alongside.
The exact arithmetic Antitese uses — how several reasons accumulate, why a bracket is worth its
weakest link, how much an objection weighs — is described in MODELO-DE-CALCULO.md,
next to the code. The numbers are an aid to reading, not a verdict: what decides is the argument,
not the bar.
Charitable reading
Always map the strongest version of the argument you are analysing, not the one easiest to attack. Where the text is ambiguous, take the reading that makes it stronger; where a premise is missing, write the most plausible one that makes the step work.
This is not good manners: it is self-interest for anyone who wants to criticise. An objection to the weak version proves nothing — your opponent replies that this is not what they meant, and they are right. An objection to the strong version is a real objection.
Charity is not agreement either. You can still conclude that the argument does not hold; the difference is that you now know exactly where. And it applies to your own work: mapping what you defend and then looking for the strongest objection you know of is the most uncomfortable and most productive use of this tool.
What a map does not do
A map is not a summary, nor a substitute for reading. It records the logical skeleton and throws away the tone, the examples, the qualifications and nearly everything that makes a text worth reading. That loss is the price of clarity.
It also decides nothing. It shows the shape of the argument, which claims carry the weight and where the joints are exposed. Whether the premises are true remains a question about the world, and no diagram will answer it.
A useful comparison: a map is to an argument what a circuit diagram is to a circuit. It leaves out the colour of the wires and tells you exactly what is connected to what — which is what you need when something does not work.
And do not confuse an argument map with a mind map or a concept map: those represent associations and relations between ideas, this one represents only inferential relations — what holds up what.
Sources
- Martin Davies, Ashley Barnett and Tim van Gelder, Using Computer-Aided Argument Mapping to Teach Reasoning, in J. Anthony Blair (ed.), Studies in Critical Thinking (Windsor Studies in Argumentation), CC BY 4.0. The eight-step procedure, the principles of abstraction and level consistency, and the discussion of evidence sources come from here. Read it.
- Jamel Ostwald, Argument Mapping — The Basics, sheets based on the heuristics of the Rationale software (Austhink). The assertibility question, the rabbit rule and holding hands come from here. Read it.
- Stephen Toulmin, The Uses of Argument (1958), the source of the claim/data/warrant model.
- The teaching tradition behind the shape of these maps — boxes, brackets, colours — comes from the Princeton seminar Philosophical Analysis Using Argument Maps (Simon Cullen, Adam Elga and Shamik Dasgupta) and from MindMup.