Runtime timing integration validation — November 14–15, 2026

Executed on native aarch64 Linux in an Ubuntu 22.04 Docker container:
  Linux 6.12.76-linuxkit; GCC 11.4.0; GDB 12.1; CMake 3.22.1.
  python3 scripts/verify-runtime-lab.py --require-linux --gdb --cmake
  (The local container run selected /evidence as its output directory.)

Passed:
  - C++11 build with -Wall -Wextra -Wpedantic; no compiler diagnostics.
  - Defined, unmangled perform_work and run_cycle text symbols; ELF notes saved.
  - Sleep: 20 invocations, 100 records, 19 same-TID entry intervals.
  - Busy: 20 invocations, 100 records, 19 same-TID entry intervals.
  - Period overrun: 5 invocations, 25 records, 4 entry intervals.
  - 14 invalid command/output cases, including /dev/full write failure.
  - Known synthetic durations, including a 0.2 ms caller-to-callee transition.
  - Interleaved different-TID input without cross-thread interval pairing.
  - 14 malformed, missing, duplicated or reordered trace cases rejected.
  - Alternative CMake build and two-invocation run.
  - Published GDB session: perform_work -> run_cycle -> main; normal exit;
    all three invocations complete. Debugger timings are not baseline timings.

Durations are retained in each run's report and CSVs. These are container
observations, not acceptance thresholds or historical application results.
The two clocks are sampled consecutively; small CPU/elapsed inversions are
possible. Repetition under different load can change the numbers.

Local evidence: output/runtime-validation-linux/report.json plus CSVs and logs.
The report records input hashes, binary hash, versions and exact commands.
CI runs the verifier on Ubuntu with --require-linux --gdb --cmake; its results
are future evidence, not claimed as already observed on an x86-64 hosted runner.

Site checks:
  npm run check, with both /execmap and empty SITE_BASEURL.
  Existing browser suite, with both prefixes: 13 passed, 1 intentional skip
  per run (viewport sweep is already covered in the desktop project).
  Separate browser inspection of the new lab and guide at mobile width;
  download links retain /execmap, and lab downloads match tested source bytes.

Not executed:
  perf stat/record/sched, BPF compile/attach, remote GDB, sanitizers,
  CPU affinity/policy/frequency changes, or lockup reproduction.
The BPF examples target bpftrace 0.21 syntax and need separate kernel validation.
Archive CSVs, private project history and old sanitizer claims were not imported.