Checklist the per-video bar — engine/sim
77.6/100
quantitative quality · weights learn from your reviews (
engine.sim.memory review thread-minus-the-hype-W49-Thu good|bad)⚠ 7 flag(s) — not yet ship-ready: copy_genericlow_vo_coveragedead_airgeneric_scene · see docs/strategy/VIDEO-CHECKLIST.md
Composition comp · template family · expected output
9:16 Reelrendered1:1 Squarepending16:9 Widepending9:16 4Kpending1:1 4Kpending16:9 4KpendingGIF (SMS)pending
expected output: 1/7 rendered — same matrix the /media preview surfaces for this asset.
Composition layer × scene 5 scenes · 27.0s · comp_id + rendered still + tier + the script
| # | Layer (comp_id · still · tier) | Beat | Timecode | Motion | Logo | Audio | VO / on-screen / caption |
|---|---|---|---|---|---|---|---|
| 1 |  matches intent shared field signature-3d | hook | 0–5.0s | kinetic-build | icon·wireframe | ♪ node_lock | The reflex when answers are off: fine-tune a model on company data. on-screen: "Fine-tune it on your data" |
expected on screen: white ground · box hero in the shared Signal Field · Faber leads · mark · kinetic-build · icon·wireframe logo · caption bottom-left spec (the prompt): comp_id shared Signal Fieldvisual markshape boxground whitetreatment wireframemotion kinetic-buildpower laser-lockinstrument laser-trace→"Fine-tune it on your data" | |||||||
| 2 |  first render · fix pending StackTrace templatedead_air | diff | 5.0–10.0s | crossfade-8f | icon·wireframe | ♪ node_lock | What we reach for first is better context. We feed the model fresh docs at ask time, so it reads the new facts instead of guessing. on-screen: Reality: retrieval beats training |
expected on screen: white ground · a StackTrace panel over a dimmed Signal Field · Faber leads · code · crossfade-8f · icon·wireframe logo · caption bottom-left spec (the prompt): comp_id StackTracevisual codeshape boxground whitetreatment wireframemotion crossfade-8fpower morphinstrument morph→Reality: retrieval beats trainingcurate codeLines, errMsg, errType, fix, frames | |||||||
| 3 |  matches intent NumberedList template | teach | 10.0–17.0s | kinetic-build | icon·wireframe | ♪ node_lock | The real cost of fine-tuning for facts is a frozen model that's already out of date by launch. Try retrieval first. on-screen: The cost: a stale, locked model |
expected on screen: white ground · a NumberedList panel over a dimmed Signal Field · Faber leads · node-graph · kinetic-build · icon·wireframe logo · caption bottom-left spec (the prompt): comp_id NumberedListvisual node-graphshape boxground whitetreatment wireframemotion kinetic-buildpower morphinstrument morph+laser→The cost: a stale, locked modelcurate items, nodes | |||||||
| 4 |  first render · fix pending StatScoreboard templatedead_airgeneric_scene | proof | 17.0–22.0s | receipts-count | icon·wireframe | ♪ node_lock | Minus the Hype |
expected on screen: white ground · a StatScoreboard panel over a dimmed Signal Field · Faber leads · receipts · receipts-count · icon·wireframe logo · caption bottom-left spec (the prompt): comp_id StatScoreboardvisual receiptsshape boxground whitetreatment wireframemotion receipts-countpower receiptsinstrument spotlight→Minus the Hypecurate pillar, stats, statsLabels | |||||||
| 5 |  first render · fix pending shared field signature-3ddead_airgeneric_scene | resolve | 22.0–27.0s | coalescence | icon·color | ♪ bed_out | Minus the Hype |
expected on screen: white ground · box hero in the shared Signal Field · Faber leads · coalescence · coalescence · icon·color logo · caption bottom-left spec (the prompt): comp_id shared Signal Fieldvisual coalescenceshape boxground whitetreatment colormotion coalescencepower coalescenceinstrument coalescence→Minus the Hype | |||||||
Format stack 3 aspects · same scenes[], re-cropped
9:16
1080×1920
Stories · TikTok · YouTube Shorts · Reels
1:1
1080×1080
LinkedIn · Facebook · Instagram
16:9
1920×1080
X/Twitter · YouTube · LinkedIn video
Channels 9 destinations
LinkedInX/TwitterYouTubeInstagramFacebookThreadsTikTokPinterestBluesky
Social captions supplemental published copy · per channel (comp_id level)
tiktok"Fine-tune it on your data," the reflex when answers are off. (AI-assisted)
What we reach for first is better context. We feed the model fresh docs at ask time, so it reads the new facts instead of guessing. The real cost of fine-tuning for facts is a frozen model that is already out of date by launch. Try retrieval first.
instagram"Fine-tune it on your data."
The reflex when answers are off.
Reach for better context first.
Feed the model fresh docs at ask time, so it reads new facts instead of guessing.
#temerarii #rag #llm #aidev #retrieval
linkedin"Fine-tune a model on your company data." The reflex when answers come back wrong.
What we reach for first is better context. We feed the model fresh docs at ask time, so it reads the new facts instead of guessing. Retrieval, not training, for anything that changes.
The real cost of fine-tuning for facts is a frozen model that is already out of date by launch. Try retrieval first.
x"Fine-tune it on your data," the reflex when answers are off.
Reach for better context first. Feed the model fresh docs at ask time so it reads new facts, not guesses. Fine-tuning for facts ships a stale model. Try retrieval first.
facebook"Fine-tune it on your data," the reflex when answers are off.
What we reach for first is better context. We feed the model fresh docs at ask time, so it reads the new facts instead of guessing. The real cost of fine-tuning for facts is a frozen model that is already out of date by launch.
Try retrieval before you train.
threads"Fine-tune it on your data," the reflex when answers are off.
What we reach for first is better context. We feed the model fresh docs at ask time, so it reads the new facts instead of guessing.
The real cost of fine-tuning for facts is a frozen model that is out of date by launch. Try retrieval first.
pinterestMyth-busting fine-tuning an LLM on company data. Why retrieval beats training for facts that change, and how feeding fresh docs at ask time avoids a stale frozen model. RAG vs fine-tuning method for AI developers.
bluesky"Fine-tune it on your data," the reflex when answers are off. Reach for better context first. Feed the model fresh docs at ask time so it reads new facts, not guesses. Fine-tuning for facts ships a stale model. Try retrieval first.
youtubeThe Fine-Tune Myth: Why Retrieval Beats Training for Facts
The reflex when answers are off is to fine-tune a model on company data. What we reach for first is better context. We feed the model fresh docs at ask time, so it reads the new facts instead of guessing. The real cost of fine-tuning for facts is a frozen model that is already out of date by launch. Try retrieval first. AI-assisted production by Temerarii.