Week 8: Scheduling and scale

What changes about prompts, review and distribution once you're producing fifty of something a week instead of one

12 April 2027

A prompt written once, by hand, for one video is a script. A prompt written to be filled in by a scheduler, over and over, is a template, and the distinction is the entire economic argument this course keeps circling. Halcyon’s four-stage pipeline (week 7) runs on eleven templates, not eleven thousand hand-written prompts — one per mood tag, each with three slots (verse, tone-word, b-roll descriptor) filled by whatever the scraper found that hour.

Batching changes what “review” means#

At one video, review means watching it. At two hundred videos generated overnight on a rented GPU block, review means setting a threshold and trusting it: a similarity score against a reference frame, a runtime check, a profanity filter on the generated script. Halcyon keeps roughly one render in twelve — not because eleven fail obviously, but because the threshold is tuned to the cost of a false positive (an ad the client would reject) against the cost of a false negative (wasted render minutes on GPU time already paid for). Nobody watches the eleven that get discarded; the threshold is the only judgement that touches them.

Platforms impose their own scheduling logic on top of the batch#

Creator folklore holds that TikTok’s recommender rewards near-daily consistency from an account more than it rewards any single video’s quality — neither platform publishes the weighting, and independent analyses mostly find quality and watch-time swamping raw posting frequency, but the belief is widespread enough to shape real behaviour: devotional-slop accounts post on a fixed clock rather than “when something good is ready,” and the schedule ends up upstream of the content, not downstream of it. The same folklore holds that YouTube Shorts tolerates burstier patterns; a paid Discord server for a niche slop serial wants irregular drops timed to a server event, because that audience is buying anticipation, not volume.

Template, plus batching, plus a schedule is a production system with its own cost structure, and cost structure is an economics argument, not a filmmaking one. That’s the case week 9 makes explicitly, using the numbers this week’s set generates.