The research notebook

Evidence before folklore.

A successful transfer is not proof that firmware booted. A call count is not a BLE capture. These pages follow the actual code and keep the remaining questions visible.

Current topics

Current audit

R12 runtime paths

The real A104 producer, the stock memory bridge, and the five-digit display encoder. What changed after rechecking the binary.

Read the R12 findings →
Mixed evidence

Ring families

Why “R02” or “R10” is not enough to choose an image. Compare exact identities, hardware canaries and review-only work.

Compare the families →
Byte-verified

Firmware containers

Standard and nested image layouts, activation metadata, and the difference between a correct checksum and a working device.

Inside the images →

Read the evidence level.

Hardware result

Observed on a physical ring. A short canary is evidence for that device and setup—not a universal compatibility or long-duration stability claim.

Offline finding

Derived from firmware bytes or execution in a bounded emulator. Hardware, radio behaviour and loader acceptance can still differ.

Open question

A proposed test or unresolved explanation. We keep it separate from results, even when the hypothesis is compelling.

Historical record

Useful context that may contain superseded conclusions. Follow the current audit before using an old notebook or candidate.

Reproduce, without a ring.

The R12 page’s packet and timer tables come from a captured run of the strict harness. Image-header checks are recomputed from the checked-in binaries whenever the site is built. No site build contacts hardware.

# From the repository root
python3 fw/sim/test_r12_runtime.py -v
python3 scripts/site/capture.py --check
python3 scripts/site/build.py --check

The execution tests need Unicorn and Capstone; the repository supports a Nix fallback. The site itself needs only Python’s standard library and ships no emulator or executable firmware.

Source notebooks & history

Detailed notebooks remain in the repository. These links preserve the research trail without presenting every old conclusion as current advice.

No page authorizes a flash. Exact identity, an appropriate stock recovery image, battery/readiness checks and an explicit device-specific decision remain necessary. Review-only candidates are not recommendations.