Back Office · office.temerarii.xyz
One asset, all the way in — composition, the wireframe + storyboard, the output format stack, and the template, all read from the SAME content-index record. The expected output matches what /media surfaces for this post.
post social-W33-Sat-4kind threadweek W33date 2026-08-22campaign thread · Satpillar staff_trainingbeat Satasset videoduration 65.8sground blackscenes 9

Checklist the per-video bar — engine/sim

82.0/100
plain languagevo coverageno dead airuniquenesscaption fitcompletenesscleanliness
quantitative quality · weights learn from your reviews (engine.sim.memory review social-W33-Sat-4 good|bad)
⚠ 4 flag(s) — not yet ship-ready: copy_generictoo_complexdup_sequence_across_assetsdup_set_across_assets · see docs/strategy/VIDEO-CHECKLIST.md

Composition comp · template family · expected output

composition SceneReelfamily / template SignalFieldReel
9:16 Reelpending1:1 Squarepending16:9 Widepending9:16 4Kpending1:1 4Kpending16:9 4KpendingGIF (SMS)pending
render pending — silent master not yet on disk
expected output: 0/7 rendered — same matrix the /media preview surfaces for this asset.

Composition layer × scene 9 scenes · 65.8s · comp_id + rendered still + tier + the script

#Layer (comp_id · still · tier)BeatTimecodeMotionLogoAudioVO / on-screen / caption
1s1
matches intent
shared field
signature-3d		open0–7.0sspatial-parallaxicon·liquid-chrome♪ bed_in
We are heading toward a world where teaching a model from examples beats writing the rules by hand.
on-screen: Soon teaching a model beats coding one
expected on screen: black ground · knot hero in the shared Signal Field · Magister leads · node-graph · spatial-parallax · icon·liquid-chrome logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual node-graphshape knotground blacktreatment liquid-chromemotion spatial-parallaxpower summoninstrument summon→Soon teaching a model beats coding one
2s2
matches intent
shared field
signature-3d		hook7.0–14.7skinetic-buildicon·wireframe♪ node_lock
Old way, you tried to write a rule for every case and the list grew until nobody could maintain it.
on-screen: The old way: hand-code every rule
expected on screen: black ground · knot hero in the shared Signal Field · Magister leads · mark · kinetic-build · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual markshape knotground blacktreatment wireframemotion kinetic-buildpower laser-lockinstrument laser-trace→The old way: hand-code every rule
3s3
matches intent
NumberedList
template		teach14.7–22.799999999999997skinetic-buildicon·wireframe♪ node_lock
Here is the move: instead of rules, collect clean examples of right and wrong, and let the model learn the line.
on-screen: Label good examples, not rules
expected on screen: black ground · a NumberedList panel over a dimmed Signal Field · Magister leads · node-graph · kinetic-build · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id NumberedListvisual node-graphshape knotground blacktreatment wireframemotion kinetic-buildpower morphinstrument morph+laser→Label good examples, not rulescurate items, nodes
4s4
matches intent
RankList
template		proof22.8–29.8sreceipts-counticon·wireframe♪ node_lock
Good examples catch the fuzzy cases that no written rule could ever quite manage to put into words.
on-screen: Examples teach what rules can't say
expected on screen: black ground · a RankList panel over a dimmed Signal Field · Magister leads · receipts · receipts-count · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id RankListvisual receiptsshape knotground blacktreatment wireframemotion receipts-countpower receiptsinstrument spotlight→Examples teach what rules can't saycurate rows, statsLabels
5s5
matches intent
CodeWindow
template		diff29.8–37.1scrossfade-8ficon·wireframe♪ node_lock
A fancy model on messy data fails. A simple model on clean, honest examples quietly wins almost every time.
on-screen: Good data beats a clever model
expected on screen: black ground · a CodeWindow panel over a dimmed Signal Field · Magister leads · code · crossfade-8f · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id CodeWindowvisual codeshape knotground blacktreatment wireframemotion crossfade-8fpower morphinstrument morph→Good data beats a clever modelcurate codeLines, windowTitle
6s6
matches intent
ChecklistCard
template		teach37.1–44.800000000000004skinetic-buildicon·wireframe♪ node_lock
Now look hard at where the model is confidently wrong, because those mistakes show you the gaps in your examples.
on-screen: Check where it's still confidently wrong
expected on screen: black ground · a ChecklistCard panel over a dimmed Signal Field · Magister leads · node-graph · kinetic-build · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id ChecklistCardvisual node-graphshape knotground blacktreatment wireframemotion kinetic-buildpower morphinstrument morph+laser→Check where it's still confidently wrongcurate items, nodes
7s7
matches intent
StatScoreboard
template		proof44.8–52.5sreceipts-counticon·wireframe♪ node_lock
Add a few examples right where it failed, retrain, and the model gets sharper without touching the code at all.
on-screen: Fix the data, the model gets sharper
expected on screen: black ground · a StatScoreboard panel over a dimmed Signal Field · Magister leads · receipts · receipts-count · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id StatScoreboardvisual receiptsshape knotground blacktreatment wireframemotion receipts-countpower receiptsinstrument spotlight→Fix the data, the model gets sharpercurate pillar, stats, statsLabels
8s8
matches intent
CheatSheet
template		step52.5–58.8skinetic-buildicon·wireframe♪ node_lock
Three steps: collect honest examples, train, then inspect the failures and feed those gaps back in.
on-screen: Step: collect, train, inspect failures
expected on screen: black ground · a CheatSheet panel over a dimmed Signal Field · Magister leads · pipeline · kinetic-build · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id CheatSheetvisual pipelineshape knotground blacktreatment wireframemotion kinetic-buildpower throwinstrument laser-trace→Step: collect, train, inspect failurescurate points, stages, steps, uses
9s9
matches intent
shared field
signature-3d		resolve58.8–65.8scoalescenceicon·liquid-chrome♪ bed_out
The job is shifting from writing clever rules to curating honest data the model can actually learn from.
on-screen: Curate the data, not the rules
expected on screen: black ground · knot hero in the shared Signal Field · Magister leads · coalescence · coalescence · icon·liquid-chrome logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual coalescenceshape knotground blacktreatment liquid-chromemotion coalescencepower coalescenceinstrument coalescence→Curate the data, not the rules

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)

tiktokSoon teaching a model from examples beats hand-coding the rules. (AI-assisted) The move: collect clean examples of right and wrong, let the model learn the line, then inspect where it's confidently WRONG and feed those gaps back in. Good data beats a clever model.
instagramTeach with examples, not rules. Collect clean right-and-wrong examples. Let the model learn the line. Fix the data where it fails. #machinelearning #ml #ai #datascience #aitools
linkedinWe're heading toward a world where teaching a model from examples beats writing the rules by hand. The move: instead of coding a rule for every case, collect clean examples of right and wrong and let the model learn the line. Good examples catch the fuzzy cases no written rule could ever put into words. A fancy model on messy data fails; a simple model on clean, honest examples quietly wins. Then look hard at where the model is confidently wrong, because those mistakes reveal the gaps in your examples. Add a few right where it failed, retrain, and it sharpens without touching the code. The job is shifting from writing clever rules to curating honest data.
xSoon teaching a model from examples beats hand-coding rules. Collect clean right-and-wrong examples, let it learn the line, then inspect where it's confidently wrong and feed those gaps back in. Good data beats a clever model. temerarii.com
facebookSoon teaching a model from examples beats writing rules by hand. Collect clean examples of right and wrong, let the model learn the line, then look at where it's confidently wrong and add examples right there. A simple model on clean data beats a fancy model on messy data. temerarii.com
threadsSoon teaching a model from examples beats hand-coding rules. Collect clean right-and-wrong examples, let it learn the line, then inspect where it's confidently wrong and feed those gaps back in. Good data beats a clever model.
pinterestMachine learning the practical way: teach a model with clean right-and-wrong examples instead of hand-coded rules, then fix the data where it's confidently wrong. ML basics, data curation, example-based learning, model debugging, data quality over model complexity.
blueskySoon teaching a model from examples beats hand-coding rules. Collect clean right-and-wrong examples, let it learn the line, then inspect where it's confidently wrong and feed those gaps back in. Good data beats a clever model. temerarii.com
youtubeMachine Learning: Curate the Data, Not the Rules We're heading toward a world where teaching a model from examples beats writing the rules by hand. This short shows the move: instead of coding a rule for every case, collect clean examples of right and wrong and let the model learn the line. Good examples catch the fuzzy cases no rule could put into words. A simple model on clean data beats a fancy model on messy data. Then inspect where the model is confidently wrong, add examples right there, and retrain. The job is shifting from clever rules to honest data. #machinelearning #ml #datascience

Cross-links every lens is a view on this one record

→ Wireframe · scenes→ Calendar week→ Media (audience preview)→ Storyboard hub→ Day view→ Channel view