# The Bach run's own log — pass lines and step lines, quoted

**This is not the log file.** The Bach trainer's log lives on the training box at
`music/out/bach-shakedown/20260830T104030Z-bach0/train.log`, and that box was not
reachable from the lane that assembled this kit. What is below is the log's own text,
**quoted verbatim** from the arc's receipts pack — `MATERIALS.md` §1.6, "Bach epoch
lines, VERBATIM", which cites `train.log:115-141` — and the article's Bach numbers were
read from exactly this text.

The receipts pack it was quoted from:
`estate/bench/ace-composers-2026-08/MATERIALS.md`, built 2026-08-31 (UTC),
sha256 `9cb722fb315995de1a385e2f9682980fe759403915fd5e6c323da275ce2fe614`.

Per-pass and per-step figures both appear: `Epoch n/10, Step m, Loss:` is a step line;
`[OK] Epoch n/10 in Ns, Loss:` is the pass mean the article's table quotes.

## The pass and step lines (`train.log:115-141`)

```
Epoch 1/10, Step 10, Loss: 0.9242
[OK] Epoch 1/10 in 49.9s, Loss: 0.9266
Epoch 2/10, Step 20, Loss: 0.6672
Epoch 2/10, Step 30, Loss: 0.6450
[OK] Epoch 2/10 in 49.5s, Loss: 0.6822
Epoch 3/10, Step 40, Loss: 0.6313
[OK] Epoch 3/10 in 49.9s, Loss: 0.6513
Epoch 4/10, Step 50, Loss: 0.6804
Epoch 4/10, Step 60, Loss: 0.6305
[OK] Epoch 4/10 in 50.2s, Loss: 0.6139
Epoch 5/10, Step 70, Loss: 0.6515
Epoch 5/10, Step 80, Loss: 0.5471
[OK] Epoch 5/10 in 50.2s, Loss: 0.6254
Epoch 6/10, Step 90, Loss: 0.5903
[OK] Epoch 6/10 in 50.3s, Loss: 0.6080
Epoch 7/10, Step 100, Loss: 0.5717
Epoch 7/10, Step 110, Loss: 0.6811
[OK] Epoch 7/10 in 50.4s, Loss: 0.5967
Epoch 8/10, Step 120, Loss: 0.6011
[OK] Epoch 8/10 in 50.3s, Loss: 0.5859
Epoch 9/10, Step 130, Loss: 0.6695
Epoch 9/10, Step 140, Loss: 0.5780
[OK] Epoch 9/10 in 50.3s, Loss: 0.6145
Epoch 10/10, Step 150, Loss: 0.6544
Epoch 10/10, Step 160, Loss: 0.5891
[OK] Epoch 10/10 in 50.3s, Loss: 0.5749
```

## What the receipts pack records around these lines

- **Wall clock, the trainer's own clock: 8m 23s.** The run doc's timestamp arithmetic
  gives 8m 30s for the same run — the article quotes the trainer's clock and names the
  gap. Both readings are on the record; neither was picked over the other silently.
- **63 training files, 16 held back by piece, 10 passes, 160 optimizer steps,
  peak VRAM 5.7 GiB.**
- **The summary block's `Loss 0.9242 -> 0.6139` is step loss**, not the pass means
  (0.9266 → 0.5749) the article's table quotes.