# S5 release and hardware record

On September 17, 2026, after the S5 SD-card handoff, the owner requested:

> push this one as a release
> to git

This authorizes tag **setup-s5-20260917**, pairing FPGA
**PI FMG260917S550Mhz** with ARM **MM2609S5**. Promotion uses the exact
binaries copied to the card, without rebuilding either component.

## SD-card handoff

At **2026-09-17 14:06:11 UTC**, the two update files and S4/S3 rollback pair
were written to the SD card, SHA-256 readback verified, flushed and safely
unmounted. The prior core, ARM update and settings were backed up locally.
Saved configuration was preserved; disk-image sizes and modification times
were unchanged. Copying the ARM upgrade file does not confirm that the owner
has subsequently flashed it on the board.

FPGA BIN SHA-256:
`2343b67d8cad5a49f08125d339ccc576b5f1909963482ea5dbf5e45c733fd9b1`.

ARM UPG SHA-256:
`cf9d0949eacf810d34c7d66c60acf164702361aaff30bb78916a81b7d1a1afd5`.

## Validation and follow-up

The selected F/8 route passes main, Normal and component timing with zero
errors and all seven independent route audits. FPGA functional regressions,
additional CPU/SRAM/retirement tests at 60 MHz, and ARM host/build checks
pass. The exact source and reports are preserved in the release archive.

The promotion request does not include a separate S5 hardware test result.
It therefore does not establish that the previously reported intermittent
50 MHz instability is fixed. Repeat the affected workload at 50 MHz,
then check optional 60 MHz operation, cold boots, Normal mode and OSD
navigation. Earlier TF030/68000 release results remain historical evidence;
no new accelerator test is claimed by this promotion.

Preserve the owner's enabled slow-RAM-priority command in User-Startup and
the same memory settings for comparable SysInfo readings. The earlier low
SysInfo readings caused by uncached fast RAM remain a separate closed issue.

The original test receipt and source archive retain their pre-promotion
status. Current release notes record the owner's publication decision while
preserving the original evidence unchanged.

## Subsequent instability report

Later on September 17, 2026, the owner reported:

> I can tell that after a while 40, 50 ,60 MHz crash. See if you can improve timing and reliability?

This report followed the S5 release. The exact time to failure, workload,
failure signature and comparison at 20/30 MHz are still unknown. The earlier
68SEC000 configuration is the working assumption, pending confirmation.
S6 addresses two narrow internal timing paths and is a test candidate for
this report; no hardware fix has yet been demonstrated. See
[the S6 changes and validation](setup-s6.txt). The released S5 binaries,
checksums and original evidence remain unchanged.
