Session Handoff [rv3028-eeprom]: MeshCore #3421 converged at 0f038054 after an all-PASS acceptance matrix; RV3028 position frozen #23

Next steps, in priority order

1. Sweep every upstream PR and issue authored by ptr727 (the commands are in link Session Handoff [rv3028-eeprom]: All upstream PRs answered; watch for maintainer review #18, step 1). Anything with updatedAt after 2026-10-07T01:00Z is new. Done means each maintainer comment is answered, or brought to Pieter as a dialog.
2. #3421 is converged at 0f038054. Copilot found nothing, all threads are resolved, and the acceptance matrix passes. The RV3028 position is FROZEN (see New learnings). Change code only if a maintainer requires it or a test fails, and park every other finding for Pieter. Any code change re-runs the full acceptance matrix A-H on the exact commit before it is pushed. Then Pieter re-requests Copilot in the web UI, asked as a dialog with the URL.
3. When Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421 or #3545 merges, rebase the other one. The two conflict with each other in AutoDiscoverRTCClock::begin(), though both merge cleanly onto dev. If the rebase lands on Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421, re-run the matrix on the new commit, under the rule in step 2.
4. Keep the two shared paragraphs word for word wherever they are carried. "Cross-project consistency" is in Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421, zephyr#121389 and ZephCore#103. "12-hour mode" is in Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421 and ZephCore#104; Pieter chose not to carry it on Zephyr. Any edit happens in all copies together, with Pieter's approval and fully qualified cross-repo references.
5. Parked, not code (Pieter's choice):
   • RTClib's DS3231 now() ignores the 12-hour flag. Revisit after the RTC PRs land.
   • The two MeshCore bugs from link Session Handoff [rv3028-eeprom]: Two RTC PRs re-pushed with cross-firmware fixes; triage Copilot headlines #19, step 4, stay ON HOLD until the RTC PRs land.
6. Cleanup stays deferred ("Watch only"). Keep the hardware builds and worktrees until a PR merges or is declined.

External blockers

• MeshCore maintainers: review of Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421, Adopt an RTC only if its time registers read like that chip 🤖🤖 meshcore-dev/MeshCore#3544, Store RV3028 backup switchover and trickle charger config in EEPROM 🤖🤖 meshcore-dev/MeshCore#3545, Write and read the RX8130CE WEEK register as one-hot 🤖🤖 meshcore-dev/MeshCore#3550, Reject a negative sensor list start index 🤖🤖 meshcore-dev/MeshCore#3433 and Pass TELEM_RAK12500_ADDRESS to the RAK12500 I2C probe 🤖🤖 meshcore-dev/MeshCore#3434, plus Do not feed an invalid GPS time into the RTC 🤖🤖 meshcore-dev/MeshCore#3422, Make GPS status reporting say what was actually checked 🤖🤖 meshcore-dev/MeshCore#3423 and Treat the GPS enable pin as a pin number, not a truth value 🤖🤖 meshcore-dev/MeshCore#3425 (draft). Nothing is owed from our side.
• Copilot re-request: after any upstream push it still needs Pieter in the web UI. Ask for it in a dialog, with the PR URL and the "Reviewers" sidebar location.
• Counterparts: drivers: rtc: rv3028: reject time accesses cut short by a switchover zephyrproject-rtos/zephyr#121389 and drivers: rv3028: disable backup switchover during EEPROM access zephyrproject-rtos/zephyr#121252, and rtc: write the RV3028 time with an A0h year mark, and only 2000-2099 liquidraver/ZephCore#103 and rtc: select 24-hour mode before a time write, and take no time in 12-hour mode liquidraver/ZephCore#104, wait on their own reviewers. Each one's peer session owns it.

Internal dependencies

• Step 3 waits on an upstream merge. Step 5 waits on the RTC PRs landing.
• A Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421 code change (step 2) must finish its matrix run before step 3's rebase, so that only one matrix runs per pushed commit.

State

Re-derive all of this; it was true at 2026-10-07T01:30Z.

• Upstream dev moved to f3bcb363, two OTA BLE-name commits unrelated to ours. All six fix branches still merge onto it cleanly.
• Counterparts:
  • drivers: rtc: rv3028: reject time accesses cut short by a switchover zephyrproject-rtos/zephyr#121389 is at 82e160257cc; drivers: rv3028: disable backup switchover during EEPROM access zephyrproject-rtos/zephyr#121252 is at 751beb04cba.
  • rtc: write the RV3028 time with an A0h year mark, and only 2000-2099 liquidraver/ZephCore#103 is at 535a37c (A0h rule). Pieter accepted its read-modify-write Status clears as a ZephCore platform difference.
  • rtc: select 24-hour mode before a time write, and take no time in 12-hour mode liquidraver/ZephCore#104 is at 7286e50 (12-hour mode).
  • ZephCore's handoff is on its own fork.
• Acceptance matrix: ALL PASS in all three projects, each on its exact submitted code. MeshCore's run was on 0f038054, with G2 done by a real USB power cut.
• Hardware: MeshCore's test board runs the acceptance build. That build is local branch local/hw-accept: 0f038054, plus the throwaway diagnostic CLI 933cae86, plus the harness test-local/rv3028_accept.py (a99d9ad6). Only the throwaway commits differ from 0f038054, and src/ is identical to it. Board state: correct time, test registers restored. A third RAK4631 attached to the host is not ours.
• Worktrees: same as link Session Handoff [rv3028-eeprom]: No upstream movement; cleanup deferred, every PR waits on maintainers #22, plus local/hw-accept checked out in the -hw-3421 worktree.

The parked decision queue

0 parked: the fork has no open decision issue. The decisions made this round were put to Pieter as dialogs and answered in the session.

What the last round did

• Replaced "rule 3" on Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421 with the A0h rule, agreed with the Zephyr and ZephCore sessions. The year is marked A0h, then all seven registers go in one burst. A year of A0h-FEh reads as not set, FFh as a failed read, and year 00 as 2000.
• Answered four Copilot rounds on Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421, each fixed and tested on hardware:
  • The 2000-2099 write range (2100 encoded as A0h).
  • The weekday from dayOfTheWeek() (the old formula was wrong on 83% of days).
  • 12-hour mode, checked and cleared inside the read/write bracket.
  • Write retries every 5 s, at most 13 attempts in all.
• Cross-checks with the other two sessions produced further changes:
  • The Control 2 RESET guard (bit 0 always reads 0, so a torn read is never written back).
  • set24HourMode() dropped from begin().
  • BSF cleared with a direct DFh write.
• Upstream text: the "Cross-project consistency" and "12-hour mode" paragraphs were added to the Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421 description, with notifying comments.
• Freeze: agreed the freeze and the shared acceptance matrix, then ran it to ALL PASS.

What not to repeat

• C3 (seconds 5Ah) can't be tested across a reboot. The chip counts an invalid seconds value on to a valid one within seconds, so write it on a fresh boot and read at once.
• Bit-banged I2C on the RAK4631: a lambda-based version with 5 µs half-periods never got an ACK. Plain functions with about 20 µs half-periods work, and the bus timeout reproduced Zephyr's T18 result.
• A power-cut case is valid only with proof on record. The first cut of this round hit the wrong board. Check that the board's USB port re-enumerated and that Count TS (14h) incremented, with time stamping set up beforehand (TSE plus TSS=1).
• A usbip re-enumeration can drop the serial port mid-run with an EIO. Reopen by the by-id path; it is not a firmware crash.
• Asking Pieter to act in prose fails. "Please re-request Copilot" at the end of a summary was missed. Every action for him goes in a dialog with the URL.
• A bare #98 in cross-repo text links to the host repository's own issue 98. Always use owner/repo#N.

New learnings

• The RV3028 freeze (Pieter's direction, relayed by the Zephyr session): agree, test on hardware, then proceed. Change code only on a failed test or a reviewer's requirement, and run the matrix A-H on the exact submitted commit. It's recorded in this session's project memory. The matrix cases are in the Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421 description's acceptance table.
• Status flag writes: writing 1 to a Status flag leaves it unchanged (tested by two sessions on two RV-3028-C7). A direct write of ~flag is the safe clear. Recorded in the Read and write the RV3028 clock atomically, and reject switchover-garbled transfers 🤖🤖 meshcore-dev/MeshCore#3421 code comments and description.