Fab-order check for KiCad and JLCPCB
Drop your schematic, board file, BOM and placement CSV. 38 checks run in this tab on the exact files you are about to upload, and nothing leaves your computer. When you want a person to sign off the order, it is $49.
.kicad_sch .kicad_pcb .kicad_pro and the BOM and placement .csv files
Runs in this tab. No upload, no account, no analytics or other third-party script on this page. How to verify that. On a phone, pick a .zip of the project.
What it catches
A respin is the entire order again plus weeks of waiting. These are the errors that survive ERC and DRC and show up when the boards arrive. The examples are from the sample report, a real open-source board.
Schematic, board, BOM and placement file are four exports that drift apart. The check lines them up: parts missing from the board, nets that differ, BOM lines with no placement, two footprints sharing one reference, malformed LCSC numbers.
Rules: pcb-sync, cpl-vs-pcb, bom-cpl-match, bom-vs-design, duplicate-refs, bom-lcsc
JLCPCB’s published limits, read from your board whatever your own design rules say: annular ring, drill size, track width, silkscreen, outline. It does not measure the gap between tracks; run KiCad’s DRC with JLCPCB’s minimum clearance for that. It also tells you whether the board qualifies for Economic PCBA or needs Standard.
Sample: 15 plated holes have an annular ring under 0.18 mm on U1 (thinnest 0.075 mm).
The things a reviewer looks at around every IC: bypass capacitors, regulator input and output capacitors, crystal load capacitors, I2C pull-ups, USB-C CC resistors, EN and BOOT pins, ESD on USB, MOSFET gates.
Sample: an LDO with no capacitor on its output, and a charger with none on either side.
Proof
This is the output for joric/nrfmicro, a public-domain open-source board. Seven serious findings. I opened the files and confirmed each one by hand, and the report says so next to each finding.
Labeled sample, not a client board. Not a statement about the quality of the original project.
Free report, then a person
Every report ends with a list called Still needs a person. It is honest about what an offline checker cannot judge. That list is exactly what the paid checks work through, by hand, on your board.
A person goes through your free report and the things software cannot see, then tells you in writing: order it, or fix these first.
Your schematic read against the datasheets, part by part, before you spend a day on layout.
Schematic, layout and the fab package, reviewed together, ending in the same sign-off.
Everything each option includes. Free for open-source boards: I review it and publish the report.
Accuracy, stated plainly
I measured it on open-source boards it had never seen, and checked every serious finding by hand. The numbers below are the latest untouched run, not a tuned one.
72% correct, 41% worth acting on. That is why every finding has an evidence line and why the paid tier has a person who confirms or dismisses each one. Seeded-bug test: when I removed a part from 29 real boards it had not been tuned on, it noticed 235 of 246 times. Method, misses and all 38 rules.
Who is behind this
UIUC electrical engineering student. I design boards in KiCad and order them assembled from JLCPCB.
It is one person’s tool, not a company’s: you deal with me, and I answer by email.
Questions
No. The checker is JavaScript that runs in this tab. This page loads no analytics and no other third-party script, and its security policy lets it talk only to this site, so it cannot send your files to anyone else. You can check: open your browser’s developer tools, go to the Network tab, drop a board, and watch. Files leave your computer only if you choose to send them to me for a paid review.
Use all three. They answer different questions.
Honest answer from the latest test on boards it had never seen: of 29 serious findings, 12 were real problems, 9 were true but harmless, and 8 were wrong (72% correct, 41% worth acting on). I then fixed the wrong ones. That fix is not counted as accuracy, because I looked at those boards first. Every finding names its rule and shows the nets or parts involved so you can check it in a minute. Full numbers, including the misses.
The free report ends with a list called “Still needs a person”. That list is the product. On your board it covers: BOM parts against JLCPCB’s live parts library (stock, lifecycle, Basic or Extended fee), pin-1 and polarity rotation for every polarised part, footprints that did not ship with KiCad against the datasheet land pattern, and every free finding confirmed or dismissed by hand. You get a one-page order-it-or-fix-these-first note.
Best: the schematic (all sheets), the .kicad_pcb, and the BOM and placement CSVs in JLCPCB’s format. A schematic alone works, and so does a netlist (.net). Checks that need a file you did not give are shown as not run, never as passed. A .zip of the project folder works too. On KiCad 8 and newer, include the .kicad_pro project file as well: bus aliases are stored there, and without it a bus drawn from an alias is not expanded.
Not for Altium, Eagle or EasyEDA. KiCad only: schematics (.kicad_sch) and boards from KiCad 6 to 10 are read directly. KiCad 5 keeps its schematic in the old .sch format, which this does not read. For a KiCad 5 project, export a netlist (.net) from the schematic and give it to the checker together with the .kicad_pcb and the BOM and placement files; the netlist stands in for the schematic.
I do. Mahir Patel: UIUC electrical engineering student. I design boards in KiCad and order them assembled from JLCPCB. I use the same checker you just ran, then do the parts it cannot. More about me.
Then I say so, and what I checked. A review lowers risk; it does not guarantee a working board. If it tells you nothing you did not already know, you get your money back.
Run the free check now. If it finds something, fix it. If you want someone to read the order before you pay for it, send me the board.