hardware-FeatherS3: TPS82130 doesn't power the FeatherS3 from 12V (two boards from one batch) — what should the unloaded 3.3V rail read?

[jkosharek]

First, thanks for this project — the controller itself works great on my unit.

Setup

• hardware-FeatherS3 PCB, JLCPCB-assembled from the repo's production files (BOM/CPL/gerbers unmodified)
• Unexpected Maker FeatherS3
• LG LAN120HSV5 wall-mounted unit, connected via CN-REMO (LMU36CHV Outside Unit)
• ESPHome 2026.7.4
• Also looking at connecting a setup to a LDN127HV4

What works
With the FeatherS3 powered over USB-C (3.3V pin seated in the socket, JST connected), everything is fine: status, settings, climate control, vane position, all reliable for hours. So the JST wiring, TJA1027, and signal path are all good.

Problem
Powered from the AC's 12V only (USB disconnected), the FeatherS3 never boots or turns off when USB-C is disconnected.

Measurements (12V applied at J1, FeatherS3 removed, DMM ~10Mohm)

• 12V at the J1 input: present
• 3.3V socket position: reads ~12V
• Across C1 (the 10uF 1210): ~12V
• A second, never-used board from the same batch measures identically

Additional data point
With the FeatherS3 seated and USB-powered, lifting the FeatherS3's 3.3V pin out of the socket turns RX into continuous invalid-checksum garbage; re-seating it restores clean communication. So the FeatherS3 appears to be back-feeding the 3.3V rail that supplies the TJA1027's logic side — the rail carries that load happily when driven from the FeatherS3, which suggests the TPS82130 output is genuinely not regulating rather than the rail being shorted.

Questions

1. With no FeatherS3 seated, what should the 3.3V rail measure on a known-good board — 3.3V, or is an unloaded reading meaningless on this design (I notice the BOM's only bulk cap is C1, so is the buck's output capacitance expected to come from the FeatherS3 itself)?
2. If unloaded-12V is abnormal: any known bad batches or JLCPCB substitutions at U1, or a likely EN-path fault to check?

Happy to post photos of the boards or scope the switch node if useful.

[cure]

Wrt question 1: this is a known issue, the fix is pending in PR #182. My older boards read roughly 12V on the 3.3V line when a FeatherS3 is not connected and the board is fed via 12V. This is because of a missing cap, which is fixed in PR #182 . You can also "rework" and add a cap yourself, see this comment.