12 V to 5 V / 1 A Buck Molecule (TPS54202)

@@ -1,38 +1,56 @@ # The process video: how it was made -The three-minute video on the [main page](https://wiki.adom.inc/john/buck-12v5v-molecule) tells the whole story of this board: the brief, the circuit, the three-engine simulation, the live layout in KiCad, the current and heat analysis and the rework it caused, the silkscreen, the part models and their wiki pages, and the molecule on its scaffold. Every frame is real footage or a native render from this project; nothing is mocked up.+The video on the [main page](https://wiki.adom.inc/john/buck-12v5v-molecule) tells this board's story start to finish: the prompt in Adom Hydrogen, the circuit, the three-engine simulation, the live layout in KiCad, the current and heat analysis and the rework it caused, the silkscreen, the part models and their wiki pages, the molecule on its scaffold, and what it cost. Every frame is real footage or a native render from this project. -- 3 min 4 s, 1080p, 30 fps, narrated, with subtitles burned in so it also works muted.-- [Download the 1080p copy (43 MB)](docs/video/buck-process-1080p.mp4). The player on the main page is a 1600 px web copy (13 MB).+- 3 min 25 s, 1080p, 30 fps, narrated, with subtitles burned in so it also works muted.+- [Download the 1080p copy](docs/video/buck-process-v2-1080p.mp4). The player on the main page is a 1600 px web copy. -## Scenes and the clips behind them+## Scenes and the footage behind them -Each scene is seen through a slow, eased camera move (a crop that pushes in or pans across the frame), and long recordings are sped up to their action. A scene is exactly as long as its narration plus a short lead-in and tail, so nothing is rushed or padded; scenes cross-dissolve into each other.+Each scene is seen through a slow, eased camera move, and long recordings are sped up to their action. A scene is as long as its narration plus a short lead-in and tail; scenes cross-dissolve. All 3D shots are rendered frame by frame from the models, so they are smooth and free of cursors and browser chrome. -| starts at | scene | source |+| starts at | scene | footage | |---|---|---|-| 0:00 | Opening | KiCad 3D viewer tour from the rework run (`run-rework`, clip recorded 2026-09-30 10:48), played at about 2x with a slow push-in; title in Familjen Grotesk. |-| 0:13 | The circuit | The KiCad schematic from the molecule page, panned and zoomed (a 34 s capture), played at 1.2x. |-| 0:39 | Simulation | Our own adom-spice dashboard run, recorded 2026-10-01 in a background pup window: ngspice's curves landing (real time), LTspice's "2 engines agree" (real time), PSpice for TI starting and its "3 engines agree" (real time). The idle minutes between them are cut. Vendor run times stay hidden (licence). |-| 1:03 | Placement | First run, placement clip: 31 s of the AI placing 24 parts, played at 2.3x, ending on a push-in to the board. |-| 1:16 | Routing and pours | Rework run, pours clip: the board with its pours stripped back to bare traces (25 s at 3.1x, a close pan), then the pours flooding back in (real time). |-| 1:30 | Current and heat | First run, Adom Fields walkthrough: 108 s of the current-density and heat views at about 7x. |-| 1:44 | The rework | First run, routing-rework clip (the sense trace moving, 36 s at 3.5x), then the rework run's second routing visit (62 s at about 8x). |-| 2:02 | Silkscreen, placed live | Rework run, the 16-minute live silkscreen session at about 88x, with a slow pan across the board. |-| 2:12 | Part models | Native KiCad 10 renders (`kicad-cli pcb render`) of the board before and after the part-model cleanup: a dissolve, then camera stops on C1, U1, C2/C3 and D1. |-| 2:29 | Component pages | Turntables of four hero GLBs from the component pages (WS2812B-2020, IN-S63BTG, TPS54202DDCR, EEE-FK1V470P), recorded in a pup window, about 3 s each. |-| 2:40 | On the scaffold | The scaffold scene from `scaffold/viewer.html`, rendered frame by frame in a headless browser (no cursor, no browser chrome) at 2880 x 1620. |-| 2:48 | Close | A rendered closing card in the brand fonts over the after render of the board. |--The screen recordings are the ones on the [build clips](clips.md) and [rework clips](clips-rework.md) pages, taken from their raw 4K masters. The earlier two-minute cut that adom-aiflow made automatically is still here: [build-video.mp4](docs/build-video.mp4).+| 0:00 | The prompt | The prompt typed into a real Claude Code thread in Adom Hydrogen, filmed with Adom Bridge's window recorder. It opens on a full shot of Hydrogen, then pushes in on the composer. |+| 0:12 | Opening | A smooth orbit of the finished molecule, rendered frame by frame from its GLB (etched parts, green D1, rounded corners). |+| 0:24 | The circuit | The cleaned-up KiCad schematic, with the Adom-style TPS54202 symbol, explored by an eased camera move. |+| 0:39 | Simulation | Our own adom-spice dashboard run: ngspice, then LTspice, then PSpice for TI, ending on "3 engines agree on all 8 measurements". Idle time cut; vendor run times hidden (licence). |+| 1:02 | Placement | First run, placement clip: the AI placing 24 parts in KiCad 10, at 2.3x. |+| 1:15 | Routing and pours | Rework run: the routed traces with the pours stripped back, then the pours flooding back in. |+| 1:29 | Current and heat | First run, the Adom Fields current-density and heat walkthrough, at about 7x. |+| 1:44 | The rework | First run's routing rework, then the rework run's second routing visit. |+| 2:01 | Silkscreen, placed live | Rework run, the 16-minute live silkscreen session at about 88x. |+| 2:12 | Part models | Native KiCad 10 renders of the board before and after the part-model cleanup, then a smooth close-up of D1 from its etched side. |+| 2:28 | On the Adom wiki | Turntables of this board's seven parts as published on the Adom wiki, each labelled with its reference. |+| 2:43 | On the scaffold | The molecule on its LrgMed scaffold, wired to the control panel, rendered frame by frame. |+| 2:51 | What it cost | A cost card: tokens and model calls for the build and rework runs, their price at Claude API rates, and the practical cost on a 20x plan. |+| 3:10 | Close | A closing card over the finished board. |++The step recordings are the ones on the [build clips](clips.md) and [rework clips](clips-rework.md) pages, taken from their raw 4K masters.++## What it cost++Counted from this board's own Claude Code thread and its sub-agents, inside the two runs' clocks (the build, 2026-09-29 00:39 to 17:01 UTC, and the rework, 2026-09-30 01:08 to 11:33 UTC). The making of this video, and other threads open at the same time, are not counted.++| | tokens |+|---|---|+| model calls (Claude Opus 5.5) | 1,458 |+| output | 892,123 |+| input (fresh) | 3,415 |+| cache reads | 583,232,019 |+| cache writes | 5,228,973 |++At Claude API pricing for Opus 5.5 ($4 input, $20 output, $0.20 cache read and $8 one-hour cache write per million tokens) that is about **$176**. On a 20x subscription plan, divide by 20: about **$9**.  ## The voice-over script -Spoken by adom-tts (the Andrew voice, rate +4 %). Part numbers and units are written out the way they should be spoken; the subtitles show the written form.+Spoken by adom-tts (the Andrew voice). Part numbers and units are written out the way they are spoken; the subtitles show the written form.++**The prompt.** It starts with one prompt: create a twelve volt to five volt, one amp buck converter, using TI's T P S five four two oh two, as an Adom molecule.  **Opening.** This is a twelve volt to five volt, one amp buck converter. Claude designed it, laid it out, and documented it, start to finish, with Adom's tools. Here's how it came together. -**The circuit.** The brief came from a YouTube video, Hardware Design with G P T six: twelve volts in, five volts out at one amp. We kept the brief, and designed the circuit again from scratch, around TI's T P S five four two oh two. Every value comes from the datasheet equations: a fifteen microhenry inductor, three twenty-two microfarad output capacitors, and a sixty-eight picofarad feed-forward capacitor that shapes the loop.+**The circuit.** Claude worked the circuit out from the datasheet. Every value comes from its equations: a fifteen microhenry inductor, three twenty-two microfarad output capacitors, and a sixty-eight picofarad feed-forward capacitor that shapes the loop.  **Simulation.** Before any copper, the design was simulated in three engines at once, with the Adom SPICE dashboard. N G spice runs right in the container. L T spice, and TI's P spice, run on a Windows PC, driven through Adom Bridge. Start-up, both load steps, and the inductor ripple all overlay: three engines, eight measurements, all in agreement. @@ -48,16 +66,14 @@ Spoken by adom-tts (the Andrew voice, rate +4 %). Part numbers and units are wri  **Part models.** Then every 3D model got the same attention. Each part number is laser-etched flat onto the part, in the part's own colours. The LED is tinted to its datasheet wavelength, with its marking on the side, so the window stays clear. And every height matches its datasheet. -**Component pages.** Those models now live on each part's wiki page. Every 3D hero carries the footprint's pads, the silkscreen, and the pin names, so you can see exactly how the part lands on the board.+**On the Adom wiki.** And every part on this board now has its model on the Adom wiki, for every Adom user: the inductor, the capacitors, the resistors and the LED, each with its etched part number, its footprint, its silkscreen and its pin names.  **On the scaffold.** Finally, the molecule takes its place on a probing scaffold, wired to the control panel, ready for live measurements. +**What it cost.** What did it cost? Building the board took about nine hundred thousand output tokens, and five hundred eighty million tokens read back from the prompt cache. At Claude A P I pricing, that's about a hundred seventy-six dollars. On a twenty X plan, divide by twenty: about nine dollars.+ **Close.** One measured run, from a prompt to a finished board. Zero design rule errors. Three simulators in agreement. And every part on it, marked and modelled. The whole story, with every clip, is on the Adom wiki. -## Tools+## Review and tools -- **Narration:** `adom-tts say`, one file per scene, edge silence trimmed; the scene length is set from it.-- **Frames:** a small Python renderer reads each source through ffmpeg, crops it with eased keyframes and composites the text: Familjen Grotesk labels, JetBrains Mono chips and Satoshi subtitles on a soft dark glow, using the brand-font helpers from `gang-takes` in [adom/adom-video-post](https://wiki.adom.inc/adom/adom-video-post).-- **Assembly:** ffmpeg cross-dissolves the scenes (0.5 s), lays each narration at its scene's start and normalises loudness (-16 LUFS).-- **Captures made for the video:** the SPICE dashboard run (background pup window), the hero turntables (pup, three.js), the schematic pan (pup), and the scaffold orbit (headless, frame by frame).-- **Review:** frames from every scene were checked on contact sheets; fixes included re-cropping around a wandering mouse cursor, cropping out browser chrome, and re-timing the camera to the subtitles.+Every scene was loaded into adom-video-post's storyboard as its own clip and reviewed one by one before assembly. Narration is synthesised per scene with adom-tts; frames are composited by a small Python renderer with the brand fonts from the gang-takes helpers in [adom/adom-video-post](https://wiki.adom.inc/adom/adom-video-post); ffmpeg cross-dissolves the scenes and normalises loudness (-16 LUFS). The [aiflow-process-video](https://wiki.adom.inc/adom/adom-aiflow) skill in adom-aiflow now carries these practices for the next board.