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 longform-W23-Tuekind longformweek W23date 2026-06-09campaign longform-youtubepillar brandbeat asset videoduration 312.9sground redscenes 6

Checklist the per-video bar — engine/sim

98.0/100
plain languagevo coverageno dead airuniquenesscaption fitcompletenesscleanliness
quantitative quality · weights learn from your reviews (engine.sim.memory review longform-W23-Tue good|bad)
⚠ 1 flag(s) — not yet ship-ready: copy_generic · see docs/strategy/VIDEO-CHECKLIST.md

Composition comp · template family · expected output

composition LongFormChaptersfamily / template LongFormChapters
9:16 Reelpending1:1 Squarepending16:9 Widerendered9:16 4Kpending1:1 4Kpending16:9 4KpendingGIF (SMS)pending
▶ open rendered mp4
expected output: 1/7 rendered — same matrix the /media preview surfaces for this asset.

Composition layer × scene 6 scenes · 312.9s · comp_id + rendered still + tier + the script

#Layer (comp_id · still · tier)BeatTimecodeMotionLogoAudioVO / on-screen / caption
1s1
matches intent
shared field
signature-3d
open0–53.5sspatial-parallaxicon·white-knockout♪ bed_in
A logo is not a brand. A logo is one drawing. A brand is the system that decides how every next thing you make looks, sounds, and behaves, so the hundredth asset costs less than the first and still reads as you. The old way was a static PDF style guide. A designer would open it, eyeball the swatches, and hand-place the colors, and three months later the blue had drifted and nobody could say which blue was right. We do not do that anymore. We treat identity as data the machine reads, not as a memory the team is supposed to hold. Over the next few chapters I will walk you through the whole method, end to end, naming the real files, so you can build one that compounds instead of one that decays.
on-screen: Identity that compounds
expected on screen: red ground · icosa hero in the shared Signal Field · Signum leads · node-graph · spatial-parallax · icon·white-knockout logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual node-graphshape icosaground redtreatment white-knockoutmotion spatial-parallaxpower summoninstrument summon→Identity that compounds
2s2
matches intent
NumberedList
template
teach53.5–113.0skinetic-buildicon·white-knockout♪ node_lock
It starts with tokens, not pixels. Before anyone draws a single frame, the identity is written as values in a file: every color, the type family, the spacing scale, the motion timing, the contrast floor, all of it sitting in one tokens source the renderer reads at build time. Tahoma is not a suggestion in a doc, it is the type token every surface pulls. Red is not a vibe, it is a hex value with a rule attached. The point of writing it as data is a quieter kind of payoff: you change the token once, and the website, the video lower-thirds, the email signature, and the deck all update from that same change. Nobody re-keys a hex code by hand across forty files. There is one number, in one place, and every surface obeys it. That is what stops drift before it starts.
on-screen: Tokens before pixels
expected on screen: red ground · a NumberedList panel over a dimmed Signal Field · Signum leads · node-graph · kinetic-build · icon·white-knockout logo · caption bottom-left
spec (the prompt): comp_id NumberedListvisual node-graphshape icosaground redtreatment white-knockoutmotion kinetic-buildpower morphinstrument morph+laser→Tokens before pixelscurate items, nodes
3s3
matches intent
ChecklistCard
template
teach113.0–166.2skinetic-buildicon·liquid-chrome♪ node_lock
Then the mark. The wordmark never gets re-typeset, ever. People want to retype it in a new font for a one-off, and that is exactly how a brand quietly becomes three brands. So we lock it as a vector and we never re-parse it. It is recolored only by a fixed, named set of treatments, the white-knockout on black being the workhorse, each treatment defined once and reused everywhere. We even enforce it with a check that refuses to let the written name be typed onto an asset as live text; the logo carries the name, the caption can say it, the voice can say it, but the frame uses the locked mark. That discipline is the boring reason the same brand reads as the same brand in a sixty-second video and a plain-text email. Same mark, same rules, no exceptions.
on-screen: One mark, six treatments
expected on screen: red ground · a ChecklistCard panel over a dimmed Signal Field · Signum leads · node-graph · kinetic-build · icon·liquid-chrome logo · caption bottom-left
spec (the prompt): comp_id ChecklistCardvisual node-graphshape icosaground redtreatment liquid-chromemotion kinetic-buildpower morphinstrument morph+laser→One mark, six treatmentscurate items, nodes
4s4
matches intent
WireframeMock
template
teach166.2–214.79999999999998skinetic-buildicon·white-knockout♪ node_lock
Reading the text is the next rule, and it is a rule, not a taste. Every color in the system names the ground it is allowed to sit on and the contrast it has to clear. Red lives on black. Red stays under eight percent of the frame, so it reads as a signal and not as noise. Here is the actual move. We do not let a designer squint at a mockup and call it fine. A gate runs the contrast math and the brand math on every asset before it ships. If a caption would be too hard to read, or the mark sits on a ground it was never cleared for, the gate fails it and it does not go out. A program checks the text every time, so nothing leaves the shop that you cannot read.
on-screen: Legibility is a rule
expected on screen: red ground · a WireframeMock panel over a dimmed Signal Field · Signum leads · node-graph · kinetic-build · icon·white-knockout logo · caption bottom-left
spec (the prompt): comp_id WireframeMockvisual node-graphshape icosaground redtreatment white-knockoutmotion kinetic-buildpower morphinstrument morph+laser→Legibility is a rulecurate nodes
5s5
matches intent
KpiGrid
template
proof214.8–270.1sreceipts-counticon·liquid-chrome♪ node_lock
Now the proof, because receipts beat claims. Our own brand book is not a document that describes the tokens from a distance and slowly falls out of sync with them. It renders straight from the tokens themselves. The human-readable page you look at and the machine config the renderer obeys are generated from the same source, which means they physically cannot disagree, because they are the same data wearing two faces. The same engine that builds the book builds the office site and renders these videos, all reading that one tokens file. So when we say one brief produces nine outputs that are all on-brand, that is not a slogan, it is the file structure. And the whole book is public. You can open it, read every rule, and check our work against what actually ships.
on-screen: The book is public
expected on screen: red ground · a KpiGrid panel over a dimmed Signal Field · Signum leads · receipts · receipts-count · icon·liquid-chrome logo · caption bottom-left
spec (the prompt): comp_id KpiGridvisual receiptsshape icosaground redtreatment liquid-chromemotion receipts-countpower receiptsinstrument spotlight→The book is publiccurate kpis, statsLabels
6s6
matches intent
shared field
signature-3d
resolve270.1–312.90000000000003scoalescenceicon·white-knockout♪ bed_out
So here is the takeaway, minus the hype. Build the system, not just the logo. Put your identity in tokens, lock your mark and never re-typeset it, write your legibility as a contrast floor a gate enforces, and render your brand book from the same source every surface reads. Do that, and consistency stops being a thing you police by hand and becomes a thing the build does for free, while every new asset gets cheaper and more recognizable than the last. That is identity that compounds. The method is all out in the open, and so is ours. Go look at the whole system, running live, at temerarii dot xyz.
on-screen: temerarii.xyz
expected on screen: red ground · icosa hero in the shared Signal Field · Signum leads · coalescence · coalescence · icon·white-knockout logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual coalescenceshape icosaground redtreatment white-knockoutmotion coalescencepower coalescenceinstrument coalescence→temerarii.xyz

Format stack 1 aspects · same scenes[], re-cropped

16:9
1920×1080
X/Twitter · YouTube · LinkedIn video

Channels 2 destinations

YouTubeBlog

Social captions supplemental published copy · per channel (comp_id level)

youtubeA Logo Is Not a Brand: Build an Identity From Design Tokens A logo is one drawing. A brand is the system that decides how every next thing looks, sounds, and behaves, so the hundredth asset costs less than the first and still reads as you. This video walks that system, naming the real files. What you will learn: - Tokens before pixels: write every color, the type family, spacing, motion timing, and contrast floor as values in one file the renderer reads. Change the token once and every surface updates. - One mark, locked: never re-typeset the wordmark. Keep it as a vector, recolor it only with named treatments, and enforce it with a check that refuses the typed name on an asset. - Legibility is a rule, not a taste: every color names the ground it sits on and the contrast it must clear, and a gate runs the math on every asset before it ships. Then the proof: our brand book renders straight from the tokens, so the human-readable page and the machine config physically cannot disagree. The whole book is public, so you can check our work against what ships. Keywords: brand identity, design tokens, brand book, logo system, contrast, visual consistency. See the whole system run in public at temerarii.xyz.

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