9.0 KiB
Bare-Metal DoE Run Provenance
Master ledger for all 30-replicate DoE runs. Three replications × three architectures = 9 runs total. Each run is itself a complete 2⁴ factorial experiment (16 configs × 30 reps = 480 runs), making this a nested/hierarchical DoE: 9 outer cells × 480 inner runs.
Outer DoE Design — Run Order Matrix
Balanced so each architecture appears in each position exactly once:
| Position | Rep 1 (seed 12345) | Rep 2 (seed 67890) | Rep 3 (seed 13579) |
|---|---|---|---|
| 1st run | amd64 | riscv64 | aarch64 |
| 2nd run | aarch64 | aarch64 | amd64 |
| 3rd run | riscv64 | amd64 | riscv64 |
riscv64 occupies all three positions across reps (3rd, 1st, 2nd) — best achievable balance given the first two orderings.
Inner DoE factors (2⁴ = 16 configurations, shuffled by seed via Fisher-Yates):
| Factor | Low | High |
|---|---|---|
| Entropy thresh | 0.50 | 0.75 |
| CV threshold | 0.10 | 0.15 |
| Temporal decay | 0.30 | 0.50 |
| Stability flag | 0 | 1 |
Config presentation order is shuffled per-run using the DoE seed (doe.4th Block 2054). Same seed → same shuffle (reproducible). Different seeds → independent permutations.
Note on inner DoE CSV extraction: EMIT-ROW output (480-row per-run summaries)
is fragmented by the 10µs heartbeat interrupt firing mid-print. These rows are not
reliably extractable from existing logs. Fix planned: atomic row buffer in doe.4th
before next campaign. Current analysis uses heartbeat stream only.
Replication 1 — Seed 12345 (2026-06-12)
Run order: amd64 → aarch64 → riscv64
| Arch | CSV filename | Rows | Wall clock | Log filename | Status |
|---|---|---|---|---|---|
| amd64 | doe-amd64-20260612-110212.csv | 261,098 | 51m 39s | qemu-amd64-20260612-110212.log | CANONICAL |
| aarch64 | doe-aarch64-20260612-132412.csv | 261,095 | 5m 05s | qemu-aarch64-20260612-132412.log | CANONICAL |
| riscv64 | doe-riscv64-20260612-135612.csv | 261,095 | 5m 16s | qemu-riscv64-20260612-135612.log | CANONICAL |
MD5: aarch64 ≠ riscv64 in heat columns from tick 34,001; all attractor metrics (exec_delta, window_width, time_trust, variance) identical across all three.
Replication 2 — Seed 67890 (2026-06-12)
Run order: riscv64 → aarch64 → amd64
| Arch | CSV filename | Rows | Wall clock | Log filename | Status |
|---|---|---|---|---|---|
| riscv64 | doe-riscv64-20260612-141129.csv | 261,095 | 5m 21s | qemu-riscv64-20260612-141129.log | CANONICAL |
| aarch64 | doe-aarch64-20260612-141722.csv | 261,095 | 5m 10s | qemu-aarch64-20260612-141722.log | CANONICAL |
| amd64 | doe-amd64-20260612-142323.csv | 261,098 | ~51m | qemu-amd64-20260612-142323.log | CANONICAL |
Replication 3 — Seed 13579 (2026-06-12)
Run order: aarch64 → amd64 → riscv64
| Arch | CSV filename | Rows | Wall clock | Log filename | Status |
|---|---|---|---|---|---|
| aarch64 | doe-aarch64-20260612-160132.csv | 261,095 | ~5m | qemu-aarch64-20260612-160132.log | CANONICAL |
| amd64 | doe-amd64-20260612-160656.csv | 261,098 | ~51m | qemu-amd64-20260612-160656.log | CANONICAL |
| riscv64 | doe-riscv64-20260612-174414.csv | 261,095 | ~5m | qemu-riscv64-20260612-174414.log | CANONICAL |
All 9 outer DoE cells complete. Campaign closed 2026-06-12.
Latin Square Seed D — Seed 54321 (2026-06-14)
Run order: amd64 → riscv64 → aarch64
Context: This is the first run using the corrected QEMU termination condition
(-lt 3 instead of -lt 2). Prior to this fix, the REPL command echo
([Hera] zuse)ok> 54321 30 EXEC-DOE) bumped the grep count to 2,
killing QEMU before EXEC-DOE ran. All three arches now produce full 261k-tick runs.
| Arch | CSV filename | Rows | Wall clock | Log filename | Status |
|---|---|---|---|---|---|
| amd64 | doe-amd64-20260614-212929.csv | 261,099 | ~43m | qemu-amd64-20260614-212929.log | CANONICAL |
| riscv64 | doe-riscv64-20260614-230220.csv | 261,096 | ~5m | qemu-riscv64-20260614-230220.log | CANONICAL |
| aarch64 | doe-aarch64-20260614-230803.csv | 261,096 | ~5m | qemu-aarch64-20260614-230803.log | CANONICAL |
All three arches agree at 261,096–261,099 ticks (within 3 ticks of boot-phase variation). APIC timer active on amd64 (apic_ticks advancing); riscv64 and aarch64 show apic_ticks=0 (APIC not wired for those ISAs in current kernel).
Superseded Runs (kept for audit trail, NOT used in analysis)
| Arch | CSV filename | Rows | Date | Reason superseded |
|---|---|---|---|---|
| amd64 | doe-amd64-20260611-012412.csv | 261,097 | 2026-06-11 | Transitional build — EXEC-DOE was trivial alias |
| aarch64 | doe-aarch64-20260611-190822.csv | 261,095 | 2026-06-11 | Superseded by Rep 1 clean re-run |
| riscv64 | doe-riscv64-20260611-191335.csv | 261,095 | 2026-06-11 | Superseded by Rep 1 clean re-run |
The original aarch64 and riscv64 (2026-06-11) were byte-for-byte identical. This was a single-session observation; Rep 1 re-runs show heat variation between arches while attractor metrics remain invariant. See report §7.5.
Development / Partial Runs (not used, not audited)
Files dated 2026-06-09 and 2026-06-10 are development iterations with varying rep counts, incomplete capsule implementations, or aborted builds. Retained for git history, excluded from all analysis.
New 3×3 Campaign — ACL-Corrected Build (2026-06-15)
Identical design to the original 3×3 (same seeds, same run orders, 30 reps), re-run after two correctness fixes:
- ACL.4th and zuse.4th moved to correct block range (4000+)
- QEMU termination condition fixed (
-lt 2→-lt 3) so EXEC-DOE runs to completion
Rep 1 — Seed 12345 (run order: amd64 → aarch64 → riscv64)
| Arch | CSV filename | Rows | Log filename | Status |
|---|---|---|---|---|
| amd64 | doe-amd64-20260614-233623.csv | 261,099 | qemu-amd64-20260614-233623.log | CANONICAL |
| aarch64 | doe-aarch64-20260615-003908.csv | 261,097 | qemu-aarch64-20260615-003908.log | CANONICAL |
| riscv64 | doe-riscv64-20260615-004335.csv | 261,096 | qemu-riscv64-20260615-004335.log | CANONICAL |
Rep 2 — Seed 67890 (run order: riscv64 → aarch64 → amd64)
| Arch | CSV filename | Rows | Log filename | Status |
|---|---|---|---|---|
| riscv64 | doe-riscv64-20260615-004822.csv | 261,096 | qemu-riscv64-20260615-004822.log | CANONICAL |
| aarch64 | doe-aarch64-20260615-005301.csv | 261,097 | qemu-aarch64-20260615-005301.log | CANONICAL |
| amd64 | doe-amd64-20260615-005725.csv | 261,099 | qemu-amd64-20260615-005725.log | CANONICAL |
Rep 3 — Seed 13579 (run order: aarch64 → amd64 → riscv64)
| Arch | CSV filename | Rows | Log filename | Status |
|---|---|---|---|---|
| aarch64 | doe-aarch64-20260615-020020.csv | 261,097 | qemu-aarch64-20260615-020020.log | CANONICAL |
| amd64 | doe-amd64-20260615-020446.csv | 261,099 | qemu-amd64-20260615-020446.log | CANONICAL |
| riscv64 | doe-riscv64-20260615-031059.csv | 261,096 | qemu-riscv64-20260615-031059.log | CANONICAL |
All 9 cells complete. New 3×3 campaign closed 2026-06-15.
latest/ Contents (current canonical data)
Points to new 3×3 Rep 3 (most recent complete replication, seed 13579, ACL-corrected build).
| File | Points to |
|---|---|
latest/amd64.csv |
runs/doe-amd64-20260615-020446.csv (new 3×3 Rep 3) |
latest/aarch64.csv |
runs/doe-aarch64-20260615-020020.csv (new 3×3 Rep 3) |
latest/riscv64.csv |
runs/doe-riscv64-20260615-031059.csv (new 3×3 Rep 3) |
All three files represent new 3×3 Rep 3 (seed 13579, ACL-corrected build, 2026-06-15).