When did things change? Finding the week, not guessing it
WholeThing finds the week a chat changed by counting messages per week across a whole WhatsApp export and trying every possible split into before and after. It keeps the split that explains the most, and only reports it when weekly messages rose or fell at least 1.6 times, with at least four weeks on each side.
How it finds the week
Most people who ask “when did things change?” have a feeling, not a date. Texting seems to have slowed down, or picked up, sometime last spring. Scrolling back doesn’t help much: thousands of messages don’t look different from one another.
WholeThing starts by counting messages per week, every week from the first message to the last, Monday to Sunday. Weeks with no messages at all are kept in as zeros. Leaving them out would make every change look sharper than it was.
Then it tries every way to cut that series in two, a before and an after, and asks the same question each time: if the chat really were one steady level before this week and another after it, how much of the week-to-week ups and downs would that explain? The split that explains the most is the candidate.
An example: if a two-year chat sat between 40 and 55 messages a week for a year and then jumped above 400, the week of that jump splits the series almost perfectly in two. If the weeks keep bouncing between 20 and 80, no split can explain the bouncing. There’s no turning point there, just a chat with ups and downs.
The counts are compared on a log scale, so a chat going from 5 messages a week to 50 weighs the same as one going from 50 to 500. A conversation halving is the same size of event, whatever its volume. Statisticians call this kind of search change point detection. WholeThing uses its simplest form: exactly one change, and every possible week tried.
The bar a change has to clear
This is the finding most likely to tell you something you didn’t know. It’s also the easiest one to make up, so the bar is high on purpose. A chat that slowly drifted has no turning point, and saying otherwise would be inventing a story.
- The chat has to span at least 12 weeks.
- Each side of the split needs at least 4 weeks.
- The split has to explain at least 25% of the week-to-week variation.
- Messages per week have to be at least 1.6 times higher or lower after the split than before.
- It then has to clear a higher confidence bar than most cards, because it claims a pattern rather than reporting a count. In practice, that means at least five weeks on the shorter side and a split that explains more than a third of the variation.
If nothing clears all of that, there’s no turning-point card. Fewer strong findings beat many weak ones.
What the card shows
The card names the month the change landed in and which way it went: “Something changed around March 2025. You started talking far more.” Or, when it went the other way: “Something changed around March 2025. The conversation thinned out.”
Under it, Before and After show the average messages per week on each side of the split, for example 45 per week before and 445 after. Those are plain averages of the real weekly counts, so you could check them against your own chat.
Something changed around March 2025. You started talking far more.
WholeThing also runs the same search on a second series: the typical reply time each week, using weeks with at least five replies. Both go through the same bar. If reply times split more cleanly than message counts, the card is about replies instead: “Around March 2025, replies started taking longer.” If they tie, message volume wins, because a chat going quiet is easier to recognize than a reply time drifting.
How it differs from the other change cards
Three cards in a report look at change, and they answer different questions:
| Card | Looks at | Answers |
|---|---|---|
| Turning point | The whole chat, week by week | When did the conversation shift most? |
| Reply rhythm | Each person’s typical reply, period by period | Did one of you start answering slower or faster? |
| Busiest era | Messages per month | Which month was the busiest? |
They can agree, and that’s often the interesting part: a turning point in March and one person’s reply time changing the same season. The longest silence is a different kind of change again: one stretch with no messages at all.
Which chats it works for, and when it’s free
The turning point needs a full WhatsApp export, ideally the whole history. Screenshots don’t carry dates for each message, and a few screens can’t show months of weekly counts anyway. The export guide shows how to get every message out on an iPhone.
Two findings in every report are free: the strongest one, and the one that answers your question. The turning point is the direct answer to “When did things shift?” for a situationship and “When did things change?” for an ex, and one of the direct answers to “Is this changing over time?” and “Has our rhythm changed?”. It’s usually the opening card when you pick “Show me everything” for an ex. When it isn’t free, it unlocks with the conversation, or with a 7-day pass for every conversation you start analyzing that week.
What it can tell you, and what it can’t
It can tell you roughly when the chat’s rhythm moved, which way, and by how much, with numbers you can check. For a lot of people, the date is the part that lands: it’s often a week they remember for some other reason.
It can’t tell you why. “We don’t text as much anymore” can mean you started seeing each other every day, moved in together, switched to calls or moved to another app. None of that is in a WhatsApp export. The week is where the split fits best; a change that happened gradually over a few weeks still gets one week, which is why the card says “around”. And it finds one change, the biggest. A long chat can have had others.
An unlocked card closes with one general line, never about the two of you: “Long chats often have a moment like this, usually when something changes outside the chat.” How it works explains the rest of the report.
Questions
- What if our texting slowly faded instead of changing at once?
- Then there’s usually no turning-point card. A slow drift doesn’t split cleanly into a before and after, and WholeThing won’t invent a date for it.
- Can it find more than one change?
- No. It reports the single week where the chat shifted most, if that shift is big enough. Reply rhythm and busiest era can show other changes.
- Does it work with screenshots?
- No. It needs dates for months of messages, which only a full WhatsApp export has.
- Will it tell me why things changed?
- No. It shows when and by how much. The reason is usually outside the chat, where no count can reach.