Communication

Argument Pattern Decoder: Read the Four Horsemen, Then Repair

Most couples can tell you that an argument went badly. Far fewer can say what pattern ran through it. The Gottman research tradition gives four recurring shapes to conflict: criticism, contempt, defensiveness, and stonewalling, known as the four horsemen. This decoder takes a recap you paste in, counts the word patterns behind each horseman, and reports how many sentences carried at least one. It then hands back three repair moves aimed at the patterns that actually showed up, ordered by how much damage each one tends to do. The tool reads words, not people. It cannot hear tone, guess intent, or tell you anything about the relationship as a whole. What it can do is take a vague feeling such as that conversation felt awful and turn it into a list of sentences worth rewriting.

Paste the recap of the argument

Four to twelve sentences, written the way it actually happened. One message per paragraph helps the sentence counter.

Whose lines did you paste

When do you want to use the repair moves

Result

How to use it

STEP 1

Paste the recap in your own words, four to twelve sentences long. Write what was actually said, not what you concluded from it. The decoder matches phrases such as you never, not my fault, and no comment, so a paraphrase in your own vocabulary will hide more than it reveals. If you do not remember the exact wording, write the closest thing you can recall and treat the counts as approximate.

STEP 2

Answer the two short questions. Whose lines you pasted decides whether the read is about one person or about the exchange, and whether the repair moves are written for right now, for later today, or for preparation. The moves are the same three ideas rewritten in three registers, so pick the timing that matches the room you are actually in.

STEP 3

Read the marker counts, not the percentage. The percentage tells you how much of the recap carried a pattern; the counts tell you which pattern. A recap with two contempt markers and zero repair phrases is a different situation from one with two criticism markers and two repair phrases, even though the flagged share is identical.

STEP 4

Take one repair move, not all three. Saying the first thirty seconds of a move out loud, in your own words, beats reciting three paragraphs. Repair attempts land when they are specific, short, and aimed at the pattern rather than at the person who produced it.

Three common mistakes

Pasting the argument itself instead of a recap. The decoder reads words, not tone or history. A wall of text lifted from the worst night of the year will flag nearly every sentence and teach you nothing, because you will have read your own worst hour as if it were the pattern of the relationship.
Treating one hit of fine or whatever as proof of stonewalling. Those are ordinary words. One occurrence is weak evidence, five of them in a single message is a pattern, and the same word in a calm message is just a word. Look at the count inside one message, not at the presence of a term.
Scoring the other person with it. The tool cannot tell what anyone meant. Contempt counted on one line might have been sarcasm aimed at a third party, and defensiveness counted on one line might have been a correct correction of a misheard sentence. The counts describe the message in front of you, never the character of the person who wrote it.
Reading the marker total as a score for the relationship. It measures one message. A relationship is a sample of hundreds of messages, and the same four patterns show up in healthy couples too. The number that matters is not this one but how often the same row repeats across weeks.

The four horsemen and what the decoder is actually counting

The framework comes from decades of observational research on couples, where four recurring patterns showed up again and again in the recordings of couples who later struggled. Criticism attacks behaviour but generalises it into identity: you never, you always, every time. Contempt attacks character or presence: ridiculous, useless, pathetic, and the sarcastic versions of the same thing. Defensiveness counter attacks or defends intent: not my fault, that is not what I said, you started it. Stonewalling shuts the conversation down: fine, whatever, I am done, going to bed, no comment.

The decoder counts phrase matches per sentence, so one pattern cannot inflate the count by repeating. In the default example the recap scores 4 criticism markers, 2 contempt markers, 3 defensiveness markers and 5 stonewalling markers across eleven sentences, and ten of those eleven sentences carry at least one pattern. Contempt is weighted highest when the three repair moves are ranked, not because it is the most common, but because contempt toward a partner is the pattern most strongly associated with later deterioration in the research literature. A pattern that shows up twice and one that shows up five times can both be worth naming, but they are not worth the same amount of attention.

Two things the decoder deliberately does not do. It does not count tone, because text has no tone. And it does not count what was missing, which in a text recap is often the most important part of the exchange.

Why the flagged share beats the marker total

Totals reward volume. A person who writes a long message will out-count a person who writes two short ones, even if the short ones are worse. That is why the headline number here is a share rather than a sum: 90.9 percent of the sentences in the example carry a pattern, which tells you the pattern is spread through the whole exchange instead of concentrated in one corner of it.

The flagged share answers how consistent the pattern is. A single flagged sentence in a long message is an incident. Nine flagged sentences in ten is a weather pattern.
The repair count is the opposite metric, and it is the one most people skip. Repair phrases such as sorry, my fault, I hear you and you are right are counted separately. An exchange with fourteen patterns and zero repair phrases is a different situation from one with fourteen patterns and four repair phrases, because the second one is still answering the question.
The longest sentence is a pressure gauge. Above roughly twenty words in one breath, the other person stops hearing content and starts hearing pressure, and the conversation tends to escalate from there.

Read the three numbers together. The share tells you how widespread the pattern was, the repair count tells you whether anything landed, and the longest sentence tells you where the room got tense. One number alone will mislead you, and it will usually mislead you in the direction of your own mood that evening.

Turning a read into an actual repair

A pattern read is worth nothing unless something changes on the next attempt, so the tool always ends with three moves instead of a diagnosis. Each move is aimed at one of the patterns that actually appeared, and each one is written in the second person, ready to say out loud.

Three things make the difference between a repair move that works and one that becomes another round of the same fight. First, specific beats global: a sentence about a specific Tuesday afternoon can be answered, while you never cannot. Second, impact before intent: naming what the other person landed on removes the need to defend yourself, and most defensive counters exist because the other person still feels unheard. Third, a break needs a return time: twenty minutes with a stated return time is a pause, silence without one is a withdrawal, and the difference in how the other person experiences it is enormous.

Do not try all three moves in one conversation. Pick the one aimed at the pattern with the most markers, say the first thirty seconds of it, and then stop talking. Repair attempts that are short, specific, and followed by actual silence are the ones people remember as the moment the fight stopped.

FAQ

What if my recap comes back with no markers at all?

The decoder only knows word patterns, so a zero result means one of three things: the recap was genuinely neutral, the recap was written in language the lexicon does not cover, or the recap describes the argument in your own summary words instead of the words that were used. The third is the most common. Try pasting the two sentences you remember most clearly and nothing else, and see whether the numbers move. A zero result is not a clean bill of health, and a full set of markers is not a verdict. Neither one measures the relationship.

Is contempt just criticism with worse manners?

The difference is the target. Criticism points at behaviour: you never think ahead, you were late again. Contempt points at the person: you are useless, that was ridiculous, honestly. The decoder separates them by lexicon, so swapping an insult for a description of the event drops the contempt count while leaving the criticism count intact. Sarcasm counts as contempt in this tool, since sarcasm aimed at a partner lands as a character attack even when no slur is used. If you are unsure which one you heard, read the sentence aloud without your voice and ask whether it describes something that happened or something that someone is.

Does a high count mean the relationship is in trouble?

No, and it is worth resisting that reading. A single message can contain all four patterns and still be an ordinary bad hour, especially after a long day, a bad day at work, or a month of poor sleep. The four horsemen are common in relationships that work fine, because they are the default moves people make when they feel cornered. What matters is repetition across many messages and over months, which is exactly what one pasted recap cannot show. Use the tool as a prompt to look at your own track record, not as a measurement of the relationship.

Can I paste my own apology and check it before sending?

Yes, and it is one of the more useful ways to use the page. Repair phrases are counted separately from horseman patterns, so a drafted apology shows you how many repair elements it carries and how many horseman patterns slipped in. An apology that contains you always, or that explains intent before naming impact, will show both a repair count and a criticism count at once. Read the flagged share as well: a long apology with every sentence flagged may be long enough that the other person stops reading before the repair lands.

What about threats, intimidation, or controlling behaviour?

This page is not built for that situation, and you should not use it that way. Scoring a threat as a stonewalling marker would be treating intimidation as a communication problem, which it is not. If a recap contains threats, intimidation, damage to property, monitoring of your movements, or pressure about money, the useful next steps are outside this tool: talk to someone you trust, and look at local support services for domestic abuse and coercive control. A pattern score is not a safety plan, and no tool on this site can be one.

Limits of this tool

Limits of this tool: it matches word patterns in the text you paste. It cannot hear tone, read intent, judge who was right, or see anything outside the pasted message.
This is a self-observation tool. Nothing here is a diagnosis, and no count is a verdict on a person or on a relationship.
Counts are not comparable between two people. Some people write long, warm messages with soft wording, and some write short, blunt ones, so the same pattern can look very different in each style.
Nothing you paste is stored or transmitted. The text is processed inside your browser, and the page does not send it anywhere.