How the Heddy Score works
The Heddy Score rates a learning video with seven stars. Each star is one test the video must pass, and the score is the number it has passed, from 0 to 7. The same rules apply to every source: YouTube creators, Oak, BBC Bitesize, and videos we make ourselves.
Version 2.0 · 2026-10-16 · an open standard, intended for CC BY 4.0The seven stars
A star is pending until someone or something has judged it. Pending never counts as a pass, so a new video starts low and earns its stars. Every star applies to every video, so the score is always out of 7.
Trusted sourcerequired
Is it the creator's own, live, embeddable upload?
Pass: Official channel; oEmbed returns 200 with the expected author; plays in the embed; not age-restricted.
Fail: A re-upload, private, removed, not embeddable or age-restricted.
Judged by: Automatic (oEmbed and YouTube metadata, weekly).
On the maprequired
Does it mainly teach the map topic it is linked to?
Pass: A reviewer watched it and confirms the main topic (and timed any second topic as its own chapter). Subject matches.
Fail: The topic is a lead-in, a passing mention, or a different subject.
Judged by: Reviewer (watches it).
Spec-truerequired
Is it accurate and right for the current UK specification?
Pass: No factual errors; matches current KS3 or GCSE content; the curator's key points agree with it; no misconception taught.
Fail: Errors, an old specification, or content that contradicts the key points.
Judged by: Reviewer, with an LLM pre-check of the key points and an automatic flag on uploads before the specification was first taught. At review time the reviewer also reads the top and newest YouTube comments for reported errors; nothing is stored or quoted.
Focused
Is it one concept, short enough to be a clip?
Pass: About 3 to 10 minutes for one idea, or longer with creator chapters we can clip to.
Fail: A long revision session or past-paper walk-through with no usable chapters.
Judged by: Automatic (duration and chapters).
Accessible
Can every student follow it?
Pass: Human-made English captions, or a reviewer confirms the auto-captions get the subject terms right; clear audio and visuals.
Fail: No usable captions, or captions that garble key terms.
Judged by: Automatic for human captions, otherwise reviewer.
Safe for childrenrequired
Is it right for 11 to 16 year olds?
Pass: No age restriction; no adult content; no pressure to follow off-platform links, sponsors or paid upsells inside the teaching.
Fail: Age-restricted, unsuitable content, or a sales pitch aimed at children.
Judged by: Automatic pre-check, then reviewer. At review time the reviewer also reads the top and newest YouTube comments for safety concerns; nothing is stored or quoted.
Proven
Do students who watch it learn from it?
Pass: After at least 30 video checks and 30 answers to our own 'Did this clip help?' tap: the 3/3 rate is at or above the topic's average, at least 70% found it helpful, and at least half of those who started it watched to the end.
Fail: After those minimums, any of the three is below its line.
Judged by: Learners (our own video checks and clip feedback only, computed by scripts/videos/rate-proven.ts). YouTube numbers are never an input.
Bands
- 7/7 · Heddy's pick
- 6/7 · Excellent
- 5/7 · Strong
- 4/7 · Meets the standard (the minimum to publish)
- 0 to 3 · Below the standard: not shown to students
Proven: earned from our own students
Star 7 stays pending until a clip has at least 30 video checks and 30 answers to our own one-tap "Did this clip help?". Then it passes only when all three of these hold, and fails if any does not:
- Learning: the share of video checks scoring 3 out of 3 is at or above the average for the topic.
- Helpful: at least 70% of the answers are "Yes".
- Watched: at least 50% of the students who started the clip in our player watched it to the end.
So a new clip can reach at most 6/7 until learners prove it. These are first-party signals only: our students' checks, answers, starts and finishes.
The Heddy Score card
Every clip has one card. It shows the seven stars and the score; our students (how many, the share who found it helpful, watched to the end, and got 3 out of 3, once at least 30 students have used it); and, when available, the clip's numbers on YouTube.
- Only the stars and our own students make the score. YouTube's views, likes and subscriber counts are never part of the score, and never an input to any star.
- YouTube's numbers are shown exactly as YouTube reports them, with the date, through the official YouTube Data API. They are cached for at most 24 hours and never stored in our database or files; YouTube's API Developer Policies (III.E.4) allow at most 30 days.
- No metric is derived from them: no ratios, no ranks, no blending.
Rules
- The publish gate. A video is shown to students only when Trusted source, On the map, Spec-true and Safe for children all pass. Those four make the 4/7 floor.
- Stars can be lost. A removed upload fails Trusted source at the weekly check, and falling results can fail Proven.
- Every score is dated. Each star records who or what judged it and when. A reviewer's name stays internal.
- A flag is not a verdict. An automatic flag, such as an upload made before the specification was first taught (GCSE 9–1: Maths and English 2015-09-01; the sciences, History and Geography 2016-09-01), asks the reviewer to look harder. It never fails a star by itself, except on the owner's October 2026 import list, where it fails Spec-true until a reviewer checks the video against the current specification.
- Comments. At review time the reviewer reads a video's top and newest YouTube comments for reported errors and safety concerns. Nothing from them is stored or quoted.
Fair by design
- Reviewers are qualified teachers. An LLM may pre-check, but it never awards On the map, Spec-true or Safe for children on its own.
- Creators can ask for a re-review, decided by a different reviewer.
- No paid placement: a score cannot be bought, and videos we make are judged the same way.
- Before a pass or fail line changes, two reviewers hand-rate 20 videos and any automatic check must agree on at least 90% of verdicts.