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-W48-Monkind longformweek W48date 2026-11-30campaign longform-youtubepillar performancebeat asset videoduration 151.3sground redscenes 6

Checklist the per-video bar — engine/sim

100.0/100
plain languagevo coverageno dead airuniquenesscaption fitcompletenesscleanliness
quantitative quality · weights learn from your reviews (engine.sim.memory review longform-W48-Mon good|bad)
✓ all static checks pass — one-focal/scene · tier-by-beat · one-track caption · colorway · cast+shape correct · no banned/fabricated. (audio + visual tiers verify on the rendered finals — Phase 2)

Composition comp · template family · expected output

composition LongFormChaptersfamily / template LongFormChapters
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 6 scenes · 151.3s · comp_id + rendered still + tier + the script

#Layer (comp_id · still · tier)BeatTimecodeMotionLogoAudioVO / on-screen / caption
1s1
matches intent
shared field
signature-3d
open0–26.6sspatial-parallaxicon·white-knockout♪ bed_in
The operating layer this week is about systems that run themselves, and that includes your website. Today is web design, but not the usual kind where you pay for a pretty page and then watch it rot. We are going to wire a site that can read its own data and improve from it. You will learn how to let the page tell you what is broken instead of you hunting for it.
on-screen: A site that fixes itself
expected on screen: red ground · dodeca hero in the shared Signal Field · Mensor leads · node-graph · spatial-parallax · icon·white-knockout logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual node-graphshape dodecaground redtreatment white-knockoutmotion spatial-parallaxpower summoninstrument summon→A site that fixes itself
2s2
matches intent
NumberedList
template
teach26.6–51.400000000000006skinetic-buildicon·white-knockout♪ node_lock
Step one. Connect an agent to Google Search Console through an MCP server. Search Console is the free tool that shows the actual words people typed to find you. Most owners never open it. Wire it in once, and now the model can pull your real query data on command, the exact phrases sending people to each page, instead of you copying numbers into a spreadsheet by hand.
on-screen: Pull your own search data
expected on screen: red ground · a NumberedList panel over a dimmed Signal Field · Mensor leads · node-graph · kinetic-build · icon·white-knockout logo · caption bottom-left
spec (the prompt): comp_id NumberedListvisual node-graphshape dodecaground redtreatment white-knockoutmotion kinetic-buildpower morphinstrument morph+laser→Pull your own search datacurate items, nodes
3s3
matches intent
ChecklistCard
template
teach51.4–76.9skinetic-buildicon·liquid-chrome♪ node_lock
Step two. Hand that query data to the agent and tell it to rewrite the page heading and the first paragraph to match the words people actually used. If folks search a plain question and your page answers with empty buzz, the model swaps the buzz for the question. You read the change, you approve it, the agent edits the file. The page now speaks the visitor's language, not yours.
on-screen: Let the model rewrite the page
expected on screen: red ground · a ChecklistCard panel over a dimmed Signal Field · Mensor leads · node-graph · kinetic-build · icon·liquid-chrome logo · caption bottom-left
spec (the prompt): comp_id ChecklistCardvisual node-graphshape dodecaground redtreatment liquid-chromemotion kinetic-buildpower morphinstrument morph+laser→Let the model rewrite the pagecurate items, nodes
4s4
matches intent
DiffCard
template
teach76.9–104.2skinetic-buildicon·white-knockout♪ node_lock
Step three. Never push a change blind. Use a headless browser, a real browser with no window, driven by the agent, to load the new page and screenshot it on a phone size and a desktop size. The big T-M checks every change this way before it goes live. If the heading wraps wrong or a button falls off the screen, you see it in the shot, not in an angry email from a customer.
on-screen: Test it before anyone sees it
expected on screen: red ground · a DiffCard panel over a dimmed Signal Field · Mensor leads · node-graph · kinetic-build · icon·white-knockout logo · caption bottom-left
spec (the prompt): comp_id DiffCardvisual node-graphshape dodecaground redtreatment white-knockoutmotion kinetic-buildpower morphinstrument morph+laser→Test it before anyone sees itcurate lines, nodes
5s5
matches intent
StatScoreboard
template
proof104.2–127.2sreceipts-counticon·liquid-chrome♪ node_lock
Our proof is plain. The office site you can visit was not hand-coded once and left alone. It is rebuilt by agents that read our own search data and our own page checks, then redo the parts that are weak. No invented traffic figures here, just the honest fact that the site fixes itself faster than a person clicking through it ever could.
on-screen: We rebuild our own pages this way
expected on screen: red ground · a StatScoreboard panel over a dimmed Signal Field · Mensor leads · receipts · receipts-count · icon·liquid-chrome logo · caption bottom-left
spec (the prompt): comp_id StatScoreboardvisual receiptsshape dodecaground redtreatment liquid-chromemotion receipts-countpower receiptsinstrument spotlight→We rebuild our own pages this waycurate pillar, stats, statsLabels
6s6
matches intent
shared field
signature-3d
resolve127.2–151.3scoalescenceicon·white-knockout♪ bed_out
So a good website is not a one-time purchase. It is a small loop: read the real search words, let the model rewrite to match them, test in a headless browser, ship. Set that loop up once and your site quietly gets clearer every week. See the live version of this loop at office dot temerarii dot xyz, then point it at your own slowest page.
on-screen: Start at office.temerarii.xyz
expected on screen: red ground · dodeca hero in the shared Signal Field · Mensor leads · coalescence · coalescence · icon·white-knockout logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual coalescenceshape dodecaground redtreatment white-knockoutmotion coalescencepower coalescenceinstrument coalescence→Start at office.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)

youtubeBuild a Website That Reads Its Own Search Data and Fixes Itself This is web design, but not the usual kind where you pay for a pretty page and then watch it rot. In this video we wire a site that can read its own data and improve from it, so the page tells you what is broken instead of you hunting for it. The moves: - Pull your own search data. Connect an agent to Google Search Console through an MCP server. Search Console is the free tool that shows the actual words people typed to find you. Most owners never open it. Wire it in once and the model can pull your real query data on command. - Let the model rewrite the page. Hand that query data to the agent and tell it to rewrite the page heading and first paragraph to match the words people actually used. If folks search a plain question and your page answers with empty buzz, the model swaps the buzz for the question. You approve, the agent edits the file. - Test it before anyone sees it. Never push a change blind. Use a headless browser, a real browser with no window, driven by the agent, to load the new page and screenshot it on a phone size and a desktop size. If the heading wraps wrong, you see it in the shot, not in an angry email. The honest proof: the office site you can visit was not hand-coded once and left alone. It is rebuilt by agents that read our own search data and our own page checks, then redo the parts that are weak. No invented traffic figures, just the fact that the site fixes itself faster than a person clicking through it. The takeaway: read the real search words, let the model rewrite to match them, test in a headless browser, ship. Set that loop up once and point it at your slowest page. See the live version at office.temerarii.xyz. Keywords: web design, Google Search Console, SEO, headless browser, MCP, AI agent.

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