A series bible for short-form
A one-page short-form bible locks cold open, slot, length band and numbering so episodes stay comparable, and shows what breaks when each field drifts.
A series is only measurable if episode four is the same object as episode one with a different payload. Most "series" fail that test by week three: the cold open becomes a logo sting, Tuesday 9am becomes "when it is ready," 30 seconds becomes 47, and Part 6 ships as a one-off with a new caption style. The numbers are then incomparable. You cannot kill the run, milk it, or improve it, because you are no longer looking at one thing.
A short-form bible is the one-page spec that prevents the drift. It is not a brand book and it is not a client creative brief. Those documents carry locks and rejected directions. This page carries the four fields that make episodes comparable: a fixed cold open, a slot, a length band, and a numbering scheme. Everything else on the page exists to keep those four still.
Why a bible is a measurement instrument
You read Viewed vs. Swiped Away, completion rate and sends per reach against a median. A median over mixed objects is a fiction. Completion rate is length-dependent by construction: vendor analyses of TikTok floors move as runtime moves (roughly 50% average watch under 30 seconds, 40% from 30–60, 30% past a minute). If episode one is 28 seconds and episode seven is 71, a drop in completion is a change of denominator, not a verdict on the format.
The same contamination hits every other field. Slot drift makes a 24-hour checkpoint compare different audience pools. Cold-open drift compares different first frames. Dropped numbering makes the backlog unsearchable and kills the "Part N" habit that turns a feed item into a run.
YouTube's Shows feature (seasons, custom artwork, bingeable units) assumes comparable episodes. A bible is what makes that grouping honest rather than decorative.
The one-page spec
One page. If it does not fit, you are writing a handbook and nobody will open it next Tuesday. Fill this table once, then stop.
| Field | What you write | Example |
|---|---|---|
| Cold open | The first frame, in film vocabulary, plus the first two seconds of audio | Object on table, labels already on screen, no presenter face, room tone only |
| Slot | Weekday and local time, with timezone | Tuesday 09:00 Europe/London |
| Length band | Inclusive seconds, not a target | 28–34s |
| Numbering | Title pattern and where the number lives | {Series}: Part {n} - {payload}; number on the end card and in the title, never only in the description |
| Caption preset | One named style | Heavy sans, two lines, claim on line one |
| Cast / assets | What is reused, not redescribed | Lead reference sheet; table; label colour |
| End beat | The unresolved question every episode withholds | Verdict card holds the score until the last 8 frames |
| Metric | One number, 24-hour checkpoint | Shorts: Viewed vs. Swiped Away |
Four of those rows are load-bearing. The rest exist so the load-bearing four cannot be "interpreted."
Cold open. The hook is the first frame, not the first line of dialogue. TikTok's auction-ad creative tips put the key message or product in the first three seconds for over 63% of highest-CTR videos. That is a line about ads, not a law. Write the frame. "Energetic start" is not a spec.
Slot. Same weekday, same hour. Early velocity is slot-dependent. A metric read at a fixed 24-hour checkpoint is not comparable across a Tuesday morning and a Saturday night. If you cannot hold the slot, you do not have a series. You have a queue. Cron for a recurring series is how the slot stays a slot once a human is no longer remembering it.
Length band. A band, not a point. 28–34 seconds is a band. "About 30" is how 41-second episodes sneak in. Pick the band from the format, then cut to it. Duration is a production constraint here, not a creative preference.
Numbering. Numbered parts beat vague continuity because they make the backlog navigable. "Part 7" tells a new viewer there is a 1–6. "Another one from us" does not. Put the number in the title. Description-only numbering is how platforms and humans both lose the thread. The widely repeated claim that finishing Part 1 causes Part 2 to be served without search is unverified folklore; what you can measure is returning viewers on episode N+1, and you can only measure that if N is in the title.
The other rows prevent silent mutations. A caption preset stops weekly typography. A cast row is the character-consistency contract: reference sheets get reused, not rewritten. An end beat that withholds something is the reason to watch the next one; if every episode resolves cleanly at 0:29, you have a feed of one-offs.
If you need a finished skeleton rather than a blank table, start from a named series format and copy its runtime, caption and first-frame rules onto this page. Those decisions have to live next to the episode you are making today.
What breaks when each field drifts
This is the diagnostic, not the style guide. When the numbers move, check the bible before you recut.
| Drift | What the sheet starts to show | Why the number is now lying | Repair |
|---|---|---|---|
| Cold open redesigned | Hook rate or Viewed vs. Swiped Away jumps around with no trend | You are A/B testing first frames without a protocol | Restore the specified first frame for three episodes, then test a new open as a labelled variant |
| Slot slips | 24-hour metric looks like a cliff or a spike | Different hour, different pool | Put it back. Re-measure three episodes before you touch the cut |
| Length walks out of band | Completion rate falls (or rises) in a step | Denominator changed | Cut back into band. Do not "explain more" |
| Numbering dropped | Returning viewers on N+1 unreadable; backlog looks like one-offs | The run is no longer a run in the UI | Restore the title pattern. Do not renumber history |
| Caption / layout churn | First-frame recognition never arrives | Regulars cannot identify the show | Lock one preset. Wait four episodes |
| Cast rewritten each time | Comments about "is this the same person" | Identity is not persistent across generations | Reuse the reference sheet; do not re-prompt the face |
Two rows get misdiagnosed constantly. Length drift looks like a retention problem and is often padding: a 30-second idea stretched for a more "complete" answer. Publish at true length. If the idea needs 50 seconds, that is a different bible.
Cold-open drift looks like "the format is tiring." Often you replaced a visual demonstration with a presenter intro and called it a refresh. Put the original first frame back for three episodes. If the median recovers, it was the open.
How to keep it next to production
A bible in a deck dies at the first generation. The page has to sit where the episode is built.
- Fill the table before episode one. If a field is still "TBD," you are not ready to publish a Part 1. You are ready to do a one-off.
- Save the production shape as a reusable workflow. Saving and reusing a video workflow is how the cold open, caption preset and asset map survive a new payload. The bible names the locks; the workflow enforces them.
- Check the cut against the band in the editor. An EDL timeline lets you change duration without rebuilding the file. The 480p preview pass is free of a credit charge and carries a short per-user cooldown, so use it to confirm the episode lands inside the band and the first frame matches the spec. The final export is charged once, regardless of clip count.
- Refuse out-of-spec publishes. If the cut is 41 seconds against a 28–34 band, it does not ship as Part N. It ships as a one-off or it gets cut. Treating exceptions as episodes is how the median dies.
Continuity across separately generated shots is its own craft, and it does not replace the bible. A perfectly continuous 70-second episode is still a broken data point if the band was 30.
Revisit the page on a schedule, not when you are bored. Once per pilot (after the committed N) is enough. Changing the cold open mid-pilot resets the learning block. If you must change it, start a new numbering prefix so the sheet can tell the two objects apart.
FAQ
How is this different from a brand guide or a creative brief?
A brand guide governs look. A brief governs constraints, rejected directions and who approved what. A short-form bible governs comparability across episodes of one run: first frame, slot, length, number. You can have all three. You cannot substitute one for the others. A beautiful brand guide with a wandering runtime still wrecks completion rate.
What if one episode genuinely needs to be longer?
Then it is not an episode of this series. Publish it without the Part N title, or write a second bible with a longer band and a different numbering prefix. Mixing 30-second and 70-second files under one title guarantees a median that means nothing. The end beat can tease the longer film. The series episode stays in band.
Do I need a bible for a three-episode test?
Yes, a thinner one: cold open, slot, length band, metric. Numbering can be "Pilot A, episodes 1–3." The point of a three-episode transplant test is that the skeleton stayed still. Without the spec, you will "improve" the second episode and unblind the test.
Who is allowed to edit the page?
One owner. Suggested edits go in a note, not in the table. The failure mode is four people each "tightening" a different field between episodes. If the owner is away, the series pauses.