Weeks 1–4
1 · Find the click
Publish four broad narratives. Test title and thumbnail before production. Borrow proven packaging structures.
Internal content studioPackage before production
Supabase metadata · R2 assets · measured outcomes
Content Studio
Topic → package → structure → production
Cold-start system
Twelve episodes are an experiment, not a content calendar.
90 days · 12 long videos
Weeks 5–8
2 · Earn the return
Turn the strongest topic into a series; add real builds and a consistent host/visual language.
Weeks 9–12
3 · Prove transaction
Add one relevant lead magnet and one paid offer. Measure email and product intent, not vanity CTA clicks.
Packaging
3 title + 2 cover variants before script
CTR
Judge after meaningful impressions; target top-third of own videos
Retention
Title promise paid off in first 30 seconds
Business
Email conversion and qualified product clicks
Turn heat into evergreen
The event earns timing. The enduring question earns lifetime views.
Model release
What is new in GPT 5.6?
Why better models still fail to create better products
Layoff / hiring news
Which company cut jobs?
How AI rewrites apprenticeship and career leverage
Sleep device launch
New wearable review
What wearables can actually measure and change
Episode equation: current event × enduring human tension × original access × visual mechanism.
Production budget
Human time target for a 10-minute episode.
180m
Package
20 min
3 titles, 2 covers, one promise
Research
35 min
Claim ledger, primary sources, counterpoint
Script
35 min
Narration, A-roll beats, visual plan
Record
25 min
Selective A-roll and clean voice
Rough cut
20 min
Script-aware A-roll spine
Visuals
25 min
B-roll, charts, reusable Remotion scenes
Fine cut
15 min
Rhythm, music, SFX, captions
Publish
5 min
Upload, chapters, CTA, measurement
48-hour launch loop
T−48h
Test 3 titles and 2 covers with five target viewers; kill ambiguity before editing
T0
Publish Chinese master to YouTube; Bilibili gets the same story with native metadata
T+6h
Xiaohongshu visual essay activates the existing 3K base and points to the full argument
T+24–48h
Newsletter carries sources and framework; X/LinkedIn carry one chart and judgment
Do not use Shorts as the channel identity. Cut clips only after the long-form episode proves a moment through retention, comments, or rewatches.
Visual language
One token set and six editorial primitives across still, GIF, and video.
Remotion rendererMP4 · GIF · still
Palette
Ink, paper, signal yellow, evidence green
Hierarchy
Claim → proof → mechanism → decision
Motion
Narrative state changes only
Reuse
16:9, 9:16, 1:1 from the same props
01
Mechanism map
Explain systems and causal chains
Nodes reveal along one causal path
02
Timeline
Company, technology, and market history
Events enter as the narration advances
03
Split compare
Old vs new workflow or competing choices
Synchronized contrast, no decorative movement
04
Evidence card
Quote, filing, chart, benchmark, or screenshot
Source first, claim second
05
Build replay
Code, product, and experiment walkthrough
Cursor, state change, result
06
Decision frame
Close with options, trade-offs, and action
One decision appears at a time
One research core, many outputs
YouTube / Bilibili
8–12 minute narrative master
Xiaohongshu
6–10 page visual essay from the claim ledger
Newsletter
Sources, caveats, framework, and product bridge
X / LinkedIn
One argument, one chart, one operator judgment
Markdown is the canonical text artifact. Platform rich text, captions, chapters, clips, and graphics are compiled from it; they are not rewritten independently.