channel-strategy
When to use
Niche or idea evaluation, content planning, cadence, weekly review. Not for writing a script (see story-writing) or cutting clips (see shorts-from-peaks).
Evaluation workflow
- Get data first:
evaluate_niche(topic)for the score,search_videosfor the topic (views, virality per video),get_peakson 3-5 of them. No tool numbers = mark the criterion as an estimate. - Score each criterion 0-100 and weigh it:
| Criterion | Weight | Signal |
|---|---|---|
| Demand | 25 | views of top videos, upload velocity, trending queries |
| Source supply | 20 | how many videos have heatmap peaks; permitted/official sources |
| Clip-ability | 15 | dramatic or visual moments, short punchlines |
| Competition | 15 | existing shorts channels, their followers/views (saturation; high = low score) |
| Legal risk | 15 | owner copyright strictness, Content ID history (high risk = low score) |
| Monetization/longevity | 10 | evergreen vs news spike, affiliate/sponsor fit |
- Total = sum(score x weight) / 100. Verdict: Go >= 70, Maybe 50-69, No-go < 50.
- Always add: "what would make it a Go" (the 1-2 weakest criteria and a fix) and 3 starter content ideas.
- Cite the numbers behind each score; tag the rest "estimate".
Channel type fit
shorts/reuse: score all six criteria; source supply and legal risk decide.ai_anime: no third-party source. Replace Source supply with "premise strength" (hook in one sentence, repeatable conflict, 20+ episodes of runway) and Legal risk with "originality risk" (resembles a known franchise?).
Cadence
- Start with 1 post/day at fixed hours (e.g. 12h, 18h, 21h BRT); 2-3/day only when the queue buffer is healthy.
- Keep 3-7 days of approved buffer; batch production weekly.
- Series: one episode per day beats several in one burst; end each on a cliffhanger.
Metrics and review
- Per post: 3 s views, average watch time, completion %, shares, follows. Use only what the app returns; if analytics are not connected, say so and ask the user for numbers.
- Weekly loop: compare top and bottom 20%, name the hook or format that differs, record the learning in a note, propose one change.
- Flop diagnosis order: hook (first 3 s) -> length -> caption legibility -> topic fit -> posting time.
Proposing next actions
- End every answer with one concrete action that fits the autonomy level, phrased as a yes/no.
- Prefer the action that produces data or a draft (cheap, reversible) before one that commits (queue, schedule).
- If the verdict is No-go, propose the nearest viable niche, not a lecture.
Rules
- Rumors are labeled; no fabricated stats; nothing publishes without approval (docs/LEGAL.md).