The problem InputScout is trying to solve
YouTube has more target-language video than any learner could watch, but a large library does not answer a learner’s immediate question: What can I watch now that interests me and is close enough to my level to follow?
InputScout is a discovery and calibration layer around public YouTube videos. It does not copy the video library, download media, import your YouTube history, or rank with YouTube views, likes, comments, or subscriber counts. Videos play through the official YouTube player or open on YouTube.
1. Start with a language profile
You choose a language and a self-described CEFR starting point from A1 through C2. InputScout maps those choices to starting values on its internal 0–100 scale:
| Starting point | Initial estimate |
|---|---|
| A1 | 8 |
| A2 | 25 |
| B1 | 42 |
| B2 | 59 |
| C1 | 76 |
| C2 | 92 |
These values initialize discovery. They are not test results, certifications, or claims that every skill sits at the same CEFR level. The score exists so the app can choose an initial range and update it consistently as you provide feedback.
You can also select topic, dialect, and duration preferences. Those preferences influence recommendations without changing your ability estimate or a video’s global difficulty.
2. Keep three kinds of evidence separate
InputScout deliberately avoids collapsing every reaction into one popularity number.
Personal fit
After a meaningful watch, you can say that a video felt easy, just right, or too hard. That observation helps adjust your estimate for the active language. It does not directly rewrite the video’s global score.
A watch qualifies when the official player reaches its natural end or when you have watched at least 20% of the duration, bounded between one and five minutes. This reduces drive-by ratings while still allowing useful feedback on longer videos.
Global relative difficulty
Occasionally, InputScout asks which of two meaningfully watched videos was harder, or whether they felt equal. Those comparisons update the videos’ relative positions within the same language. Comparisons are more useful than asking every learner to invent an absolute number.
Difficulty remains language-relative. A score of 60 in Spanish cannot be compared with 60 in French. It also does not mean “60% understood.”
Helpfulness for comprehensible input
An optional yes/no response can indicate whether a video was useful for comprehensible input. This signal contributes to recommendation quality, but it never changes your ability or the video’s global difficulty.
3. Update your estimate cautiously
For a fit rating, InputScout creates an observation around the video’s difficulty: above it for “easy,” at it for “just right,” and below it for “too hard.” The learner estimate moves only part of the way toward that observation.
The movement is smaller when the video itself has low difficulty confidence. If you change an earlier fit rating, the current evidence is recomputed from your self-assessment baseline so the old and new labels are not both applied.
This is an experimental calibration model. It is designed to respond gradually, not to infer a complete language ability from a few clicks.
4. Calibrate video difficulty through comparisons
Pairwise comparisons use an Elo-like update. A comparison that matches the current order makes a smaller adjustment; a surprising result makes a larger one. Newer items can move more quickly, while the update size decreases as comparison evidence accumulates.
InputScout also shows confidence separately from score:
- New: confidence below 0.25.
- Calibrating: confidence from 0.25 through 0.69.
- Established: confidence at 0.70 or above.
Low confidence does not mean a video is bad. It means the community has provided less evidence about its placement. A small exploration bonus prevents new items from disappearing simply because they are new.
5. Rank recommendations
The default target is slightly above the current learner estimate. The documented core recommendation score combines:
| Signal | Core weight |
|---|---|
| Difficulty fit | 55% |
| Topic preference | 20% |
| Dialect preference | 15% |
| InputScout helpfulness votes | 10% |
A small deterministic exploration bonus is added for low-confidence items. You can override recommendations with filters for difficulty, dialect, topic, duration, learner-directed content, known subtitles, saved or completed state, and title, channel, or app-owned tags.
Interest is intentionally part of the model. A theoretically perfect difficulty match is not useful if you do not want to watch it.
6. Recenter without overriding you
The default near-level filter follows your estimate. When fit feedback changes that estimate, an unpinned range recenters automatically. If you manually choose and pin a difficulty range, InputScout preserves your choice.
This distinction lets the app adapt by default without treating its recommendation as more authoritative than your own decision.
7. Track activity without pretending it is YouTube history
InputScout can track progress from playback inside the app and from time you manually report. It does not import or record official YouTube watch history. Progress minutes, saved items, completion state, ratings, and comparisons are app-owned records.
The app can show a weekly input goal and recent activity, but it does not use fake streak pressure. Minutes are a planning aid, not proof that acquisition occurred.
8. Preserve the source and the creator
InputScout retrieves official metadata through approved YouTube APIs and uses the official IFrame player. YouTube retains its controls, links, advertising, creator attribution, and platform behavior. Videos can become unavailable or change independently of InputScout.
InputScout difficulty, confidence, fit, helpfulness, and recommendation signals are not YouTube data and are not supplied, reviewed, or endorsed by YouTube.
What the system can and cannot do
InputScout can narrow a catalogue, learn from feedback, show uncertainty, and make its ranking factors legible. It cannot measure every dimension of proficiency, guarantee that a recommendation will feel right, know what you watched outside the app, or promise a learning result.
Treat the score as a compass. Your comprehension, attention, interest, and willingness to continue are the terrain.