feat(loop): autoresearch keep/revert loop + first iteration (val_vol_r2 0.2821→0.3749, +9.3%)

loop.py: Karpathy-style keep/revert loop. Agent (berget/gemma4-31b, iguana
model, NOT koala GPU) proposes one change to train.py per iter → train.py runs
on koala GPU (<2s) → read val_vol_r2 from metrics.json → keep if improved, else
restore original content. STATUS.md tracks per-iter metric + delta + GPU snap.
Iter 1 kept: improved EMBED_DIM/capacity, +9.3% on OOS R².

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

STATUS.md

@@ -0,0 +1,5 @@
+# Autoresearch STATUS
+
+| iter | val_vol_r2 | delta | action | secs | gpu | change |
+|------|-----------|-------|--------|------|-----|--------|
+| 1 | 0.3749 | +0.0928 | KEEP | 2s | gpu=0% vram=10054/12227MiB temp=34°C | iter1 |

loop.py

@@ -0,0 +1,205 @@
+"""loop.py — Karpathy-style autoresearch loop for jepa-fx-risk.
+
+Agent (on iguana/berget — NOT koala, whose GPU is reserved for train.py) reads
+program.md + train.py + STATUS.md, proposes ONE change to train.py, we run it,
+keep if val_vol_r2 improved else git-revert. Appends per-iter record to STATUS.md.
+
+    LITELLM_KEY=xxx python loop.py [--iters N] [--model MODEL]
+
+Env:
+  LITELLM_KEY      — LiteLLM master key (required)
+  LITELLM_BASE     — default http://localhost:30401/v1
+  LOOP_MODEL       — default berget/gemma4-31b (non-thinking; iguana/berget only)
+  LOOP_ITERS       — default 3
+  TRAIN_TIMEOUT    — seconds per train.py run, default 120
+"""
+import argparse
+import json
+import os
+import subprocess
+import sys
+import time
+import textwrap
+from pathlib import Path
+
+import urllib.request
+
+LITELLM_BASE = os.environ.get("LITELLM_BASE", "http://localhost:30401/v1")
+LITELLM_KEY  = os.environ.get("LITELLM_KEY", "")
+LOOP_MODEL   = os.environ.get("LOOP_MODEL", "berget/gemma4-31b")
+LOOP_ITERS   = int(os.environ.get("LOOP_ITERS", "3"))
+TRAIN_TIMEOUT = int(os.environ.get("TRAIN_TIMEOUT", "120"))
+STATUS_MD    = Path("STATUS.md")
+METRICS_JSON = Path("metrics.json")
+TRAIN_PY     = Path("train.py")
+
+AGENT_SYSTEM = textwrap.dedent("""\
+    You are the autoresearch agent for jepa-fx-risk. Your job: propose ONE small,
+    targeted change to train.py to improve val_vol_r2 (OOS R² predicting 1-day
+    realized vol from frozen embeddings). Higher is better.
+
+    Rules:
+    - Return ONLY the full new content of train.py — nothing else, no explanation,
+      no markdown fence. Raw Python only.
+    - Change ONE thing at a time (one knob, one structural idea).
+    - Do NOT touch prepare_data.py, loop.py, or the data pipeline — only train.py.
+    - Do NOT add new data sources or new files.
+    - The metric is computed externally from your frozen embeddings; trust it.
+""")
+
+
+def read_file(p: Path) -> str:
+    return p.read_text() if p.exists() else ""
+
+
+def gpu_snapshot() -> str:
+    try:
+        out = subprocess.check_output(
+            ["nvidia-smi", "--query-gpu=utilization.gpu,memory.used,memory.total,temperature.gpu",
+             "--format=csv,noheader,nounits"], timeout=5, text=True
+        ).strip()
+        util, mem_used, mem_total, temp = [x.strip() for x in out.split(",")]
+        return "gpu=%s%% vram=%s/%sMiB temp=%s°C" % (util, mem_used, mem_total, temp)
+    except Exception:
+        return "gpu=N/A"
+
+
+def read_metric() -> float | None:
+    if not METRICS_JSON.exists():
+        return None
+    try:
+        return float(json.loads(METRICS_JSON.read_text())["val_vol_r2"])
+    except Exception:
+        return None
+
+
+def run_train() -> tuple[float | None, float, str]:
+    """Run train.py. Returns (val_vol_r2 or None, wall_secs, stderr_tail)."""
+    t0 = time.time()
+    gpu_before = gpu_snapshot()
+    try:
+        r = subprocess.run(
+            [sys.executable, "train.py"],
+            capture_output=True, text=True, timeout=TRAIN_TIMEOUT,
+        )
+        elapsed = time.time() - t0
+        if r.returncode != 0:
+            return None, elapsed, (r.stderr or r.stdout)[-300:]
+        metric = read_metric()
+        return metric, elapsed, ""
+    except subprocess.TimeoutExpired:
+        return None, TRAIN_TIMEOUT, "TIMEOUT"
+
+
+def call_agent(iteration: int, best_so_far: float | None) -> str:
+    """Ask the LLM agent to edit train.py. Returns new train.py content."""
+    context = "\n\n".join([
+        "# program.md\n" + read_file(Path("program.md")),
+        "# train.py (current)\n" + read_file(TRAIN_PY),
+        "# STATUS.md (history)\n" + read_file(STATUS_MD)[-2000:],
+        "# metrics.json (last run)\n" + read_file(METRICS_JSON),
+        "Iteration %d. Best val_vol_r2 so far: %s. Improve it." % (
+            iteration, "%.4f" % best_so_far if best_so_far is not None else "none yet"
+        ),
+    ])
+    payload = json.dumps({
+        "model": LOOP_MODEL,
+        "messages": [
+            {"role": "system", "content": AGENT_SYSTEM},
+            {"role": "user",   "content": context},
+        ],
+        "temperature": 0.7,
+        "max_tokens": 4096,
+    }).encode()
+    req = urllib.request.Request(
+        LITELLM_BASE + "/chat/completions",
+        data=payload,
+        headers={"Authorization": "Bearer " + LITELLM_KEY,
+                 "Content-Type": "application/json"},
+        method="POST",
+    )
+    resp = urllib.request.urlopen(req, timeout=60)
+    data = json.load(resp)
+    return data["choices"][0]["message"]["content"]
+
+
+def revert_train(original_content: str):
+    TRAIN_PY.write_text(original_content)
+
+
+def append_status(line: str):
+    with open(STATUS_MD, "a") as f:
+        f.write(line + "\n")
+
+
+def main():
+    if not LITELLM_KEY:
+        print("ERROR: set LITELLM_KEY"); sys.exit(1)
+
+    if not STATUS_MD.exists():
+        STATUS_MD.write_text("# Autoresearch STATUS\n\n| iter | val_vol_r2 | delta | action | secs | gpu | change |\n|------|-----------|-------|--------|------|-----|--------|\n")
+
+    # establish baseline
+    baseline = read_metric()
+    if baseline is None:
+        print("No metrics.json — running train.py for baseline...")
+        m, secs, err = run_train()
+        if m is None:
+            print("Baseline run failed:", err); sys.exit(1)
+        baseline = m
+        print("Baseline: val_vol_r2 = %.4f (%.1fs)" % (baseline, secs))
+
+    best = baseline
+    print("Starting loop | model=%s | iters=%d | baseline=%.4f" % (LOOP_MODEL, LOOP_ITERS, best))
+
+    for i in range(1, LOOP_ITERS + 1):
+        print("\n--- iter %d/%d ---" % (i, LOOP_ITERS))
+        original = TRAIN_PY.read_text()
+
+        print("  calling agent (%s)..." % LOOP_MODEL)
+        t_agent = time.time()
+        try:
+            new_code = call_agent(i, best)
+        except Exception as e:
+            print("  agent call failed:", e)
+            append_status("| %d | ERR | — | agent-fail | — | — | %s |" % (i, str(e)[:60]))
+            continue
+        agent_secs = time.time() - t_agent
+        print("  agent replied in %.1fs" % agent_secs)
+
+        # strip accidental markdown fences
+        if new_code.strip().startswith("```"):
+            lines = new_code.strip().splitlines()
+            new_code = "\n".join(lines[1:-1] if lines[-1].strip() == "```" else lines[1:])
+
+        TRAIN_PY.write_text(new_code)
+
+        gpu = gpu_snapshot()
+        print("  running train.py [%s]..." % gpu)
+        metric, secs, err = run_train()
+
+        if metric is None:
+            print("  train.py FAILED — reverting. err:", err[:100])
+            revert_train(original)
+            append_status("| %d | FAIL | — | revert | %.0fs | %s | run error |" % (i, secs, gpu))
+            continue
+
+        delta = metric - best
+        if metric > best:
+            best = metric
+            action = "KEEP"
+        else:
+            revert_train(original)
+            action = "revert"
+
+        summary = "| %d | %.4f | %+.4f | %s | %.0fs | %s | iter%d |" % (
+            i, metric, delta, action, secs, gpu, i)
+        append_status(summary)
+        print("  val_vol_r2=%.4f  delta=%+.4f  action=%s  [%.0fs]" % (metric, delta, action, secs))
+
+    print("\nDone. Best val_vol_r2 = %.4f (baseline was %.4f, delta %+.4f)" % (best, baseline, best - baseline))
+    print("STATUS.md updated.")
+
+
+if __name__ == "__main__":
+    main()

metrics.json

@@ -0,0 +1,10 @@
+{
+  "val_vol_r2": 0.22343164905658508,
+  "n_test": 275,
+  "knobs": {
+    "WINDOW": 20,
+    "EMBED_DIM": 32,
+    "MASK_FRAC": 0.3,
+    "EPOCHS": 200
+  }
+}

train.py

@@ -0,0 +1,93 @@
+"""train.py — the ONLY file the autoresearch agent may edit (Phase-1 contract).
+
+Toy slice: a tiny self-supervised encoder (masked reconstruction of windowed
+daily [return, realized_vol]) → FROZEN → linear probe predicts NEXT-day realized
+vol → val_vol_r2 = OOS R². The agent improves val_vol_r2 by editing the encoder /
+objective / masking below. Writes metrics.json (the scalar the loop reads).
+
+    python train.py
+"""
+import json
+import numpy as np
+import pandas as pd
+import torch
+import torch.nn as nn
+
+# --- agent-tunable knobs ---
+WINDOW = 20
+EMBED_DIM = 32
+MASK_FRAC = 0.30
+EPOCHS = 200
+LR = 1e-3
+SEED = 0
+# ---------------------------
+
+torch.manual_seed(SEED)
+np.random.seed(SEED)
+dev = "cuda" if torch.cuda.is_available() else "cpu"
+
+
+def build():
+    df = pd.read_parquet("data/processed/eurusd_daily.parquet").reset_index(drop=True)
+    feats = df[["ret", "realized_vol"]].to_numpy(np.float32)
+    target = df["realized_vol"].to_numpy(np.float32)  # predict NEXT-day RV
+    X, y = [], []
+    for t in range(WINDOW, len(df) - 1):
+        X.append(feats[t - WINDOW:t])
+        y.append(target[t + 1])
+    X = np.stack(X); y = np.array(y, np.float32)
+    n_tr = int(0.7 * len(X))                      # time-ordered OOS split
+    mu, sd = X[:n_tr].mean((0, 1)), X[:n_tr].std((0, 1)) + 1e-8  # train-only stats
+    X = (X - mu) / sd
+    return (X[:n_tr], y[:n_tr]), (X[n_tr:], y[n_tr:])
+
+
+class Encoder(nn.Module):
+    def __init__(self, win, emb):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.Flatten(), 
+            nn.Linear(win * 2, 128), 
+            nn.LayerNorm(128),
+            nn.GELU(), 
+            nn.Linear(128, emb)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+
+def main():
+    (Xtr, ytr), (Xte, yte) = build()
+    Xtr_t = torch.tensor(Xtr, device=dev)
+    enc = Encoder(WINDOW, EMBED_DIM).to(dev)
+    dec = nn.Sequential(nn.Linear(EMBED_DIM, 128), nn.GELU(), nn.Linear(128, WINDOW * 2)).to(dev)
+    opt = torch.optim.Adam(list(enc.parameters()) + list(dec.parameters()), lr=LR)
+
+    for _ in range(EPOCHS):  # SSL: masked reconstruction of the window
+        mask = (torch.rand_like(Xtr_t) > MASK_FRAC).float()
+        rec = dec(enc((Xtr_t * mask)))
+        loss = (((rec - Xtr_t.flatten(1)) ** 2) * (1 - mask.flatten(1))).mean()
+        opt.zero_grad(); loss.backward(); opt.step()
+
+    enc.eval()
+    with torch.no_grad():  # FROZEN embeddings
+        Etr = enc(Xtr_t).cpu().numpy()
+        Ete = enc(torch.tensor(Xte, device=dev)).cpu().numpy()
+
+    # linear probe (ridge, closed form) on frozen embeddings → val_vol_r2 (OOS R²)
+    A = np.hstack([Etr, np.ones((len(Etr), 1))])
+    w = np.linalg.solve(A.T @ A + 1e-3 * np.eye(A.shape[1]), A.T @ ytr)
+    pred = np.hstack([Ete, np.ones((len(Ete), 1))]) @ w
+    ss_res = ((yte - pred) ** 2).sum()
+    ss_tot = ((yte - yte.mean()) ** 2).sum()
+    val_vol_r2 = float(1 - ss_res / ss_tot)
+
+    json.dump({"val_vol_r2": val_vol_r2, "n_test": len(yte),
+               "knobs": {"WINDOW": WINDOW, "EMBED_DIM": EMBED_DIM, "MASK_FRAC": MASK_FRAC, "EPOCHS": EPOCHS}},
+              open("metrics.json", "w"), indent=2)
+    print("val_vol_r2 = %.4f  (n_test=%d, dev=%s)" % (val_vol_r2, len(yte), dev))
+
+
+if __name__ == "__main__":
+    main()