Features

Nine things Circin does, explored.

Circin has more features than these nine. These are the nine worth slowing down for — the ones where what the thing does and what anyone can honestly claim for it are two different sentences. Each one below is followed as far as the code and the evidence will go, and then stopped out loud: how it actually works today, what you can do with it, why it might help with the study that suggests so, what that study cannot say — including where it points away from the choice Circin made — and where the feature runs out. Watching one video and closing the app is a complete way to use Circin. Nothing here is something you are expected to do.

No study on this page is a study of Circin. Nobody has evaluated this app. Everything cited here is evidence about a mechanism — curiosity, explaining, returning to something, checking a source — found in other settings, with other materials, mostly in laboratories and classrooms. It is a reason a feature is shaped the way it is, not proof that the feature works. Where a claim would need Circin’s own data, Circin does not have any.

Badge on a card	What it means	What it still cannot say
Meta-analysis	Many studies pooled into one estimate	Whether it holds for short video, or for Circin
Replicated experiments	Controlled experiments, found more than once	Much beyond the materials and settings tested
Single study	One study, one sample	Whether it would be found again
Theory	A framework that organises earlier work	Anything on its own; it is not a test

Swipe the table →

KeyEvery research card below carries one of these four badges. A badge describes the study, never Circin: none of it was run on this app, and each card names its own limit under “What it cannot say”.

Finding your way in

Categories: eight Subjects, fifty-four smaller rooms

Eight Subjects — Cognitive Edge, Math, Science, Language Arts, History, Technology, Civics and Current Events, Business & Economics — with fifty-four subcategories beneath them. Every one has its own page on the web: circin.org/subject/math/ and, under it, circin.org/subject/math/algebra/, fifty-four in all. In the app the Explore tab lays the eight out as tiles, and a Subject page carries a Follow button. Following a Subject moves its videos up in For You, which ranks in a fixed order and nothing else: the subcategory you starred, then the Subject you starred, then Subjects you follow, then anything you asked for more of, then the interests on your profile. You can also build a temporary mix from several Subjects and topics at once — it interleaves them rather than narrowing to where they overlap. A creator classifies a video by picking whole Subjects or specific subcategories, and may mark one as the main one, or skip the step: only a clip and a title are required to publish.

What you can do with it

You can tap a Subject on Explore, follow one from its own page, star a single subcategory as the one you care most about, or build a mix and leave it when you are done.
Settings → Manage topics lists what you have followed and unfollows any of it.
Or set none of it: For You works with nothing chosen at all.

DrawnFollowing or starring moves videos up; it never removes the rest. A drawing of the sort, not a screenshot. Following a Subject lives on the phone and travels only in an account backup: there is no server follow and no creator is told. For You works with nothing chosen at all.

The limit. Following a Subject is a device-side preference, not a server follow: it lives on that phone, travels only in an account backup, and no creator is ever told. It needs an account — a guest who taps Follow is asked for one, in wording written for that action. A mix will not open on styles or levels alone; it answers "Choose at least one subject or topic." And most of what fills a Subject page is Circin's sample library — a title and a caption on a colour, with no video file behind it, which does not play. Zero creators have joined so far, so nothing on a Subject page is a real video yet.

The research beside it

Meta-analysis

Curiosity is not only a mood in the moment. Pooled across the published literature, intellectual curiosity stands beside intelligence and effort as a direct predictor of academic performance — which is the reason Circin is built so a question can be followed rather than met once.

What it found von Stumm, Hell and Chamorro-Premuzic built path models on a meta-analytically derived correlation matrix from the literature on intelligence, personality and academic performance. Typical Intellectual Engagement — the trait marker of intellectual curiosity — was a direct, correlated predictor of academic performance alongside intelligence and Conscientiousness, and the additive predictive effect of curiosity and effort rivalled the influence of intelligence, while intelligence remained the single most powerful predictor on its own.

von Stumm, S., Hell, B., & Chamorro-Premuzic, T. (2011). The hungry mind: Intellectual curiosity is the third pillar of academic performance. Perspectives on Psychological Science, 6(6), 574–588. doi.org/10.1177/1745691611421204

What it cannot say

Correlational and trait-level. It describes people who are curious across years of schooling, not what a feed does to anybody, and its own models put intelligence first. Nothing in it is about a phone, a Subject page, or what following one does to a person.

Meta-analysis

Choosing your own Subjects is a freedom, not a method. When learners were handed control of their own path through educational technology, the pooled effect on learning was almost nothing — g = 0.05 across 29 effects from 18 studies.

What it found A meta-analysis updating Niemiec, Sikorski and Walberg searched 85 distinct articles, of which 18 met inclusion criteria, yielding 29 computed effects. The overall effect of providing learner control within educational technology was almost zero (g = 0.05), and remained near zero across most characteristics of the control offered and most classroom contextual factors.

Karich, A. C., Burns, M. K., & Maki, K. E. (2014). Updated meta-analysis of learner control within educational technology. Review of Educational Research, 84(3), 392–410. doi.org/10.3102/0034654314526064

What it cannot say

Eighteen studies is a small base, mostly formal instructional settings with defined outcomes, and "learner control" covers very different designs. A near-zero average is not evidence of harm — it is evidence that control is a courtesy, and Circin should call it that. Following one Subject also narrows what you meet as surely as it deepens it.

Explore, in the live app. Circin’s sample cards have no video behind them; a creator’s post plays.

The shape of an explanation

Content style: four shapes of explanation

Four names for what a video will ask of you. Lesson is a clear explanation of a concept; Learn Along follows steps and practises with the creator; Hands-On makes, codes, experiments or produces; Analysis evaluates ideas and applies knowledge. The wording is the app's own — it appears beside each one during setup. A creator picks one or several when publishing, and may mark one as the main style. Styles show on a video's classification row, they are a real filter in search, and they are a chip in Choose what to watch. Inside a mix a style is a ranking and not a filter: the code sorts matching videos to the front of each Subject's pool and removes nothing, and the comment in the source says so. On the Explore tab, tapping a style chip runs a search for that word rather than filtering the feed. All 128 sample cards carry a style — 37 Lesson, 32 Analysis, 30 Hands-On, 29 Learn Along.

What you can do with it

You can filter a search to one or more styles, put styles on a mix so that kind of video comes up first, pick styles during setup, or never touch them.
Somebody who only ever watches Lessons has used Circin properly.

DrawnA style narrows a search and only re-orders a mix. Levels behave the same way. A style is a word the creator picked: nothing checks that a Lesson is a lesson, and Circin never matches a style to a kind of person.

The limit. A style is a word the creator typed, not a property anything checks. Nothing requires a creator to pick one — only a clip and a title are needed to publish — and the server accepts any string up to forty characters in that field, so the four names are enforced by the app's own composer and nowhere else. A video with no style passes every style filter rather than being excluded. And Circin never matches a style to a kind of person: there is no such thing as your style here, and there is no evidence that choosing your own styles teaches anyone more.

The research beside it

Theory

What a person actually does while learning predicts outcomes better than what they are shown. The ICAP framework sorts cognitive engagement into four modes by observable behaviour — Passive, Active, Constructive, Interactive — and holds that learning increases as engagement moves up that order. That is the honest reason to distinguish a video you watch from one that asks you to do something.

What it found The ICAP framework categorises cognitive engagement into four modes by observable learner behaviour — Passive, Active, Constructive, Interactive — and its central hypothesis is that learning increases as engagement moves up that order. The paper synthesises a large body of classroom and laboratory studies as the basis for the ordering.

Chi, M. T. H., & Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219–243. doi.org/10.1080/00461520.2014.965823

What it cannot say

ICAP is a framework synthesising prior work, not a single causal test, and the ordering is contested at the boundaries between modes. It points straight away from Circin: ICAP classifies what the learner does, and Circin classifies what the video is. Nothing stops anyone watching a Hands-On video without lifting a finger, nobody is obliged to do the task, and Circin neither knows nor asks. A style label describes an explanation's form; it is not a claim about anyone's engagement.

Meta-analysis

Studying a complete worked-through explanation measurably helps people early in learning something, which is the shape a Lesson video takes: across 55 studies and 181 effect sizes, the average effect on mathematics performance was g = 0.48, roughly an 18-percentile improvement.

What it found A meta-analysis screened 8,033 abstracts from published and grey literature down to 43 articles reporting 55 studies and 181 effect sizes, covering elementary school through adulthood. Using robust variance estimation, the average effect of studying worked examples on mathematics performance was g = 0.48 (p = 0.01) — roughly an 18-percentile improvement over learners who did not study them. Moderators tested included correct versus incorrect examples, pairing with self-explanation prompts, and whether examples came during skill acquisition or during practice.

Barbieri, C. A., Miller-Cotto, D., Clerjuste, S. N., & Chawla, K. (2023). A meta-analysis of the worked examples effect on mathematics performance. Educational Psychology Review, 35(1), 11. doi.org/10.1007/s10648-023-09745-1

What it cannot say

Mathematics only, inside instruction, with follow-on problem solving — strongest during skill acquisition and weaker once a learner is practising. It points away from Circin twice: the benefit is measured in example-then-practice sequences Circin does not provide, and it is about studying an example, not scrolling past one. Choosing Lesson in a feed is not studying a worked example in a course, and Circin measures nothing about whether anyone studied anything.

Content styles, in the live app. Circin’s sample cards have no video behind them; a creator’s post plays.

How much it assumes

Skill choice: Introductory to Advanced

Four levels, each with the app's own one-line description shown beside it in the filter sheet: Introductory, a quick entry point with little assumed knowledge; Beginner, builds essential terms and foundations; Intermediate, applies established basics to multi-step ideas; Advanced, covers complex or specialised material. A creator can tag a video with several and mark one as its main level. You can filter a search by level and put levels on a mix. Circin runs no test, holds no estimate of what you know, and changes nothing inside a video based on who is watching.

What you can do with it

You can set the levels you want to see, change them whenever you like, and change them back.
Nothing asks you to rate yourself once and live with it, and nothing corrects you afterwards.

The limit. The level filter does much less than it looks. 108 of the 128 sample cards carry no level at all, and a video with no level passes every level filter — so filtering to Introductory today removes almost nothing from the sample library. A video tagged with several levels resolves to no single level, so a card marked Introductory-to-Advanced never matches a level preference either. In a mix, a level re-orders rather than excludes. And Circin does not adapt, does not meet you where you are, and has no path that fades support as you go: a series groups videos in the order a creator chose, and nothing measures you or adjusts.

The research beside it

Replicated experiments

How much you already know changes what helps. Techniques that reliably help novices — integrating text into a diagram, studying a worked example instead of solving problems — become unnecessary and then actively unhelpful as a learner's knowledge grows, because processing the now-redundant support itself costs working memory. That is the real reason one video cannot be the right video for everyone.

What it found The expertise reversal effect names the finding that instructional techniques highly effective for inexperienced learners lose effectiveness and can have negative consequences for more experienced ones. The effect is explained within a cognitive load framework in which the learner's existing knowledge changes what a given technique does, and it has been replicated across a wide range of instructional materials and participant groups.

Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38(1), 23–31. doi.org/10.1207/S15326985EP3801_4

What it cannot say

Controlled experiments with instructional materials — diagrams, worked examples, short lessons — in which experimenters measured prior knowledge by test and assigned the matching version. Circin measures nothing: a level is a word the creator typed and a filter you set. The effect shows that a mismatch can cost you; it does not show that a self-chosen label prevents one. What the literature actually supports is adaptive, assessment-driven tailoring, and Circin has none of it, by choice.

Meta-analysis

People judge their own ability with moderate accuracy, and more accurately in a narrow domain than in general — a mean correlation of .29 across 22 meta-analyses, stronger when the self-evaluation was domain-specific. That is the honest case for asking what level you want in a particular subject, rather than asking you to rate yourself once.

What it found A metasynthesis of 22 meta-analyses compared self-evaluations of ability (academic ability, intelligence, language competence, medical skills, sports ability, vocational skills) with objective performance measures (standardised tests, grades, supervisor evaluations). The mean correlation across meta-analyses was moderate (M = .29, SD = .11), with individual meta-analytic effects from .09 to .63, and was stronger when self-evaluations were domain-specific and when the performance task was objective, familiar, or low in complexity.

Zell, E., & Krizan, Z. (2014). Do people have insight into their abilities? A metasynthesis. Perspectives on Psychological Science, 9(2), 111–125. doi.org/10.1177/1745691613518075

What it cannot say

This is a caveat printed beside the feature, never an argument that Circin should decide for you. A self-chosen level is an estimate, and a moderate correlation means it will sometimes be off in both directions; Circin will not correct it, because it does not test you. The domains studied are school, work and sport abilities, not choosing videos, and a commentary in the same journal argued the correlation alone understates how wrong self-assessments can be in absolute terms.

Skill levels, in the live app. Circin’s sample cards have no video behind them; a creator’s post plays.

Saying something back

Comment lenses: eight ways in

A handle above the comment box opens eight lenses — Question It, Perspective, Analyze, Apply It, Counterpoint, Evidence, Connect, Build On It. Each carries a one-line description of what it is for and an editable opening phrase: Counterpoint opens "Another way to read this: ", Evidence opens "What backs this up: ". The starter is a suggestion, so it is only swapped or cleared while you have not written past it. The label is stored on the server with the comment and travels back to every reader, and a thread's filter chips are built only from the lenses the people in it actually used. Question It also marks the comment as a question, which is what makes "Ask creators" possible under it. An Evidence comment can cite one of the video's own sources, and the Sources sheet links back to the Evidence comments on that video.

What you can do with it

You can pick a lens and get a box that already has a phrase in it to rewrite or delete, filter a thread down to the kind of thinking you want to read, or write a plain comment and never open the handle at all.

A viewera drawn commentCounterpoint1

2 Another way to read this: … and the rest is whatever the writer wanted to say.

3 Filter chips above the thread: Questions · Creator Replies · Counterpoint

The lens labelStored on the server with the comment, and sent back to every reader.
The starter phraseAlready in the box, editable. It is only swapped or cleared while you have not written past it.
The thread’s filterQuestions and Creator Replies are always offered. A lens chip such as Counterpoint appears only once someone in that thread has used it.
No checkNothing verifies that a lens was used well, and a reply carries no lens at all.

ExampleOne lens, as it travels with a comment. The starter is the app’s own wording for Counterpoint; the comment itself is drawn, not quoted from anyone. A plain comment with no lens is a complete comment.

The limit. A lens is a label, not a check: nothing verifies that an Evidence comment carries evidence, and a public comment has an audience, which can turn explaining into performing. Lenses belong to top-level comments only, so the row disappears while a reply is being written and no label is stored on a reply. Commenting needs an account, comments are text only, and a creator can switch comments off for a post. A plain comment with no lens is a complete comment, and the row says so.

The research beside it

Meta-analysis

A prompt that asks why, or where this would apply, gets a different kind of answer than an empty box does. Prompted self-explanation produced small-to-moderate improvements in procedural knowledge, conceptual knowledge and transfer on immediate measures — and the effect was stronger where the prompt scaffolded a good explanation rather than simply asking for one.

What it found Rittle-Johnson, Loehr and Durkin meta-analysed studies that prompted self-explanation in mathematics learning. Prompting produced small to moderate improvements in procedural knowledge, conceptual knowledge and procedural transfer on immediate measures. Moderator analyses found the immediate effect stronger when scaffolding of high-quality explanation was provided, and unchanged by whether time on task was equated across conditions. Evidence that prompted self-explanation reliably helps within a classroom context, or that it survives a delay, was much more limited.

Rittle-Johnson, B., Loehr, A. M., & Durkin, K. (2017). Promoting self-explanation to improve mathematics learning: A meta-analysis and instructional design principles. ZDM Mathematics Education, 49(4), 599–611. doi.org/10.1007/s11858-017-0834-z

What it cannot say

Mathematics, with a task and a test at the end; a comment has readers and no test at all. Circin's lens is an editable opening phrase and a label — closer to a bare prompt than to the scaffolding that made the difference in those studies — and it is optional, where the studies prompted everyone. Two boundaries land here directly: the benefit that appeared immediately did not reliably survive a delay, and evidence that prompted self-explanation helps inside a real setting was much thinner. Nothing here says a thread retains anything.

Comments, in the live app. Circin’s sample cards have no video behind them; a creator’s post plays.

Your own verdict

The private review: five stars nobody sees

One rating per account per video, one to five stars, changeable, and cleared by tapping the same star again. Nobody is shown what you gave. The number under a video is the server's aggregate — the average and how many people have rated — and your star moves it, barely, alongside everyone else's, which is the honest behaviour. Three stars or more puts the video in a private Rated tab in Collections, with chips for 5, 4 and 3 stars and a highest-first or lowest-first sort; that tab is read straight from your ratings, so it cannot drift out of step with them, and its empty state tells you how many you rated under three. A rating made on a kid profile is never sent anywhere.

What you can do with it

You can rate what you watched and keep the Rated tab as a record of what was worth the time, or rate nothing at all — it changes nothing about what you are shown, because no feed reads your ratings.

DrawnNobody is shown what you gave. Your star moves the average, barely, alongside everyone else’s. A video nobody has rated says so instead of showing 0.0, and if two ratings land in the same instant, the average is recounted from each account's own row.

The limit. A video nobody has rated says so rather than showing 0.0 — sample card and real video alike. Only the ?demo=1 walkthrough fills a sample card's stars with a placeholder derived from its id, and even there it is flagged so nothing downstream can present it as a measurement. Each account's own star is the record; if two ratings land in the same instant and the average misses one, it is recounted from those records. And rating a video adds it to a device-side "liked" list that feeds a Liked rail on your own profile — but not to the Liked collection in Collections, which fills from saving. Two lists carry the same name and do not hold the same things. That is a fault, not a design.

The research beside it

Meta-analysis

A rating after watching is a judgement about the video, and people's judgements of their own comprehension track their actual comprehension only weakly — a weighted mean correlation of 0.178 across 502 effects from 115 studies and 15,889 participants. Which is part of why nobody sees yours and nothing is scored.

What it found Yang, Zhao, Yuan, Luo and Shanks pooled 502 effects from 115 studies and 15,889 participants on metacomprehension — how well readers can tell the texts they have understood from the ones they have not. Without any intervention the weighted mean correlation between judgement and performance was 0.178. Several interventions did improve accuracy, notably delayed summary writing and delayed keyword generation, and combining interventions tended to add up.

Yang, C., Zhao, W., Yuan, B., Luo, L., & Shanks, D. R. (2023). Mind the gap between comprehension and metacomprehension: Meta-analysis of metacomprehension accuracy and intervention effectiveness. Review of Educational Research, 93(2), 143–194. doi.org/10.3102/00346543221094083

What it cannot say

Texts read in studies, not short video, and it concerns a different judgement — how well did I understand this — than the one five stars asks for. Read the way it points, it is a caution about the star rather than a case for it: a rating is not evidence you learned anything. It also names something Circin has not built. What improved accuracy in those studies was delayed work afterwards — writing a summary later, generating keywords later — and Circin asks none of that of you.

The review sheet, in the live app. Circin’s sample cards have no video behind them; a creator’s post plays.

When nobody has made it yet

The Knowledge Demand board

A shared board of questions people actually asked. There are two ways on. The board's own form takes a question, a Subject and subcategory, and an option to ask anonymously. Or, under your own Question comment, "Ask creators" sends the first 300 characters up as a request — the app tells you that before it does. The server decides who a request is aimed at from the video itself and ignores whatever the client sent, and it refuses a second ask of the same comment, somebody else's comment, a reply, and a comment that is not a question. If the video is someone else's, that creator gets a notice; answers land in your Inbox. Upvotes are one per account. A creator can attach a published video as an answer, and the player then carries a backlink — "In response to: …" — to the question it answers. Reading the board needs no account; asking, upvoting and answering do.

What you can do with it

You can put a question you already asked in front of the people who make videos, upvote one somebody else asked so a creator can see how many wanted it, or read the board and answer nothing.

DrawnA question you already asked can travel to the people who make videos. The dotted step is the one nobody owes you: a request can sit on the board with its upvotes and no answer. The board’s own form is the other way on, and the only one that can be anonymous.

The limit. Nobody is obliged to answer — a request simply sits there with its upvotes. An anonymous ask starts at zero upvotes and its author cannot upvote it, because the row that records one vote per account would name them. Outside the walkthrough the board holds only what was really asked, which today is close to empty — and if questions are rare even with a tutor's whole attention, a board will be quiet. An upvote count is the number of accounts that upvoted, counted from each one's own row, so two in the same instant both count.

The research beside it

Single study

In tutoring, the quality of the questions students asked tracked achievement; the number of questions did not. Asking is also rarer than it feels — in the same research programme, even one-to-one tutoring produced only about six genuine information-seeking questions an hour from a typical student, with tutors asking 80% of the questions.

What it found Graesser and Person analysed question asking in tutoring sessions — 27 college students on research methods and 13 seventh-graders on algebra. Student achievement was positively correlated with the quality of student questions once students had some experience and training in tutoring, while the frequency of questions was not correlated with achievement. Asking is also rarer than it feels: in the same research programme, even one-to-one tutoring produced only about six genuine information-seeking questions an hour from a typical student, with tutors asking 80% of the questions and introducing 93% of the topics (Graesser, Wiemer-Hastings, Wiemer-Hastings & Kreuz, 1999, Journal of Cognitive Systems Research 1, 35–51).

Graesser, A. C., & Person, N. K. (1994). Question asking during tutoring. American Educational Research Journal, 31(1), 104–137. doi.org/10.3102/00028312031001104

What it cannot say

Two small tutoring corpora, one subject each, correlations rather than causes, and questions spoken to a person rather than posted to a board. It points at what the board cannot do: an upvote counts how many people wanted an answer, which is nearer frequency than quality, and nothing on the board rates a question or teaches anyone to ask a better one. The board routes demand to creators. It does not train anybody.

The Knowledge Demand board, in the live app. Circin’s sample cards have no video behind them; a creator’s post plays.

Where it came from

Sources, attached to the moment they back

A creator credits sources on a video — a title, a kind, a year, a note, an optional link and an optional moment. The Sources sheet lists them under two groups that are the creator's own, "What this is based on" and "Where to go next", with anything ungrouped under "Also referenced"; the headings appear only if the creator used them. A source with a moment carries a timestamp button that jumps the video there, and unlike the notes sheet this works on a real creator's video as well as on a sample card. A row can say "Also cited by N videos on Circin" and open everything citing the same title. The footer credits the creator by name for the titles, notes, groupings and moments, and says plainly that Circin does not verify them. A link leaves through a screen that names the host with www. stripped, because that is the one fact Circin can state about a link without fetching it — and in Kids mode that screen needs the parent PIN. An Evidence comment can cite one of the video's own sources, and the sheet links back to those comments.

What you can do with it

You can open the Sources sheet from a video, tap a moment to hear the part a source is said to back, follow a link out to the book, the paper or the course, or report a source that is wrong.
Or watch and never open it, which is what most people do.

DrawnA source can point at the second it is said to back. Nothing checks it: the moment is a pointer the creator typed, and Circin cannot tell what backs what. Whether a video carries a source at all is the creator’s decision.

The limit. Nothing checks a source. The moment is a pointer the creator typed — Circin cannot detect what backs what. The server silently drops a link that is not http(s), a moment landing past the video's own length, a source with no title, and anything past the twelfth, and tells the creator none of that; only "basis" and "more" are accepted as groups, so no third label can reach a viewer. Whether there is a source at all is the creator's decision: Circin does not require one, and three of the 128 sample cards carry any.

The research beside it

Single study

It gives you something to check. Even where citations are one click away at enormous scale, they are clicked rarely — one page view in roughly 300 on English Wikipedia (0.29% overall, 0.56% on desktop, 0.13% on mobile) — and clicks were more frequent on shorter and lower-quality pages, suggesting people reach for the source when the page itself has not given them what they came for.

What it found Piccardi, Redi, Colavizza and West instrumented English Wikipedia for a month and logged every click from an article to a cited reference. About one in 300 page views produced a reference click: 0.29% overall, 0.56% on desktop and 0.13% on mobile. Matched observational analyses found clicks more frequent on shorter and lower-quality pages, suggesting people reach for the source when the page itself has not given them what they came for.

Piccardi, T., Redi, M., Colavizza, G., & West, R. (2020). Quantifying engagement with citations on Wikipedia. Proceedings of The Web Conference 2020, 2365–2376. doi.org/10.1145/3366423.3380300

What it cannot say

One platform, one month, an encyclopaedia rather than video, and a count of clicks — not of anyone's judgement, belief or learning. It points away from any claim that attaching sources makes people check: the mobile rate was the lowest of all, and Circin is a phone app. What it supports is smaller and still worth building — the source is there for the reader who does want to go further, and the finding that clicking rises where the page fell short is exactly why Circin's sources point outward, to where the subject actually lives.

The Sources sheet, in the live app. Circin’s sample cards have no video behind them; a creator’s post plays.

Your own time

Time on Circin, counted and then left alone

One clock, ticking a second at a time, only while something is really playing: not in a background tab, not behind a profile or Settings, not paused, not on a card autoplay never started — and yes in the lesson view. A tick is worth exactly one second and never a wall-clock gap, so a timer the phone suspended cannot credit the time it slept through; the count runs slightly low and never high. A day is a small record — the total, which feed it came from, the subject, the creator and the video. Your Activity draws seven real days, and beneath the bars, once at least five minutes carry a breakdown, it names where the time went, up to five subjects, three creators and three videos. A Clear in the card header removes the lot. Which videos and which creators is watch history by another name, so it follows the watch-history switch you already have rather than a second one to find.

What you can do with it

You can look at it, or not.
There is no goal, no streak, no reminder, no weekly digest and nothing to beat, and clearing it costs nothing.

The limit. It never leaves the device and sits outside the account backup by construction, so it does not follow you to another phone, and eight days is all that is kept. Under five minutes with a breakdown behind them the detail stays hidden rather than showing a 100%-Science bar built out of forty seconds. A kid's second lands in the day total and in its own bucket but carries no subject, creator or video, so a parent's card adds up without becoming a log of what their child watched. And this is the one place the page has to correct itself: Circin does count the minutes that were really playing. What is true is narrower — that record stays on the device, feeds nothing, and never reaches the ordering of any feed.

The research beside it

Meta-analysis

It can show you measured minutes instead of an impression. Across 106 comparisons of self-reported against logged digital media use, self-reports correlated only moderately with the logs and were rarely an accurate reflection of them — so a counted number tells you something you did not already know.

What it found A pre-registered systematic review and meta-analysis pooling dozens of studies found that self-reported digital media use correlated only moderately with logged use, and that self-reports were rarely an accurate reflection of the logged figure, in either direction.

Parry, D. A., Davidson, B. I., Sewall, C. J. R., Fisher, J. T., Mieczkowski, H., & Quintana, D. S. (2021). A systematic review and meta-analysis of discrepancies between logged and self-reported digital media use. Nature Human Behaviour, 5(11), 1535–1547. doi.org/10.1038/s41562-021-01117-5

What it cannot say

That is about measurement accuracy only. It gives no reason at all to believe that seeing an accurate number changes what anybody does.

Meta-analysis

Monitoring progress does move behaviour in the studies — d+ = 0.40 across 138 randomised studies and 19,951 people — but only where there was a goal to monitor against, and more where the record was physically written down or reported publicly. Which is precisely what Circin leaves out.

What it found Harkin, Webb, Chang, Prestwich, Conner, Kellar, Benn and Sheeran meta-analysed 138 randomised studies (N = 19,951) that assigned people to an intervention promoting monitoring of goal progress or to a control. Monitoring promoted goal attainment, d+ = 0.40 (95% CI 0.32–0.48); the effect was mediated by how often progress was actually monitored, and was larger when progress was physically recorded or reported publicly.

Harkin, B., Webb, T. L., Chang, B. P. I., Prestwich, A., Conner, M., Kellar, I., Benn, Y., & Sheeran, P. (2016). Does monitoring goal progress promote goal attainment? A meta-analysis of the experimental evidence. Psychological Bulletin, 142(2), 198–229. doi.org/10.1037/bul0000025

What it cannot say

Health and behaviour-change goals, not screen time, and every study had a goal — the ingredient this design deliberately does not supply. Read honestly, it is evidence against a claim Circin could have made: a number with no goal, no streak and nothing reported to anyone is the weakest version of what those studies tested. The card exists because the minutes are yours and you are entitled to see them counted properly, not because seeing them is known to change anything. Leaving out goals and streaks is a value Circin holds, and the meta-analytic evidence points the other way.

A diagram of the rule, not a chart of anyone’s week: the bars are illustrative. Circin keeps this on the device and sends none of it anywhere.

Handing the phone over

Circin Kids, and a kid profile

Kids mode is watch-only. A kid profile carries a first name, a @username and an optional parent email; it lives in the device's own account list, and the server refuses a kid profile outright, so nothing about a child syncs anywhere. The parent email stays on that phone unless a parent asks for a PIN reset code, which is the one thing that sends it to Circin. Comments, creating and the options menus are switched off. A kid profile keeps rating and saving and both stay on the device — no rating a child makes is pushed to the server, and a kid profile never reaches the views route, so a child's watching appears in no creator's numbers. Leaving needs a parent PIN, with a recovery code. A parent can set an optional daily limit — 15, 30, 45, 60 or 90 minutes, or none — counted by the same clock as everything else, with a screen at the end that a parent can extend by fifteen minutes with the PIN. It is the only clock on Circin that stops anything. And a kid profile can be grown up into a full account at 13, keeping its saves and collections.

What you can do with it

A parent can make a profile, choose the Subjects it sees from the Kids tab, set a limit if they want one, and hand the phone over.
No ads reach it.
Setting no limit is a complete way to use it.

DrawnNothing about a child syncs anywhere, and leaving takes the parent PIN. One exception, and a parent starts it: asking for a PIN reset code sends the parent email to Circin. A PIN is a lock, not mediation, and the daily limit is kept per device, so two children on one phone share it.

The limit. A PIN is a lock, not mediation. Circin Kids has no watch-together mode and no conversation prompts for a parent, and removing comments and messages removes a channel — which is a much narrower claim than making a child safe. Nothing here is shown to improve any outcome. The library is a keyword rule over the sample cards, and no creator's video can be approved for Kids yet, because the owner account is not configured on the server. The daily limit and the minutes used are stored per device and not per profile, so two children on one phone share one budget. Nothing syncs and there is no server-side backup or recovery of a kid profile: a lost device is a lost profile. And a source link opens outside Circin only with the parent PIN.

The research beside it

Theory

"Educational" is not a tested claim in an app store. A review of the science of learning for children's apps reported that the roughly 80,000 apps then listed as educational in Apple's App Store were largely unregulated and untested, because there is no time, money or resource to evaluate each one as it enters the market — and set out four pillars for judging whether an app deserves the word: active, engaged, meaningful and socially interactive learning, inside a supported learning goal.

What it found Hirsh-Pasek, Zosh, Golinkoff, Gray, Robb and Kaufman reviewed the science of learning for the design of children's apps, reporting that the roughly 80,000 apps then listed as 'educational' in Apple's App Store were largely unregulated and untested, because there is no time, money or resource to evaluate each app as it enters the market. They set out four pillars — active, engaged, meaningful and socially interactive learning, within a supported learning goal — as the basis for judging whether an app deserves the word.

Hirsh-Pasek, K., Zosh, J. M., Golinkoff, R. M., Gray, J. H., Robb, M. B., & Kaufman, J. (2015). Putting education in "educational" apps: Lessons from the science of learning. Psychological Science in the Public Interest, 16(1), 3–34. doi.org/10.1177/1529100615569721

What it cannot say

A framework synthesised from earlier research, not a trial of any app, and written mainly about younger children than Circin Kids' viewers. It points squarely away from Circin on three of its four pillars: watching is the passive end of "active", social interaction is switched off here by design, and today's Kids library is selected by a subject-and-style keyword rule — 64 of the 128 sample cards pass it — rather than by a person who has watched them. The app's own control plane says it outright: a sample is a rule, not a review. Circin Kids is a safer room to watch in. It is not an educational product and is not claimed as one.

Single study

The largest population test of specific screen-time caps for young children did not support them once family circumstances were taken into account — 19,957 caregiver interviews about 2- to 5-year-olds, with wellbeing measured as attachment, resilience, curiosity and positive affect. Which is why Circin makes a limit available and refuses to recommend a number.

What it found Analysis of 19,957 telephone interviews with caregivers of 2- to 5-year-olds assessed children's digital screen use against psychological wellbeing measured as caregiver attachment, resilience, curiosity and positive affect in the past month. Evidence did not support implementing the limits then recommended by the American Academy of Pediatrics (under 1 hour or under 2 hours per day) once child ethnicity, age, gender, household income and caregiver educational attainment were taken into account.

Przybylski, A. K., & Weinstein, N. (2019). Digital screen time limits and young children's psychological well-being: Evidence from a population-based study. Child Development, 90(1), e56–e65. doi.org/10.1111/cdev.13007

What it cannot say

Cross-sectional, caregiver-reported on both sides, children aged 2 to 5, one country — and a null result is not evidence that limits are useless, only that this design found no support for those thresholds. The finding is contested: paediatric bodies have continued to issue duration guidance and the same authors' work has drawn methodological rebuttals. It licenses Circin saying there is no number it can honestly hand you. It never licenses "limits do not matter", and it says nothing whatsoever about a nine-year-old, or about Circin.

Setting up a kid profile, in the live app. Circin’s sample cards have no video behind them; a creator’s post plays.

The parent PIN, in the live app. Circin’s sample cards have no video behind them; a creator’s post plays.

Refusals

What Circin will not claim

Each of these would make this page more persuasive. None of them is supportable, so they are written down for you to hold Circin to their absence.

"Short video suits today's eight-second attention span."

No attention-span number belongs on this page, including a flattering one. The eight-second figure has no traceable scientific source, and the 10–15 minute lecture limit it descends from does not survive a review of the primary data. Bradbury (2016), Advances in Physiology Education 40(4), 509–513.

"Sixty seconds is right because working memory holds about four things."

A limit on what can be held in mind at once says nothing about how long a video should be, and the conversion from chunks to seconds is invented. The number is also still argued over, from three to seven. Cowan (2001), Behavioral and Brain Sciences 24(1), 87–114.

"Short-form teaches faster, or better, than long-form."

Nobody has compared them on learning outcomes. The largest study of educational video watching measured how much of a video people watched, which is attention, not learning. Guo, Kim & Rubin (2014), Learning @ Scale, 41–50.

"Microlearning is a proven method."

The field has no agreed definition, duration claims run from one to fifteen minutes, and a review of forty studies described the results as scattered and mixed. The word settles nothing, so Circin does not lean on it. Monib, Qazi & Apong (2024), Heliyon 11(2), e41413.

"Circin's feed makes you smarter."

No recommender system has been shown to improve learning outcomes, and nothing in this evidence base speaks to Circin's ordering — which is a handful of rules you set yourself, with no model behind it and nothing learning from you.

"Circin is engineered to be addictive." / "A dopamine hit for learning."

Circin works with how people already use the internet and tries to give those minutes somewhere better to go. That is the whole of the claim. Addiction is a contested label even for phone use itself. Panova & Carbonell (2018), Journal of Behavioral Addictions 7(2), 252–259.

"Short-form video is destroying attention spans."

The mirror-image scare is unsupported too, and self-reported media use corresponds only loosely with what logs show. Circin neither fearmongers about your phone nor absolves itself. Parry et al. (2021), Nature Human Behaviour 5(11), 1535–1547.

"Watching explanations builds skill."

Across six experiments with 2,225 people, repeatedly watching others perform raised their belief that they could do it while measured performance did not move. You can watch, and only watch — that is a complete way to use Circin — and it is not practice. Kardas & O'Brien (2018), Psychological Science 29(4).

"If it felt clear, you understood it."

A clear sixty seconds can feel like understanding. The sense of fluency is a poor guide to what you will still know next week, which is the reason a satisfying video — or a five-star rating — is not treated here as evidence that learning happened. Bjork, Dunlosky & Kornell (2013), Annual Review of Psychology 64, 417–444.

"The two best-supported study techniques are built into Circin."

Retrieving and spacing are the best-supported techniques, and they are the shapes Circin borrowed — explaining in your own words, and coming back to something later. Borrowing a shape is not delivering the thing: a comment is not a test, a saved collection is not a review schedule, and Circin measures neither. Dunlosky et al. (2013), Psychological Science in the Public Interest 14(1), 4–58.

"Showing sources makes viewers harder to fool."

Credibility information shown beside content had negligible average effects on what people read and did not reduce misperceptions. Sources are here because you are owed them, not because a panel fixes anything. Aslett et al. (2022), Science Advances 8(18), eabl3844.

"Circin teaches media literacy."

Circin has no lessons and nothing in the app is built to teach. Where judging sources was actually taught, it took six 50-minute lessons with practice and feedback inside a school course. Wineburg et al. (2022), Journal of Educational Psychology 114(5), 893–909.

"Circin adapts to how you learn best."

There is no adequate evidence for matching teaching to a diagnosed learning style, and Circin diagnoses nothing. You pick subjects, styles and levels; that is a preference you set, not a profile of you. Pashler, McDaniel, Rohrer & Bjork (2008), Psychological Science in the Public Interest 9(3), 105–119.

"You can learn a subject in sixty seconds." / "Circin replaces school, teachers or books."

You cannot, and it does not. Video's measured benefit is strongest when it is added to teaching and much weaker when it stands in for it, which is why the sources under a video point at the longer thing. Noetel et al. (2021), Review of Educational Research 91(2), 204–236.

"Circin is proven to work."

No study has evaluated Circin. Every mechanism on this page is evidence about a mechanism, in other settings, with other materials. If that ever changes, the study will be named right here with its own caveat.

Come curious. Leave knowing something you didn't before.

That is the whole promise, and it is deliberately small. Circin is a doorway: the book, the course and the person who teaches it are still where a subject actually lives, and the sources under a video are there to send you to them. Watch one thing and close the app, and you have used Circin properly.

Try Circin See how it works

Circin is new, so there is not much in the feed yet — it fills as creators publish.

Free · made for your phone · no account needed to watch

If a sentence on this page outruns the code or the research behind it, that is a fault in the page. Tell us and it gets corrected here, the way a correction travels with a video — added, not quietly swapped.