The project's drive board is a mixed-construction PCB, with SMD 0805 passives alongside through-hole ICs and electrolytics [project CLAUDE.md §2]. Since it is a precision analog signal chain, one bad joint is not obvious. A cold joint is one where the solder never really melted onto both surfaces, so it looks finished when it is not and costs a night of debugging weeks later. So I treated learning to solder the way I treated the oscilloscope in §M3a, with deliberate practice on things that do not matter so that the drive board survives.
An FM radio from Coupang
For practice, I used an FM radio assembly kit from Coupang, sold in Korean as an "FM 라디오 조립 키트" (FM radio assembly kit, soldering electronic circuit, DIY, assembly required) [25]. The kit is a bag of resistors, capacitors, ICs, and a speaker, plus a bare board. When I finished the last joint, the radio actually played [builder, 2026-08-06].
Two consumables turned out to matter as much as the iron:
- Rosin-core solder wire, a Korean-made 70 g spool for electronics (신성 실납 70g) [27]. The wire is thin, and its rosin core means every touch of the iron carries a little of its own flux.
- Syringe flux paste [26], which I did not know I needed at first. Extra flux makes solder flow where you want it instead of balling up where it lands, and the syringe format puts it on the pad and nowhere else. I learned the rule every guide repeats, that when a joint is giving you trouble, the answer is more flux and almost never more heat.
The skill progression
I learned the skills in an order I would recommend to anyone, starting with through-hole joints on the radio kit until the motion is boring. Then come 0805 surface-mount passives, which look impossible at first and become routine once you learn to let surface tension hold the part in place. Desoldering and rework (wick, pump) follow. Inspection runs alongside all of it, so that you learn to tell a good joint (a shiny cone that wets both surfaces) from the matte, blobby, or starved joints that fail later.
The certification exam
I populated the project's drive board twice, which I had not planned. The first population was on the first manufactured version of the board, whose published design files later turned out to have a broken ground scheme. The full story of how the defect was found and what the board's original authors did about it is §M6's detective story.
When the corrected board (v2) arrived from JLCPCB, I populated it in a single overnight session on June 11–12, with all fresh parts and the DAC0800s in sockets [project CLAUDE.md §2]. I did it overnight because the project had a long queue behind this board, and the fastest path was to redo it immediately with the same parts list, the same order of operations, and hands already trained on this layout. The build order is in its own document, phase1_PCB_soldering_guide.md (34 KB of which-parts-first, polarity traps, and per-stage checks), effectively this chapter's technical appendix, published alongside it.
The joint I forgot
Once a board is populated, it goes through the project's bring-up protocol, which starts with Stage 0, electrical checks before the expensive ICs go into their sockets, with continuity and resistance measured against written expected values, stage by stage. On the v2 board, my memory is that the round of checks came up clean, with nothing flagged and nothing to reflow [builder, 2026-08-06], but the June notes disagree.
The one bad joint of my soldering story, the ground end of R3 on DAC channel 1, belongs to the first board and was caught and reflowed during that board's checkout [builder correction, 2026-08-06]. I had completely forgotten that joint existed until it came up again during this write-up. The June bring-up notes filed it under v2. My memory puts it on v1, which is what I have written above.
That memory is weak evidence, because it had lost the joint entirely, which makes it less reliable than a note written the same week. I have no third piece of evidence to break the tie. The disagreement is recorded in the corrections log, and until something settles it, which board carried that joint stays open.
One bad joint across two full board populations is a hit rate I'll happily take as a beginner. The unsettling part is how much of this project's history apparently lives in markdown files instead of in my memory.
Advice for beginners
Buy the practice kit before you need the skill, and finish it even after it gets boring. Buy flux in a syringe and use more of it than feels reasonable. Socket every IC you can't afford to lose. Test in stages against written expected values, because your eyes will pass joints your multimeter won't. Keep the log, because six weeks later it is the only record left, even on the day you decide to overrule a line of it.
The board still had to produce a clean 15 V sawtooth, and getting there from power-up is its own story (§M7). Before that comes the day I opened the published design files and found the ground plane missing, and that story starts in §M3c.