> ## Documentation Index > Fetch the complete documentation index at: https://docs.trackplay.io/llms.txt > Use this file to discover all available pages before exploring further. # Retention > The second-by-second curve of who kept watching, built from each session's furthest point, with the moments that mattered marked on top of it. Play counts and average watch time never tell you which second lost the sale. The Retention curve does: one line, built from where every session actually stopped, with your CTAs, your Pitch, your conversions and the key moments of your script drawn on top of it. Open a video and click **Retention**. ## The curve TrackPlay buckets every session into fixed 5-second windows of the video's timeline, by that session's **furthest point reached**, not by how long it watched or whether it rewound. A viewer who skipped ahead to 2:00 and never went back still counts at 2:00. The curve is cumulative: bucket by bucket, from the end of the video backward, it counts every session that reached at least that point. That cumulative count is then scaled against its own peak, the count at 0:00, so the curve always starts at 100%. **Retention reads as relative to your own peak, not as a percentage of total plays.** If 2% of sessions never produced a single progress tick (a page view with no play, for instance), the curve does not know about them: it starts at 100% at the count it actually has data for at 0:00, and every later point is relative to that. **The plain "All viewers" curve, with no cohort and no filter chip active, is not scoped to the date range picker.** It reads all-time from the underlying progress data. The moment you select a cohort (Buyers vs non-buyers, Buyers only) or apply any attribution filter chip, the curve becomes properly bounded by your selected date range, because those paths carry the date window with them. If you are comparing a specific period, select a cohort or a filter chip rather than reading the default curve. ## Hover for detail Move across the curve and a panel updates with, for every line on the chart: the time, the retention percentage, the raw viewer count behind it, and how far that point has dropped from the series' own peak (shown once the drop passes a tenth of a point). Hovering also seeks the preview player next to the chart to that same second, so you can watch what is happening at the point you are reading. ## Drag to zoom and loop Drag across a span of the timeline to zoom the chart into it. TrackPlay rescales both axes to the visible data and starts the preview player at the start of your selection, looping it back to that start every time it reaches the end of the selection. This is a local, in-browser loop of the preview clip next to the chart. It does not change any setting on the live player your viewers see. Click Reset to return to the full curve. ## Cohorts Three ways to view the same curve, plus a fourth that appears only when comparing videos: | Cohort | Shows | | - | - | | **All viewers** | Every session, the default view. | | **Buyers vs non-buyers** | Two overlaid curves, split by whether the session converted. | | **Buyers only** | The buyer curve alone, isolated. | A session counts as a buyer when it has a real, non-deleted conversion recorded against this video. If your selected period has zero buyers, TrackPlay disables the buyer-based views with a note explaining why, rather than plotting an empty line. Overlay buyers against non-buyers and look for the point where the two lines separate. That is the moment your video sells. Whatever happens right before it is worth protecting from every future edit. ## Conversion density A row of bars along the bottom of the chart shows where your buyers' watching actually maxed out: each bar is a count of buyer sessions whose furthest point landed in that 5-second window. Unlike the curve above it, this is not cumulative. It is a plain histogram of buyers' stopping points, so a tall bar marks a real cluster of purchases tied to that moment, not everyone who watched past it. ## Markers: measurements versus annotations Four kinds of markers sit on the timeline, and they answer two different kinds of questions. When your video has a Pitch Time, the Pitch marker is the median of a dedicated event, `pitch-time-reached`, fired once per session by the player when it crosses it. TrackPlay takes the exact median of every session's reported pitch time and draws one line at that second. This is a measurement of actual viewer behavior. When there is no measurement yet, TrackPlay still draws the line. It uses the Pitch Time you set, and if you set none, the second your script turns from content to selling, which TrackPlay reads from the transcript. That line is labelled **Pitch (detected)** so you always know which kind you are looking at. Each CTA gets a numbered marker at the median second it appeared, alongside two counts: how many sessions it was **shown** to, and how many **clicked** it. A CTA configured on the video but never yet shown to anyone still appears on the timeline, at zero, so you can see it is wired up correctly before traffic proves it out. Named markers for your configured timed events are read directly from the video's own settings, not computed from traffic. TrackPlay draws them at the position your configuration says they should be, whether that video has one session or none. This is the opposite of the Pitch marker: it is true by definition, not by measurement. A brand-new video with a timed event configured and zero plays will still show that marker exactly where you placed it. The chart itself only renders once the video has at least one session with playback data, so on a genuinely empty video you will see an empty state instead of a chart carrying nothing but annotations. Every video with a transcript gets its key moments marked automatically: the hook, each story and case study, the teaching sections, the offer, the price, the bonuses, the guarantee, the close. TrackPlay reads them from the transcript as soon as it exists, labels each one the way you would ("Case study #3", "Bonus stack"), and places it at the second those words are spoken. You do not set anything up. If you turned on chapters for the video (Customize Player, Chapters), your own chapters are drawn instead, because you know your script better than we do. Hover the curve and the panel beside it names the section you are in, so a drop reads as "we lost them during the second case study" rather than "we lost them at 21:15". They are drawn as slate hairlines, next to the teal timed-event markers. Where a key moment lands within 10 seconds of a timed event, only the timed event is labelled. The **Key moments** (or **Chapters**) button in the chart header hides them. When the transcript changes (a re-transcription or a correction), the key moments are read again within the hour. ## Where people who reached the offer drop off Below the main curve, a second chart takes only the sessions that reached your offer and splits them into the ones who bought and the ones who left. The whole-audience buyers-versus-non-buyers view cannot answer this, because most of its non-buyers never got to the offer at all. "Reached the offer" means one of two things: * **Shown a CTA**, when the video has CTAs. * **Watched past the Pitch**, when it has none. The Pitch is the one described above, measured or detected. The pitch-based version only appears once the video has recorded at least one purchase. With no buyers there is nothing to split, and the main curve already shows the rest. ## Comparing by device, orientation, or source Break the curve into segments instead of one line, by **device**, **orientation**, or **traffic source**. TrackPlay ranks segments by session volume and draws up to 5, disclosing how many smaller segments and sessions were left out rather than hiding them silently. A faint dashed baseline shows the total curve behind every segment for reference. Orientation is the one dimension where a session can land in two segments at once: a viewer who rotates their phone mid-playback appears in both portrait and landscape. TrackPlay explains this on the chart itself when it happens, rather than letting the segment counts silently fail to add up to the total. ## Overlaying multiple videos Click **Add Video** to bring another video's retention curve onto the same chart, up to 7 at once, each in its own color. Every overlaid video uses the same cohort and date range as the one you are viewing, so the comparison stays apples to apples. This mode and segment comparison are mutually exclusive: turn one on and the other's controls step aside. ## CSV export Export writes one row per 5-second bucket: the time, then a Users and a Retention % column for the main curve, followed by the same pair of columns for every segment or every overlaid video currently on the chart. Nothing is re-queried for the export. It is built from exactly what the chart is already showing you. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.