Phase 4: dependency-free ASCII loss-curve plotter

- scripts/plot_loss.py: reads reports/loss-<config>.csv, renders train/val
  loss as a terminal ASCII chart, saves it to reports/loss-<config>.txt.
  Verified on a synthetic curve. Keeps deps at torch + numpy.
- README: documented plotting choice and the Phase 4 run commands.

README.md

@@ -109,9 +109,19 @@ raw text ──► tokenizer training ──► tokenized corpus ──► pretr
 - Runs on MPS; expected wall-clock for v1 is a few hours for ~100–200M tokens
   (roughly Chinchilla-optimal for this size — we don't need to see the whole
   corpus).
-- Loss curve logged to a plain CSV and plotted locally.
-- Deliverables: `src/train.py`, `configs/v1.yaml` (or `.py`), checkpoints in
-  `checkpoints/` (gitignored), loss curves in `reports/`.
+- Loss curve logged to a plain CSV (`reports/loss-v1.csv`) and rendered as an
+  ASCII chart by `scripts/plot_loss.py` — no plotting dependency; the chart is
+  also saved to `reports/loss-v1.txt` as a committable artifact.
+- Deliverables: `src/train.py`, `configs/v1.py`, `scripts/plot_loss.py`,
+  checkpoints in `checkpoints/` (gitignored), loss log + chart in `reports/`.
+
+Commands:
+
+```sh
+python src/train.py --config v1 --overfit   # sanity check (loss -> ~0)
+python src/train.py --config v1             # real run (resumable: --resume)
+python scripts/plot_loss.py --config v1     # ASCII loss curve
+```
 
 ### Phase 5 — Sampling & evaluation
 - Autoregressive sampler with temperature and top-k.

scripts/plot_loss.py

@@ -0,0 +1,90 @@
+"""Render the training loss curve as an ASCII chart — no plotting dependency.
+
+Reads reports/loss-<config>.csv (written by src/train.py), prints a terminal
+chart of train and val loss, and saves the same rendering to
+reports/loss-<config>.txt as a committable artifact.
+
+Usage:
+    python scripts/plot_loss.py [--config v1]
+"""
+
+import argparse
+import csv
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parent.parent
+
+TRAIN_CH = "."
+VAL_CH = "*"
+
+
+def read_csv(path: Path):
+    iters, train, val = [], [], []
+    with open(path, newline="") as f:
+        for row in csv.DictReader(f):
+            iters.append(int(row["iter"]))
+            train.append(float(row["train_loss"]))
+            val.append(float(row["val_loss"]))
+    return iters, train, val
+
+
+def render(iters, series, width=64, height=18) -> str:
+    """series: list of (name, values, char). Returns a multi-line chart string."""
+    all_vals = [v for _, vals, _ in series for v in vals]
+    lo, hi = min(all_vals), max(all_vals)
+    if hi == lo:
+        hi = lo + 1.0
+    imin, imax = min(iters), max(iters)
+    if imax == imin:
+        imax = imin + 1
+
+    def col(it: int) -> int:
+        return round((it - imin) / (imax - imin) * (width - 1))
+
+    def row(v: float) -> int:
+        return round((hi - v) / (hi - lo) * (height - 1))  # row 0 = top = hi
+
+    grid = [[" "] * width for _ in range(height)]
+    for _, vals, ch in series:
+        for it, v in zip(iters, vals):
+            grid[row(v)][col(it)] = ch
+
+    lines = []
+    for r in range(height):
+        yval = hi - r / (height - 1) * (hi - lo)
+        lines.append(f"{yval:6.3f} |" + "".join(grid[r]))
+    lines.append(" " * 7 + "+" + "-" * width)
+    axis = f"{imin:<{width // 2}}{imax:>{width - width // 2}}"
+    lines.append(" " * 8 + axis)
+    legend = f"  legend: '{TRAIN_CH}' train   '{VAL_CH}' val   (x: iter, y: loss)"
+    lines.append(legend)
+    return "\n".join(lines)
+
+
+def main() -> None:
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--config", default="v1")
+    args = ap.parse_args()
+
+    csv_path = ROOT / "reports" / f"loss-{args.config}.csv"
+    if not csv_path.exists():
+        sys.exit(f"no loss log at {csv_path} — run training first")
+
+    iters, train, val = read_csv(csv_path)
+    if len(iters) < 2:
+        sys.exit(f"only {len(iters)} data point(s) logged so far — need at least 2")
+
+    chart = render(iters, [("train", train, TRAIN_CH), ("val", val, VAL_CH)])
+    header = (f"loss curve — {args.config}  "
+              f"(final train {train[-1]:.4f}, val {val[-1]:.4f} @ iter {iters[-1]})")
+    out = header + "\n" + chart
+
+    print(out)
+    txt_path = ROOT / "reports" / f"loss-{args.config}.txt"
+    txt_path.write_text(out + "\n")
+    print(f"\nsaved {txt_path}")
+
+
+if __name__ == "__main__":
+    main()