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-W43-Thukind longformweek W43date 2026-10-29campaign longform-youtubepillar it_devbeat —asset videoduration 140.9sground whitescenes 7

Checklist the per-video bar — engine/sim

98.2/100
plain languagevo coverageno dead airuniquenesscaption fitcompletenesscleanliness
quantitative quality · weights learn from your reviews (engine.sim.memory review longform-W43-Thu 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 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 7 scenes · 140.9s · comp_id + rendered still + tier + the script

#Layer (comp_id · still · tier)BeatTimecodeMotionLogoAudioVO / on-screen / caption
1s1
matches intent
shared field
signature-3d		open0–21.3sspatial-parallaxicon·ember-fill♪ bed_in
Today is interviews and discussion, and the trap is obvious: you record an hour, and you get ten minutes of anything worth keeping. The week is the plan behind the proof, and an interview is proof when it is planned. So here is how The Big T-M plans a conversation so the useful parts are not an accident.
on-screen: Interviews that say something
expected on screen: white ground · box hero in the shared Signal Field · Faber leads · node-graph · spatial-parallax · icon·ember-fill logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual node-graphshape boxground whitetreatment ember-fillmotion spatial-parallaxpower summoninstrument summon→Interviews that say something
2s2
matches intent
NumberedList
template		teach21.3–43.6skinetic-buildicon·wireframe♪ node_lock
The first move is to plan the questions as an ordered list, each one aimed at a single point you want made. Not a vague chat. A map. We store the questions with the answers we are hoping to surface. When you know the destination, you can steer the conversation back to it instead of drifting for an hour and praying.
on-screen: Write the questions as a map
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→Write the questions as a mapcurate items, nodes
3s3
matches intent
ChecklistCard
template		teach43.6–64.5skinetic-buildicon·wireframe♪ node_lock
Next, the tool step. Run the recording through a speech-to-text model that gives you word-level timestamps. Now the whole interview is searchable text tied to the exact second it was said. You stop scrubbing a timeline by ear. You read the transcript, find the line you want, and you already know where it lives in the footage.
on-screen: Transcribe with real timestamps
expected on screen: white ground · a ChecklistCard panel over a dimmed Signal Field · Faber leads · node-graph · kinetic-build · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id ChecklistCardvisual node-graphshape boxground whitetreatment wireframemotion kinetic-buildpower morphinstrument morph+laser→Transcribe with real timestampscurate items, nodes
4s4
matches intent
TerminalRun
template		teach64.5–87.5skinetic-buildicon·wireframe♪ node_lock
Then hand that transcript to the model and ask it to mark the strongest moments against your question map. It returns the timestamps for the lines that actually land. You cut from those marks. The model does the listening pass so you do not sit through the full hour twice, and the clips you keep are the ones tied to a real point.
on-screen: Let the model pull the clips
expected on screen: white ground · a TerminalRun panel over a dimmed Signal Field · Faber leads · node-graph · kinetic-build · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id TerminalRunvisual node-graphshape boxground whitetreatment wireframemotion kinetic-buildpower morphinstrument morph+laser→Let the model pull the clipscurate nodes
5s5
matches intent
JsonDiff
template		teach87.5–109.5skinetic-buildicon·wireframe♪ node_lock
The proof here is simple and honest: when the questions are planned and the transcript is timestamped, the edit comes together fast and stays on message. There is no pile of footage nobody can use. The plan did the hard thinking up front, so the conversation earns its place in the calendar instead of becoming a file you never open.
on-screen: The method survives the edit
expected on screen: white ground · a JsonDiff panel over a dimmed Signal Field · Faber leads · node-graph · kinetic-build · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id JsonDiffvisual node-graphshape boxground whitetreatment wireframemotion kinetic-buildpower morphinstrument morph+laser→The method survives the editcurate fileName, lines, nodes
6s6
matches intent
ComparisonTable
template		proof109.5–121.1sreceipts-counticon·wireframe♪ node_lock
That is the method behind these pieces too. Planned questions, timestamped transcript, model-marked moments. It is not a secret kept in an editor's head; it is steps you can run today.
on-screen: This is how we cut talk
expected on screen: white ground · a ComparisonTable panel over a dimmed Signal Field · Faber leads · receipts · receipts-count · icon·wireframe logo · caption bottom-left
spec (the prompt): comp_id ComparisonTablevisual receiptsshape boxground whitetreatment wireframemotion receipts-countpower receiptsinstrument spotlight→This is how we cut talkcurate colA, colB, rows, statsLabels
7s7
matches intent
shared field
signature-3d		resolve121.1–140.9scoalescenceicon·ember-fill♪ bed_out
So an interview as a system: write the questions as a map, transcribe with real timestamps, and let the model pull the clips. Do that and a long conversation becomes a tight piece without you living in the timeline. See how it fits the rest of the plan at office dot temerarii dot xyz.
on-screen: Plan the talk, then trim it
expected on screen: white ground · box hero in the shared Signal Field · Faber leads · coalescence · coalescence · icon·ember-fill logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual coalescenceshape boxground whitetreatment ember-fillmotion coalescencepower coalescenceinstrument coalescence→Plan the talk, then trim it

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)

youtubeHow to build custom apps and dashboards you own, with AI and code Renting your tools means renting your business. This video shows how to build the small custom apps and dashboards you actually need, with AI and code, so you own the operating layer and the IP instead of paying rent forever. The method: Start from the real job, not a template. Name the one thing the tool must do, the data it reads, and the one screen a person needs. A custom app is small when it is scoped to one job, and small is what you can actually own and maintain. Let the agent write the first version. Hand a coding agent the job, the data shape, and your brand tokens, and have it scaffold a working app you can run. You are reading and steering the code, not typing every line, but the code is yours, in your repo, not locked in someone's platform. Read from one source of truth. The dashboard pulls from the same files and data the rest of your system uses, so the numbers on screen match the numbers everywhere else. No second copy quietly drifting out of date. Keep it honest. Show real data or an honest empty state, never a fabricated number to fill a chart. A dashboard that lies is worse than no dashboard, because it teaches you to trust the wrong thing. Own the code and the IP. Because it lives in your repo under version control, you can change it, extend it, and hand it to the next person. The tool does not vanish when a vendor raises its price or shuts down. See how we run our own operating layer in public at office.temerarii.xyz. Keywords: custom apps, internal dashboards, AI coding, Claude Code, own your IP, no code alternative, build vs buy.

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

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