Behavioral Metrics · 11 min read
Repair Attempts in Text Message Records
A repair attempt is a message that tries to lower the temperature. Counting them changes what a conflict thread looks like, and most message reviews never look for them.
The short answer
A repair attempt is a message that tries to interrupt an escalating exchange: an apology, a concession, a change of subject away from the conflict, an offer of a call, a de-escalating acknowledgement. The concept comes from observational research on couple conflict, where the presence and reception of repair attempts distinguishes arguments that resolve from arguments that spiral. Textimony surfaces candidates for these with an analyzer named repair_attempts, version 1.3. The reason it belongs in an evidence review is that almost nobody looks for it. A review built only on harassment, threats, and insult categories will faithfully return every hostile message and produce a thread that appears uniformly hostile, because that is the only question it asked. Whether either participant tried to stop the exchange, and what happened when they did, is a different question with a different answer, and it is frequently the one that changes how a period reads.
Why one-sided extraction distorts a record
Selection is the quiet problem in message evidence. Pull every angry message from a two-year thread and you have a document that is completely accurate and substantially misleading, because the same procedure applied to any long relationship produces a similar-looking file. The messages are real; the impression is an artifact of the query. Federal Rule of Evidence 106 addresses a version of this directly: when part of a writing is introduced, an adverse party may require introduction of any other part that in fairness ought to be considered at the same time. A repair-attempt pass is the practical form of that principle applied before anyone else raises it. If a thread contains de-escalation attempts, discovering them during your own review is materially better than discovering them during cross-examination.
What counts as a repair attempt
The recognisable forms are narrower than "anything conciliatory". Direct apologies are the clearest. Concessions that give ground on the disputed point are next. Acknowledgements that reflect the other position back without conceding it -- "I hear that you are upset about the pickup time" -- function as repair even when nothing is granted. Procedural moves also qualify: proposing a call, suggesting the conversation resume later, or explicitly asking to stop. These are easy to miss in a keyword review because they contain no emotional vocabulary at all. A message reading "can we talk about this tomorrow" carries no sentiment a lexicon would flag, and in a spiralling thread it is often the most consequential message in the exchange. The category is deliberately reported as candidates rather than findings, because the same words do different work in different threads. "Sorry" can be an apology, sarcasm, or a preface to a renewed accusation, and only the surrounding messages settle which.
Reception matters more than count
The count on its own is close to meaningless. What carries information is what happened next. A thread where repair attempts are made and reciprocated looks like ordinary conflict between people who disagree. A thread where they are made repeatedly by one participant and consistently rejected, mocked, or ignored looks like something else -- and that difference is visible in the record without anyone characterising a person. This is why repair attempts are read as pairs: the attempt, and the following messages from the other participant. Reviewing them in isolation reproduces exactly the selection problem the category exists to correct, and a repair attempt lifted out of the exchange that answered it is as misleading as a hostile message lifted out of what provoked it.
What this does not establish
It does not diagnose a relationship, and the research the concept draws on was built to describe interaction patterns rather than to support legal conclusions about individuals. A message record processed by software is not a clinical assessment and should never be presented as approaching one. It also does not settle responsibility. A participant who made more repair attempts was not necessarily the more reasonable one, and a participant who made none may simply have left the conversation, which is itself a form of de-escalation this measurement cannot see. The value is that it forces a review to ask a second question about a thread instead of confirming the first one.
What is a repair attempt in a text conversation?
A message that tries to lower the temperature of an escalating exchange: an apology, a concession, an acknowledgement of the other position, a proposal to talk later or by phone, or an explicit request to stop. The concept comes from observational research on how couples handle conflict.
Why look for de-escalation messages in evidence review?
Because a review that only searches for hostility returns only hostility, and that file looks similar for almost any long thread. Finding the de-escalation attempts tests whether a period was one-sided escalation or ordinary conflict, and it is far better to find them during your own review than to have them produced by the other side.
Does a record of apologies mean someone admitted fault?
No, and treating it that way is a serious misreading. People apologise to end arguments, to keep contact, and out of habit as often as to concede a point. A repair-attempt candidate marks a message worth reading in context, not an admission of anything.
How does Textimony identify repair attempts?
The repair_attempts analyzer, version 1.3, surfaces candidate messages that match recognisable de-escalation forms and keeps each one linked to its source. They are reported as candidates for a person to read, because identical wording can be sincere, sarcastic, or a preface to renewed conflict.
Is this relevant under Rule 106?
It is the same underlying concern. Rule 106 lets an adverse party require other parts of a writing that in fairness should be considered alongside what was introduced. Running a de-escalation pass over the same thread is the practical way to know what those parts are before someone else identifies them.
Does a higher repair-attempt count favour one participant?
Not on its own. Volume of attempts says nothing about who was reasonable, and a participant who simply disengaged may have de-escalated in a way no message-level measurement can detect. What carries information is whether attempts were reciprocated, rejected, or ignored.
Published by
Textimony. Editorial status: Source-linked informational guide. Updated: 2026-07-30.
Sources
Federal Rule of Evidence 106: Remainder of or Related Statements; Federal Rule of Evidence 403: Excluding Relevant Evidence; Observational research on marital conflict and repair (PMC)