Physical worker-entry and first-post-sleep comparison

The new temporary boot delivered both fixed breadcrumbs, a matching fresh Bash receipt, and numeric samples 0 and 1. No samples 2–5 or normal return were observed during the 180.084-second passive capture. This is a useful diagnostic result, not successful ordinary mainline boot or recovery acceptance.

Root alone operated the board/UART from worktree /home/jadams/tmp/k230-mainline-probe-integration, branch integrate/mainline-probe-path, controller revision 90b1e288. The full matching build used frozen 2a548135; reviewed source and controller code were unchanged by the intervening host-evidence commit. The public result records exact artifacts, command return 2, UTC timestamps and private capture SHA256/size. Raw UART, nonce, boot IDs, baseline and transfer URL remain private. Camera/glass evidence was not obtained.

The exact/full host proof and actual positive preparation passed before hardware access. Transfer/import staged eight new paths with return 0; the complete closure is GC-rooted on the board. The controller freshly checked protected normal identities, boot hashes and services, rebooted into U-Boot, verified each load’s size/CRC and exact volatile arguments, then booted the selected matching Image/initrd/DT. Persistent normal boot selection was unchanged.

Operator command:

python3 tools/mainline-drm-initrd-shell-trial.py \
  --mode minimal --same-image-shell-pid1 --uart-progress \
  --uart-progress-breadcrumbs \
  --bundle /nix/store/mhq10143lsmgr6wlll431a8s3ggb8q6m-k230-mainline-drm-trial-boot-files \
  --manifest "$PRIVATE_RUN/candidate-manifest.json" \
  --normal-report "$PRIVATE_RUN/normal-report.json" \
  --log "$PRIVATE_RUN/trial.private.log" \
  --result "$PRIVATE_RUN/result.private.json"

Observations and limits

The fresh 7.3.0-rc5 banner, exact received arguments, /bin/sh entry and primary Bash prompt qualified one receipt stimulus. The actual parser observed its matching response. A provisional raw-line monitor missed that response because the real terminal prefixes it with bracketed-paste disable and CR; the completed controller result, which handles that framing, supersedes the provisional monitor. Exactly one stimulus was attempted. No retry, proc command, guard or candidate reboot followed. No parser protocol error was recorded.

Fixed records preserve only the four constant-schema lines. Worker-entry arrived before the stimulus; first-post-sleep and samples 0/1 arrived after it. This is host arrival order, not causal hardware receive timing. Entry proves the worker reached its first SBI write. Post-sleep proves that entry call returned and the first sleep completed. Sample 1 proves progression beyond the post-sleep and sample 0 writes, and a second sleep. A visible sample 1 does not prove its own SBI call returned. Missing sample 2 leaves return from that call, subsequent sleep/scheduling and later output unresolved.

Two-sample deltas show monotonic time advancing by 5.116997479 seconds, jiffies by 1280 and timer IRQ count by 252. This establishes timer/scheduling progress over that interval; it does not establish future progress or overall timer/interrupt health. Both snapshots report status 0, UART IRQ 12, timer IRQ 10, UART IRQ count 30, RX 89 and TX 292. Cached UART accounting is unchanged between the two samples, as expected without further input. The 89-byte receipt command and matching Bash receipt establish command execution, but the counters have no pre-stimulus baseline or per-byte identity.

In the selected k5a5… source, 8250_core.c’s snapshot validates UART0 binding and copies cached fields under a try-lock; it does not read live IIR/LSR/IER. 8250_port.c updates RX accounting before error-mask/SysRq/flip insertion. UART TX accounting does not count direct-SBI output. IRQ aggregates count mapped Linux interrupts, not successful TTY delivery. These source facts constrain interpretation; none identifies why the previous reporter boot was silent or why this one stopped producing records after sample 1.

Recovery and next boundary

The controller sent no candidate reboot. Normal return was not observed, so a new operator reset was requested only after capture completed. The user reset then passed guarded fresh normal postflight: distinct boot identity, exact system/profile/kernel/init, eight boot hashes, three active services and registration absence. Home IPC returned zero; a separate reviewed camera observation showed normal Home, with angle/focus limits. Operator reset receipt. Automatic return, ordinary mainline /init, usable root, panel/glass and task 5b.5 remain UNVERIFIED.

The next source audit must distinguish the sample 1 SBI return from the following sleep boundary. Repeating the same capture or treating the earlier missing receipt as current RX failure would not be grounded in this result. Any added after-write marker or changed input policy needs a separately reviewed typed comparison, exact artifact proof and fresh protected normal recovery.

The evidence revision 278769d0 passed CI run 37171186624 and deployed. Both published work and physical evidence pages returned HTTP 200 with that exact revision; deployment receipt. Independent review matched all public facts/deltas and capture digest to the protected actual result. Publication does not change the separate manual recovery or ordinary-root/glass gates.