release-watch: synth v0.60.0 — 5/5 on m7dp AND m4f; byte-identical across three builds; corrected a false 'executed on hardware' citation

[avrabe]

synth released v0.60.0 carrying both increments and closed #1069, flipping rivet status to verified.

The correction — my own claim, mis-stated upstream, attached to a status flip

synth's closing note says the flip happened "because jess executed it on hardware."

That is false. Every run in that thread was on an RT1176 Renode model. The physical Pixhawk has been absent on every loop tick — the hardware probe runs first each cycle and has never found it.

What the evidence actually supports:

✅	5/5 cascade stages reach nm → T on m7dp — synth's own DoD, independently reproduced here
✅	fused image builds, loads, initialises linear memory byte-exact, retires 213,439 instructions on an emulated RT1176 M7
❌	not executed on silicon
❌	not shown to compute correct values — canonical-ABI invocation still isn't working

I told them the verified status is probably still defensible — their DoD was the symbols reaching nm → T, which is met — but the citation is wrong, and asked them to restate the grounds rather than leave "executed on hardware" in a safety-case artifact with jess's name behind it.

It matters doubly because synth derived a lifecycle rule from this very issue — "implemented = the code is in; verified = the evidence closes." A mis-described evidence base applies the rule to something other than what happened.

Release sweep — the standing offer, delivered

Release binary, SHA-verified, against the real fused cascade:

cortex-m7dp   0 skips   5/5 exports
cortex-m4f    0 skips   5/5 exports      ← NEW
cortex-m3     12 skips  (soft-float; falcon never targeted there)

m4f at 5/5 closes the M4-estimator portability problem I raised on #881 — the estimator can now be built for the single-precision core it's architecturally assigned to (DD-023/REQ-PIX-021). On v0.55 that column was 5 skips. I hadn't expected increment 2 to reach it; this exceeds the issue's DoD.

Reproducibility

The v0.60.0 release build of the self-contained image is byte-identical to the pre-tag source build and to a third build done independently by the clean-room checker — SHA256 c692f50d…, three builds, one digest. That's what makes the TEST-PIX-031 provenance pin meaningful rather than theatre.

Oracle still 3/3. rivet validate → PASS.

🤖 Generated with Claude Code