cleanup: remove superseded experiment scripts, notebooks, research lab; untrack local tool config

Deployed path unaffected: script.py -> train/planner/submit closure and
best.pickle module deps (models_cv, planner) verified importable after
removal; submit.py's data/train_features.parquet kept. History retains
all removed files (exp_x28 adjudication etc. referenced from comments).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.claude/settings.json

@@ -1,34 +0,0 @@
-{
-  "permissions": {
-    "allow": [
-      "Bash(kill 48443 48565)",
-      "Bash(PYTHONPATH=/Users/THIF/repos/BatterySwapAI2026-Example ./venv/bin/python /tmp/ag2_analyze.py /tmp/ag2_xrank_s89ab.csv /tmp/ag2_xrank_scdef.csv)",
-      "Bash(PYTHONPATH=/Users/THIF/repos/BatterySwapAI2026-Example ./venv/bin/python /tmp/ag2_analyze.py /tmp/ag2_xrank_s0123.csv /tmp/ag2_xrank_s4567.csv /tmp/ag2_xrank_s89ab.csv /tmp/ag2_xrank_scdef.csv)",
-      "Bash(PYTHONPATH=/Users/THIF/repos/BatterySwapAI2026-Example ./venv/bin/python /tmp/ag2_folds.py /tmp/ag2_xrank_s0123.csv /tmp/ag2_xrank_s4567.csv /tmp/ag2_xrank_s89ab.csv /tmp/ag2_xrank_scdef.csv)",
-      "Bash(PYTHONPATH=__TRACKED_VAR__ nohup ./venv/bin/python /tmp/all48_opt.py)",
-      "Bash(./ag2_mk.sh)",
-      "Bash(PYTHONPATH=/Users/THIF/repos/BatterySwapAI2026-Example nohup ./venv/bin/python /tmp/ag2_xrank2.py 0,1,2,3,4,5,6,7 48 g07)",
-      "Bash(PYTHONPATH=/Users/THIF/repos/BatterySwapAI2026-Example nohup ./venv/bin/python /tmp/ag2_xrank2.py 8,9,10,11,12,13,14,15 48 g815)",
-      "Bash(PYTHONPATH=__TRACKED_VAR__ nohup ./venv/bin/python /tmp/all48.py shape)",
-      "Bash(pkill -f 'xrank2.py 0,1\" ')",
-      "Bash(pkill -f 'until \\\\[ -s /tmp/ag2_xrank2')",
-      "Bash(PYTHONPATH=/Users/THIF/repos/BatterySwapAI2026-Example ./venv/bin/python /tmp/ag2_analyze.py /tmp/ag2_xrank2_g07.csv /tmp/ag2_xrank2_g815.csv)",
-      "Bash(PYTHONPATH=/Users/THIF/repos/BatterySwapAI2026-Example ./venv/bin/python /tmp/ag2_analyze.py /tmp/ag2_xrank_g07.csv /tmp/ag2_xrank_g815.csv)",
-      "Bash(PYTHONPATH=/Users/THIF/repos/BatterySwapAI2026-Example ./venv/bin/python -c ' *)",
-      "Bash(ps -eo pid,etime,command)",
-      "Bash(python3 -c ' *)",
-      "Bash(./venv/bin/python -c ' *)",
-      "Bash(ORIGINS=20,21 timeout 1200 ./venv/bin/python /tmp/screen48.py cliff)",
-      "Bash(time ORIGINS=20,21 ./venv/bin/python /tmp/screen48.py cliff)",
-      "Bash(time PYTHONPATH=/Users/THIF/repos/BatterySwapAI2026-Example ORIGINS=20,21 ./venv/bin/python /tmp/screen48.py cliff)",
-      "Bash(PYTHONPATH=__TRACKED_VAR__ nohup ./venv/bin/python /tmp/screen48.py model)",
-      "Bash(python3 -c \"print\\(','.join\\(str\\(i\\) for i in range\\(8,48\\)\\)\\)\")",
-      "Bash(ORIGINS=__CMDSUB_OUTPUT__ PYTHONPATH=__TRACKED_VAR__ nohup ./venv/bin/python /tmp/screen48.py gbm)",
-      "Bash(ps -eo pid,etime,args)",
-      "Bash(PYTHONPATH=__TRACKED_VAR__ nohup ./venv/bin/python /tmp/screen48.py fine)",
-      "Bash(PYTHONPATH=__TRACKED_VAR__ nohup ./venv/bin/python /tmp/unobs.py)",
-      "Bash(./venv/bin/python -m py_compile script.py batteryswap_example/*.py)",
-      "Bash(PYTHONPATH=__TRACKED_VAR__ ./venv/bin/python -c ' *)"
-    ]
-  }
-}

.gitignore

@@ -8,3 +8,6 @@ __pycache__/
 
 # browser/tool session artifacts
 .playwright-mcp/
+
+# local tooling config
+.claude/

analysis_swaps.py

@@ -1,138 +0,0 @@
-"""Characterize upward voltage jumps per device to detect undocumented battery swaps.
-
-Leak-free: uses only raw train timeseries (daily medians), no labels except for
-post-hoc split of detected events by recorded EOL.
-"""
-from pathlib import Path
-import json
-
-import numpy as np
-import pandas as pd
-
-from batteryswap_public.utils import load_dataset
-from batteryswap_example.features import daily_all
-
-BETA = 0.0049
-
-locations, timeseries, eol_times, scenarios = load_dataset(Path('dataset') / 'train')
-dd = daily_all(timeseries)
-dd['v_corr'] = dd['v'] - BETA * (dd['t'] - 20.0)
-dd = dd.sort_values(['device_id', 'day']).reset_index(drop=True)
-
-devices = dd['device_id'].unique()
-eol_rec = eol_times.dropna()  # devices with a RECORDED (non-NaT) EOL
-print(f"n_devices = {len(devices)}")
-print(f"n_devices with recorded EOL = {eol_rec.index.nunique()}")
-
-# ---- per-device consecutive-observation jumps -------------------------------
-g = dd.groupby('device_id', sort=False)
-dd['v_prev'] = g['v_corr'].shift(1)
-dd['day_prev'] = g['day'].shift(1)
-dd['jump'] = dd['v_corr'] - dd['v_prev']
-dd['gap_days'] = (dd['day'] - dd['day_prev']).dt.days
-
-jumps = dd.dropna(subset=['jump']).copy()
-
-# max upward jump per device
-idx = jumps.groupby('device_id')['jump'].idxmax()
-maxj = jumps.loc[idx, ['device_id', 'day', 'day_prev', 'v_prev', 'v_corr',
-                       'jump', 'gap_days']].copy()
-maxj = maxj.rename(columns={'v_prev': 'from_v', 'v_corr': 'to_v',
-                            'day': 'swap_date'})
-maxj['has_eol'] = maxj['device_id'].isin(eol_rec.index)
-maxj = maxj.sort_values('jump', ascending=False).reset_index(drop=True)
-
-qs = maxj['jump'].quantile([0.50, 0.90, 0.95, 0.99])
-print("\n=== Distribution of max upward jump across devices ===")
-for q, v in qs.items():
-    print(f"  q{int(q*100):02d} = {v:.4f} V")
-print(f"  max  = {maxj['jump'].max():.4f} V, min = {maxj['jump'].min():.4f} V")
-
-print("\n=== Top-40 devices by max upward jump ===")
-top40 = maxj.head(40)
-for _, r in top40.iterrows():
-    print(f"  {r['device_id']:>10s}  jump={r['jump']:.3f}  from={r['from_v']:.3f} "
-          f"to={r['to_v']:.3f}  gap={int(r['gap_days'])}d  "
-          f"date={r['swap_date'].date()}  EOL={'Y' if r['has_eol'] else 'N'}")
-
-# ---- candidate rules over ALL jump events (not just per-device max) ---------
-def apply_rule(df, thr, level_gate):
-    m = df['jump'] >= thr
-    if level_gate:
-        m &= (df['v_prev'] <= 2.85) & (df['v_corr'] >= 2.80)
-    return df[m]
-
-print("\n=== Candidate rules (events over all consecutive-day jumps) ===")
-rules = []
-for thr in (0.10, 0.15, 0.20):
-    for gate in (False, True):
-        ev = apply_rule(jumps, thr, gate)
-        ndev = ev['device_id'].nunique()
-        rules.append((thr, gate, len(ev), ndev))
-        gs = 'from<=2.85 & to>=2.80' if gate else 'no level gate'
-        print(f"  jump>={thr:.2f}, {gs:23s}: {len(ev):4d} events on {ndev:4d} devices")
-
-# stability: which device sets coincide?
-sets = {}
-for thr in (0.10, 0.15, 0.20):
-    for gate in (False, True):
-        ev = apply_rule(jumps, thr, gate)
-        sets[(thr, gate)] = set(ev['device_id'])
-print("\n=== Device-set overlaps (rows=rule, same set marked) ===")
-keys = list(sets)
-for i, k in enumerate(keys):
-    for k2 in keys[i+1:]:
-        a, b = sets[k], sets[k2]
-        if a == b:
-            rel = "IDENTICAL"
-        elif b <= a:
-            rel = f"{k2} subset of {k}, diff={sorted(a-b)}"
-        elif a <= b:
-            rel = f"{k} subset of {k2}, diff={sorted(b-a)}"
-        else:
-            rel = f"sym-diff={sorted(a ^ b)}"
-        print(f"  {k} vs {k2}: {rel}")
-
-# gap between clear swaps and noise: sort all max jumps, look at ordered values
-srt = maxj['jump'].sort_values(ascending=False).values
-print("\n=== Ordered max jumps, top 25 (look for a natural break) ===")
-print("  " + ", ".join(f"{v:.3f}" for v in srt[:25]))
-
-# ---- recommended rule events -------------------------------------------------
-# recommendation decided after inspecting output; emit full detail for all three
-# thresholds with the gate so the caller can verify stability
-for thr in (0.10, 0.15, 0.20):
-    ev = apply_rule(jumps, thr, True).copy()
-    ev['has_eol'] = ev['device_id'].isin(eol_rec.index)
-    ev['eol_time'] = ev['device_id'].map(eol_rec)
-    ev['post_eol'] = ev['has_eol'] & (ev['day'] >= ev['eol_time'].dt.normalize())
-    print(f"\n=== Events for rule jump>={thr:.2f} + gate (from<=2.85, to>=2.80) ===")
-    for _, r in ev.sort_values(['device_id', 'day']).iterrows():
-        et = r['eol_time'].date() if r['has_eol'] else '-'
-        print(f"  {r['device_id']:>10s}  date={r['day'].date()}  "
-              f"from={r['v_prev']:.3f}  to={r['v_corr']:.3f}  jump={r['jump']:.3f}  "
-              f"gap={int(r['gap_days'])}d  EOL={'Y' if r['has_eol'] else 'N'}  "
-              f"eol_time={et}  post_eol={'Y' if r['post_eol'] else 'N'}")
-    multi = ev.groupby('device_id').size()
-    multi = multi[multi > 1]
-    print(f"  -> {len(ev)} events, {ev['device_id'].nunique()} devices, "
-          f"{ev['has_eol'].sum()} events on EOL devices; "
-          f"devices with multiple swaps: {list(multi.index)} ({list(multi.values)})")
-
-# dump machine-readable summary for the structured answer
-out = {
-    'quantiles': {f"q{int(q*100)}": round(float(v), 4) for q, v in qs.items()},
-    'rules': [{'thr': t, 'gate': g, 'events': e, 'devices': d} for t, g, e, d in rules],
-}
-for thr in (0.10, 0.15, 0.20):
-    ev = apply_rule(jumps, thr, True).copy()
-    ev['has_eol'] = ev['device_id'].isin(eol_rec.index)
-    ev['eol_time'] = ev['device_id'].map(eol_rec)
-    ev['post_eol'] = ev['has_eol'] & (ev['day'] >= ev['eol_time'].dt.normalize())
-    out[f'events_{thr}'] = [
-        {'device_id': r['device_id'], 'swap_date': str(r['day'].date()),
-         'from_v': round(float(r['v_prev']), 4), 'to_v': round(float(r['v_corr']), 4),
-         'has_eol': bool(r['has_eol']), 'post_eol': bool(r['post_eol'])}
-        for _, r in ev.sort_values(['device_id', 'day']).iterrows()]
-Path('/tmp/swap_analysis.json').write_text(json.dumps(out, indent=1))
-print("\nwrote /tmp/swap_analysis.json")

exp_knn_screen.py

@@ -1,140 +0,0 @@
-"""
-Tier-0 forensic screen for donor-curve k-NN EOL estimation (user idea,
-2026-08-19): estimate days-to-EOL by aligning a device's recent corrected-
-voltage history against INDIVIDUAL dead devices' trajectories (not the
-population master curve), k-NN weighted by fit error.
-
-Walk-forward legal: donors restricted to deaths recorded <= origin; history
-restricted to data <= origin.
-
-Decision tests (no panel run):
-  T1 rank-of-misses: where do the champion's 132 saved-plan misses rank under
-     knn_tau among non-swapped transmitting descent-band devices?
-     (knee-slope channel scored median 35, zero top-5 -> closed)
-  T2 AUC vs dies42 on descent-band pairs, vs tau_projected (incumbent timer)
-  T3 timing: Spearman(knn_tau, realized days_to_eol) on dying pairs, vs
-     tau_projected - a win here would matter for day placement even if
-     inclusion doesn't move.
-"""
-
-import pickle
-import sys
-from pathlib import Path
-
-import numpy as np
-import pandas as pd
-from scipy.stats import spearmanr
-from sklearn.metrics import roc_auc_score
-
-sys.path.insert(0, '.')
-from batteryswap_example.features import daily_all, BETA_DEFAULT, WINDOW_DAYS
-import exp_x27 as x
-
-TAU_MAX = 300
-FIT_DAYS = 90
-K = 5
-V_BAND = 2.75      # score only descent-band devices; healthy matches are noise
-
-ts = pd.read_parquet('dataset/train/battery_metrics.parquet')
-d = daily_all(ts).sort_values(['device_id', 'day'])
-d['vc'] = d['v'] - BETA_DEFAULT * (d['t'] - 20.0)
-eol = (pd.read_csv('dataset/train/eol_times.csv', parse_dates=['end_time'])
-       .set_index('device_id')['end_time'])
-ctx = x.load_context(Path('dataset'))
-plans = pickle.load(open('/tmp/x27/plans_panel.pkl', 'rb'))
-table = pd.read_parquet('dataset/train/features.parquet')
-
-# donor bank: per dead device, daily-grid trajectory vc(tau), tau=days before EOL
-donors = {}
-for dev, e in eol.dropna().items():
-    h = d[(d['device_id'] == dev) & (d['day'] <= e)]
-    tau = (e.normalize() - h['day']).dt.days.values
-    ok = (tau >= 0) & (tau <= TAU_MAX)
-    if ok.sum() < 10:
-        continue
-    grid = np.arange(0, TAU_MAX + 1)
-    curve = np.interp(grid, tau[ok][::-1], h['vc'].values[ok][::-1])
-    donors[dev] = (e, np.maximum.accumulate(curve))
-
-print(f'donor bank: {len(donors)} usable of {eol.notna().sum()} deaths', flush=True)
-
-taus = np.arange(1, TAU_MAX - FIT_DAYS)
-
-def knn_tau_at(origin, devices):
-    """k-NN donor-matched days-to-EOL (minus staleness) for given devices."""
-    avail = [(e, c) for dev2, (e, c) in donors.items() if e <= origin]
-    if len(avail) < 5:
-        return pd.Series(np.nan, index=devices)
-    bank = np.stack([c for _, c in avail])           # donors x (TAU_MAX+1)
-    h = d[(d['day'] <= origin) & (d['day'] > origin - pd.Timedelta(days=FIT_DAYS))]
-    h = h[h['device_id'].isin(devices)]
-    out = pd.Series(np.nan, index=devices)
-    for dev, g in h.groupby('device_id'):
-        if len(g) < 5:
-            continue
-        delta = (origin - g['day']).dt.days.values
-        idx = taus[:, None] + delta[None, :]         # taus x points
-        # errors: donors x taus
-        err = np.abs(g['vc'].values[None, None, :] - bank[:, idx]).mean(axis=2)
-        best_t = err.argmin(axis=1)                  # per-donor best alignment
-        best_e = err[np.arange(len(bank)), best_t]
-        top = np.argsort(best_e)[:K]
-        w = 1.0 / (best_e[top] + 1e-3)
-        stale = float((origin - g['day'].max()).days)
-        out[dev] = float(np.average(taus[best_t[top]], weights=w)) - stale
-    return out
-
-rows, auc_rows, timing = [], [], []
-for idx in range(8, 42):
-    T = ctx[idx]['start_time']
-    sc_eol = ctx[idx]['eol']
-    pairs = table[table['origin_idx'] == idx].set_index('device_id')
-    band = pairs[(pairs['v_corr_last'] < V_BAND) & (pairs['staleness'] <= 14)]
-    scores = knn_tau_at(T, band.index)
-    valid = scores.dropna()
-    if len(valid) < 10:
-        continue
-    dies = band.loc[valid.index, 'dies42']
-
-    if dies.sum() > 0 and dies.sum() < len(dies):
-        auc_rows.append(dict(
-            origin=idx,
-            knn=roc_auc_score(dies, -valid),
-            tau=roc_auc_score(dies, -band.loc[valid.index, 'tau_projected']
-                              .fillna(1e6))))
-    dying = band.loc[valid.index][band.loc[valid.index, 'dies42'] == 1]
-    for dev2 in dying.index:
-        timing.append((valid[dev2], dying.at[dev2, 'days_to_eol'],
-                       dying.at[dev2, 'tau_projected']))
-
-    plan = plans[('logbag', idx)].set_index('battery')['day']
-    swapped = set(plan[plan <= T + pd.Timedelta(days=WINDOW_DAYS)].index)
-    dying_now = sc_eol[(sc_eol > T) & (sc_eol <= T + pd.Timedelta(days=WINDOW_DAYS))]
-    missed = [b for b in dying_now.index if b not in swapped]
-    ranked = valid[~valid.index.isin(swapped)].sort_values()
-    pos = {b: i + 1 for i, b in enumerate(ranked.index)}
-    for b in missed:
-        rows.append(dict(origin=idx, dev=b, rank=pos.get(b, np.nan),
-                         knn_tau=valid.get(b, np.nan), pool=len(ranked)))
-    print(f's_{idx}: band={len(valid)} misses={len(missed)} '
-          f'ranked={sum(1 for b in missed if b in pos)}', flush=True)
-
-r = pd.DataFrame(rows)
-a = pd.DataFrame(auc_rows)
-t = pd.DataFrame(timing, columns=['knn_tau', 'true', 'tau_proj']).dropna()
-
-print('\n=== T1 rank-of-misses (knee-slope channel scored median 35, 0 in top-5) ===')
-print(r['rank'].describe().round(1).to_string())
-print('misses at rank <=3: %d | <=5: %d | <=10: %d (of %d ranked, %d total)'
-      % ((r['rank'] <= 3).sum(), (r['rank'] <= 5).sum(),
-         (r['rank'] <= 10).sum(), r['rank'].notna().sum(), len(r)))
-print('\n=== T2 AUC on descent-band pairs (mean over %d origins) ===' % len(a))
-print('knn_tau %.3f  vs tau_projected %.3f  (knn wins %d/%d origins)'
-      % (a['knn'].mean(), a['tau'].mean(), (a['knn'] > a['tau']).sum(), len(a)))
-print('\n=== T3 timing on dying pairs (n=%d) ===' % len(t))
-print('spearman(knn_tau, true days_to_eol) = %.3f | tau_projected = %.3f'
-      % (spearmanr(t['knn_tau'], t['true']).statistic,
-         spearmanr(t['tau_proj'], t['true']).statistic))
-print('median |error| days: knn %.1f | tau_proj %.1f'
-      % ((t['knn_tau'] - t['true']).abs().median(),
-         (t['tau_proj'] - t['true']).abs().median()))

exp_vt_screen.py

@@ -1,151 +0,0 @@
-"""
-Tier-0 screen for V-T coupling features (user idea, 2026-08-19): rolling
-correlation/slope between voltage and temperature as a degradation signal
-(rising internal resistance -> stronger V-T coupling before the level moves).
-
-Variants, all leak-free (data <= origin only):
-  bvt_60     within-day V-T slope (hourly data), median over last 60d
-  bvt_delta  bvt_60 minus the device's own first-90-observed-days baseline
-  cvt_60     within-day V-T Pearson corr, median over last 60d
-  rcorr_90   the user's literal version: corr of 14d-rolling-median-detrended
-             daily median v and t over the last 90d
-Within-day variants are seasonality-proof by construction; rcorr_90 removes
-it by detrending.
-
-Adjudication: cell AUCs (all / healthy>2.55 / band<=2.75) vs v_drop and
-tau_projected + rank-of-misses on the champion's saved panel plans.
-"""
-
-import pickle
-import sys
-from pathlib import Path
-
-import numpy as np
-import pandas as pd
-from sklearn.metrics import roc_auc_score
-
-sys.path.insert(0, '.')
-from batteryswap_example.features import WINDOW_DAYS
-import exp_x27 as x
-
-MIN_TEMP, MAX_TEMP = 10, 30
-
-print('loading...', flush=True)
-ts = pd.read_parquet('dataset/train/battery_metrics.parquet').reset_index()
-ts['device_id'] = ts['device_id'].astype(str)
-ts['day'] = ts['end_time'].dt.normalize()
-table = pd.read_parquet('dataset/train/features.parquet')
-ctx = x.load_context(Path('dataset'))
-plans = pickle.load(open('/tmp/x27/plans_panel.pkl', 'rb'))
-
-# ---- day-level within-day V-T slope and corr, vectorized via groupby sums
-g = ts.groupby(['device_id', 'day'])
-agg = g.agg(n=('voltage', 'size'), sv=('voltage', 'sum'), st=('temperature', 'sum'),
-            svv=('voltage', lambda s: (s ** 2).sum()))
-# cross/second moments not expressible via named-agg lambdas cheaply; do manually
-ts['vt'] = ts['voltage'] * ts['temperature']
-ts['tt'] = ts['temperature'] ** 2
-agg['svt'] = g['vt'].sum()
-agg['stt'] = g['tt'].sum()
-n = agg['n']
-cov = agg['svt'] / n - (agg['sv'] / n) * (agg['st'] / n)
-var_t = agg['stt'] / n - (agg['st'] / n) ** 2
-var_v = agg['svv'] / n - (agg['sv'] / n) ** 2
-ok = (n >= 6) & (var_t > 0.25)
-day = pd.DataFrame({
-    'slope': (cov / var_t).where(ok),
-    'corr': (cov / np.sqrt(var_t * var_v)).where(ok & (var_v > 0)),
-}).reset_index()
-print(f'day-level: {day.slope.notna().sum()} usable device-days '
-      f'of {len(day)}', flush=True)
-
-# daily medians for the detrended rolling-corr variant
-dm = ts.groupby(['device_id', 'day']).agg(v=('voltage', 'median'),
-                                          t=('temperature', 'median')).reset_index()
-dm_by_dev = {k: v.set_index('day') for k, v in dm.groupby('device_id')}
-day_by_dev = {k: v.set_index('day') for k, v in day.groupby('device_id')}
-first_day = dm.groupby('device_id')['day'].min()
-
-
-def vt_feats(origin, devices):
-    out = {}
-    for dev in devices:
-        f = {}
-        dd = day_by_dev.get(dev)
-        if dd is not None:
-            h = dd[dd.index <= origin]
-            recent = h[h.index > origin - pd.Timedelta(days=60)]
-            base = h[h.index <= first_day[dev] + pd.Timedelta(days=90)]
-            if recent['slope'].notna().sum() >= 5:
-                f['bvt_60'] = recent['slope'].median()
-                f['cvt_60'] = recent['corr'].median()
-                if base['slope'].notna().sum() >= 5:
-                    f['bvt_delta'] = f['bvt_60'] - base['slope'].median()
-        m = dm_by_dev.get(dev)
-        if m is not None:
-            h = m[(m.index <= origin) & (m.index > origin - pd.Timedelta(days=90))]
-            if len(h) >= 20:
-                rv = h['v'] - h['v'].rolling('14D', min_periods=5).median()
-                rt = h['t'] - h['t'].rolling('14D', min_periods=5).median()
-                ok2 = rv.notna() & rt.notna()
-                if ok2.sum() >= 15 and rt[ok2].std() > 0.2:
-                    f['rcorr_90'] = float(np.corrcoef(rv[ok2], rt[ok2])[0, 1])
-        out[dev] = f
-    return pd.DataFrame(out).T
-
-
-VARIANTS = ['bvt_60', 'bvt_delta', 'cvt_60', 'rcorr_90']
-auc_rows, resid = [], {v: [] for v in VARIANTS + ['tau']}
-for idx in range(8, 42):
-    T = ctx[idx]['start_time']
-    pr = table[table['origin_idx'] == idx].set_index('device_id')
-    f = vt_feats(T, pr.index)
-    for col in VARIANTS:
-        if col not in f:
-            f[col] = np.nan
-    for cell, mask in [('all', pr['dies42'].notna()),
-                       ('healthy', pr['v_corr_last'] > 2.55),
-                       ('band', pr['v_corr_last'] <= 2.75)]:
-        y = pr.loc[mask, 'dies42']
-        if y.sum() == 0 or y.sum() == len(y):
-            continue
-        row = dict(origin=idx, cell=cell,
-                   vdrop=roc_auc_score(y, pr.loc[mask, 'v_drop'].fillna(0)),
-                   tau=roc_auc_score(y, -pr.loc[mask, 'tau_projected'].fillna(1e6)))
-        for col in VARIANTS:
-            s = f.loc[mask.index[mask], col] if col in f else pd.Series(dtype=float)
-            s = s.reindex(mask.index[mask])
-            if s.notna().sum() >= 10 and y[s.notna()].nunique() == 2:
-                # sign: degradation could raise coupling -> score by |value| high
-                row[col] = roc_auc_score(y[s.notna()], s.dropna())
-        auc_rows.append(row)
-
-    plan = plans[('logbag', idx)].set_index('battery')['day']
-    swapped = set(plan[plan <= T + pd.Timedelta(days=WINDOW_DAYS)].index)
-    sc_eol = ctx[idx]['eol']
-    missed = {b for b in sc_eol[(sc_eol > T) &
-                                (sc_eol <= T + pd.Timedelta(days=WINDOW_DAYS))].index
-              if b not in swapped}
-    pool = [b for b in pr.index if b not in swapped and pr.at[b, 'staleness'] <= 14]
-    orders = {'tau': sorted(pool, key=lambda b: pr.at[b, 'tau_projected']
-                            if np.isfinite(pr.at[b, 'tau_projected']) else 1e9)}
-    for col in VARIANTS:
-        s = f[col].reindex(pool) if col in f else pd.Series(np.nan, index=pool)
-        orders[col] = sorted(pool, key=lambda b: -(s.get(b) if np.isfinite(s.get(b, np.nan)) else -1e9))
-    for m, o in orders.items():
-        resid[m].append([sum(1 for b in o[:k] if b in missed) for k in (3, 5, 10)])
-    print(f's_{idx} done', flush=True)
-
-a = pd.DataFrame(auc_rows)
-print('\n=== mean AUC by cell (NaN-skipped per origin) ===')
-print(a.groupby('cell')[VARIANTS + ['vdrop', 'tau']].mean().round(3).to_string())
-print('\nhealthy cell, per-variant: origins where variant beats v_drop:')
-h = a[a.cell == 'healthy']
-for col in VARIANTS:
-    if col in h:
-        sub = h[[col, 'vdrop']].dropna()
-        print(f'  {col}: {(sub[col] > sub.vdrop).sum()}/{len(sub)}')
-print('\n=== residual misses in top-3/5/10 ===')
-for m in resid:
-    v = np.array(resid[m]).sum(axis=0)
-    print(f'{m:10s} {v[0]}/{v[1]}/{v[2]}')

exp_x27.py

@@ -1,687 +0,0 @@
-"""
-Phase 27 experiment runner: new-direction candidates from the 2026-08-19
-fresh-eyes review. New file - the frozen modules (models_cv.py, features.py,
-aux/fht_features.py) are imported, never edited; custom models/planners are
-defined here and run through a local copy of the walk-forward loop.
-
-Data findings motivating the slate (analysis in this file's `facts` stage):
-  - every one of the 82 recorded deaths crossed 2.45 V before its EOL, and
-    42d before the crossing all of them sat in the 2.47-2.77 descent band;
-    the healthy->threshold descent takes median 229d (min 4). Detection is
-    saturated; the residual is timing.
-  - the crossing->recorded-EOL lag is median 33d, p90 ~140d - wider than the
-    42d window. Label arrival, not physics, is the dominant randomness.
-  - all 82 deaths transmitted up to the EOL day (median pre-EOL silence 0d):
-    labels require telemetry; devices dark at their would-be death ghost out.
-
-Candidates (all walk-forward, panel s_8..s_41):
-  X1 lag-rank    - parametric lognormal+cure arrival model used for RANKING
-                   only; inclusion count stays sum(p42) from logbag. The
-                   phase-17 LAH refutation was a count explosion; this
-                   decouples the channels.
-  X2 twostage    - generative recorded-EOL date model: Wiener crossing CDF
-                   (not-crossed) convolved with lognormal+cure label lag;
-                   crossed devices get the truncated lag directly. Plus a
-                   count-anchored variant (logbag p42 multiset, twostage order).
-  X3 arr-feature - the arrival clock as plain logistic features
-                   (days_below_245 raw; stacked arr42 probability).
-  X4 dark cliff  - staleness>14 indicator (labels need telemetry; a linear
-                   staleness term cannot express the cliff).
-  X5 adrop       - add/drop moves in the day optimizer, priced by exact
-                   mobility deltas (opportunistic inclusion).
-  X6 noise       - label-process simulator: re-score FIXED champion plans
-                   under resampled crossing->EOL lags to measure how much
-                   panel-mean noise the label process alone contributes.
-  X7 policy      - 3-parameter direct rule (tau_projected threshold, dark
-                   cliff, day fraction) - "was the ML stack needed" control.
-
-Stages (argv[1]): facts | smoke | panel | adrop | noise | policy | full
-"""
-
-import hashlib
-import os
-import pickle
-import sys
-import time
-from pathlib import Path
-
-import numpy as np
-import pandas as pd
-from scipy.stats import norm
-
-from batteryswap_public.evaluate import evaluate_plan
-
-import batteryswap_example.models_cv as mc
-from batteryswap_example.models_cv import (
-    FEATURES, DAY_QUANTILES, SklearnRUL, LogBagFeats, DayQuantiles,
-    training_pairs, load_context, cost_curves, optimize_days, mobility_cost,
-    plan_tsroute, assemble, _batch_days, summarize, selection_stability,
-    ALL_ORIGINS)
-from batteryswap_example.features import load as load_features, WINDOW_DAYS
-
-OUT = Path('/tmp/x27')
-OUT.mkdir(exist_ok=True)
-
-CHAMPION = dict(model='logbag', planner='tsroute', p_star=0.8, day_q=0.05,
-                ec_margin=0.8, ec_pmax=25.0, split_days=7)
-
-
-def module_md5s():
-    """Frozen-file guard (parallel-session protocol): record what we ran against."""
-    for mod in ('batteryswap_example/models_cv.py', 'batteryswap_example/features.py'):
-        h = hashlib.md5(Path(mod).read_bytes()).hexdigest()
-        print(f'md5 {mod} {h}', flush=True)
-    # knob-existence asserts: a checkout race would silently no-op these
-    for attr in ('alpha', 'ec_margin', 'ec_pmax', 'split_days', 'floor_scale',
-                 'recent_deaths', 'p_be', 'floor_day', 'p_conf'):
-        assert hasattr(plan_tsroute, attr), attr
-
-
-def load_table(dataset_path=Path('dataset')):
-    """Replicates models_cv.main()'s feature merges, plus phase-27 columns."""
-    table = load_features(dataset_path)
-    table['v_proj'] = table['v_last'] + table['slope_lastk'] * table['staleness']
-    table['r_vlast'] = table.groupby('origin_idx')['v_last'].rank(pct=True)
-    table['r_tau'] = table.groupby('origin_idx')['tau_projected'].rank(pct=True)
-    from batteryswap_example.aux_features import load as load_aux
-    aux = load_aux(dataset_path)
-    dup = [c for c in aux.columns if c in table.columns
-           and c not in ('device_id', 'origin_idx')]
-    table = table.merge(aux.drop(columns=dup), on=['device_id', 'origin_idx'],
-                        how='left')
-    from batteryswap_example.fht_features import load as load_fht
-    table = table.merge(load_fht(dataset_path), on=['device_id', 'origin_idx'],
-                        how='left')
-    table['fht_days'] = ((table['v_corr_last'] - 2.40)
-                         / table['wnr_m'].clip(lower=1e-4)
-                         - table['staleness']).clip(-300, 600)
-    # phase-27: staleness cliff indicator (all 82 recorded deaths transmitted
-    # up to their EOL day - labels require telemetry)
-    table['dark14'] = (table['staleness'] > 14).astype(float)
-    return table
-
-
-# ------------------------------------------------------------- label-lag model
-
-class LagModel:
-    """Lognormal + cure-mass model of the crossing -> recorded-EOL lag.
-
-    One observation per crossed device (latest origin row, LogbagLAH's
-    construction): event = EOL on file (lag = eol - cross_date), censored
-    otherwise (at days_below_245). Cure mass pi absorbs ghosts. 3-parameter
-    MLE replaces phase-17's 14d-bucket KM (82 events cannot support a
-    nonparametric curve; they can support 3 parameters).
-    """
-
-    MU0, SIG0, PI0 = np.log(33.0), 1.0, 0.3
-
-    def fit(self, train):
-        crossed = train['days_below_245'].notna() & (train['days_below_245'] >= 0)
-        sub = train[crossed]
-        per_dev = sub.sort_values('origin').groupby('device_id').tail(1)
-        cross_date = per_dev['origin'] - pd.to_timedelta(per_dev['days_below_245'],
-                                                         unit='D')
-        has_eol = per_dev['eol'].notna()
-        lag = np.where(has_eol, (per_dev['eol'] - cross_date).dt.days,
-                       per_dev['days_below_245'])
-        lag = np.clip(lag.astype(float), 0.5, None)
-        event = has_eol.values
-        self.n_events = int(event.sum())
-        self.mu, self.sigma, self.pi = self.MU0, self.SIG0, self.PI0
-        if self.n_events >= 15:
-            from scipy.optimize import minimize
-            L = np.log(lag)
-
-            def nll(p):
-                mu, ls, lp = p
-                sig, pi = np.exp(ls), 1.0 / (1.0 + np.exp(-lp))
-                z = (L - mu) / sig
-                ll_e = np.log1p(-pi) + norm.logpdf(z) - np.log(sig * lag)
-                ll_c = np.log(pi + (1.0 - pi) * norm.sf(z) + 1e-12)
-                return -np.where(event, ll_e, ll_c).sum()
-
-            r = minimize(nll, x0=[self.MU0, np.log(self.SIG0), 0.0],
-                         method='Nelder-Mead',
-                         options=dict(maxiter=2000, xatol=1e-4, fatol=1e-4))
-            self.mu = float(r.x[0])
-            self.sigma = float(np.exp(r.x[1]))
-            self.pi = float(1.0 / (1.0 + np.exp(-r.x[2])))
-        return self
-
-    def F(self, t):
-        t = np.maximum(np.asarray(t, dtype=float), 1e-9)
-        return norm.cdf((np.log(t) - self.mu) / self.sigma)
-
-    def arrival(self, a, h=WINDOW_DAYS):
-        """P(EOL recorded within next h days | crossed a days ago, no label yet)."""
-        a = np.asarray(a, dtype=float)
-        Fa, Fah = self.F(a), self.F(a + h)
-        denom = self.pi + (1.0 - self.pi) * (1.0 - Fa)
-        return np.where(denom > 1e-12, (1.0 - self.pi) * (Fah - Fa) / denom, 0.0)
-
-    def arrival_curve(self, a, tgrid):
-        """Vector version over a horizon grid: devices x len(tgrid)."""
-        a = np.asarray(a, dtype=float)[:, None]
-        Fa = self.F(a)
-        Fat = self.F(a + tgrid[None, :])
-        denom = self.pi + (1.0 - self.pi) * (1.0 - Fa)
-        return np.where(denom > 1e-12, (1.0 - self.pi) * (Fat - Fa) / denom, 0.0)
-
-
-class CrossingModel:
-    """Wiener first-passage CDF to the 2.45 crossing (WienerFHT's math with the
-    barrier at the crossing, not the EOL). Unnormalized through silence: mass
-    that crossed while dark is DROPPED (dark devices don't get labels), which
-    bakes the staleness cliff into the physics instead of renormalizing it
-    away as WienerFHT does."""
-
-    K_SHRINK, BARRIER = 6.0, 2.45
-
-    def fit(self, train):
-        ok = train['wnr_n'] >= 3
-        self.m_pop = float(np.nanmedian(train.loc[ok, 'wnr_m'])) if ok.any() else 1e-4
-        self.s2_pop = float(np.nanmedian(train.loc[ok, 'wnr_s2'])) if ok.any() else 1e-4
-        return self
-
-    def cross_pmf(self, rows, tgrid):
-        """Unconditional-first-passage pmf on origin-relative days tgrid."""
-        w = rows['wnr_n'].fillna(0).values
-        m = np.where(w > 0, rows['wnr_m'].fillna(self.m_pop).values, self.m_pop)
-        s2 = np.where(w > 0, rows['wnr_s2'].fillna(self.s2_pop).values, self.s2_pop)
-        m = (w * m + self.K_SHRINK * self.m_pop) / (w + self.K_SHRINK)
-        s2 = np.maximum((w * s2 + self.K_SHRINK * self.s2_pop) / (w + self.K_SHRINK),
-                        1e-8)
-        d = rows['v_corr_last'].values - self.BARRIER
-        d = np.where(np.isnan(d), 0.35, d)
-        d = np.maximum(d, 1e-6)
-        stale = rows['staleness'].fillna(0).clip(lower=0).values
-
-        from scipy.special import log_ndtr
-
-        def F(times):
-            sig = np.sqrt(s2[:, None] * times)
-            z1 = (m[:, None] * times - d[:, None]) / sig
-            z2 = -(m[:, None] * times + d[:, None]) / sig
-            with np.errstate(over='ignore'):
-                term2 = np.exp(np.minimum(
-                    2.0 * m[:, None] * d[:, None] / s2[:, None] + log_ndtr(z2), 0.0))
-            return np.clip(norm.cdf(z1) + term2, 0.0, 1.0)
-
-        edges = stale[:, None] + np.concatenate([[0.0], tgrid])[None, :]
-        Fe = F(np.maximum(edges, 1e-6))
-        return np.diff(Fe, axis=1)  # devices x len(tgrid), NOT renormalized
-
-
-class TwoStage:
-    """X2: generative recorded-EOL date model.
-
-    crossed  -> truncated lognormal+cure arrival curve
-    else     -> Wiener crossing pmf (x) (1-pi) * lognormal lag CDF
-    p42 and conditional day quantiles both derive from the same curve.
-    """
-
-    def fit(self, train):
-        self.lag = LagModel().fit(train)
-        self.cross = CrossingModel().fit(train)
-        return self
-
-    def predict(self, rows):
-        n = len(rows)
-        tgrid = np.arange(1, WINDOW_DAYS + 1, dtype=float)
-        R = np.zeros((n, WINDOW_DAYS))
-
-        a = rows['days_below_245'].values.astype(float)
-        crossed = np.isfinite(a) & (a >= 0)
-        if crossed.any():
-            R[crossed] = self.lag.arrival_curve(a[crossed], tgrid)
-        if (~crossed).any():
-            sub = rows[~crossed]
-            pmf = self.cross.cross_pmf(sub, tgrid)      # devices x 42
-            FL = self.lag.F(tgrid)                       # lag CDF at 1..42
-            # R(t) = sum_{s<=t} pmf[s] * (1-pi) * F_L(t-s)
-            conv = np.zeros_like(pmf)
-            for j, t in enumerate(tgrid.astype(int)):
-                # F_L(t-s) for s=1..t ; F_L(0)=0
-                fl = np.concatenate([FL[:t][::-1], np.zeros(len(tgrid) - t)])
-                conv[:, j] = (pmf * fl[None, :]).sum(axis=1)
-            R[~crossed] = (1.0 - self.lag.pi) * conv
-
-        p42 = np.clip(R[:, -1], 0.0, 1.0)
-        out = pd.DataFrame({'p42': p42}, index=rows.index)
-        qcdf = R / np.maximum(p42[:, None], 1e-12)
-        for q in DAY_QUANTILES:
-            idx = (qcdf >= q).argmax(axis=1)
-            hit = qcdf[:, -1] >= q
-            out[f'day_q{q}'] = np.where(hit, tgrid[idx], WINDOW_DAYS).astype(float)
-        return out
-
-
-class TwoStageAnchor:
-    """X2b: logbag's p42 multiset (count preserved exactly) reordered by the
-    TwoStage date model; TwoStage's own day quantiles."""
-
-    def fit(self, train):
-        self.a = SklearnRUL('logbag').fit(train)
-        self.b = TwoStage().fit(train)
-        return self
-
-    def predict(self, rows):
-        out = self.b.predict(rows)
-        pa = self.a.predict(rows)['p42']
-        rank = out['p42'].rank(method='first')
-        vals = np.sort(pa.values)
-        out['p42'] = pd.Series(vals, index=rank.sort_values().index).reindex(out.index)
-        return out
-
-
-class LagRank:
-    """X1: champion logbag untouched (count = sum p42), but crossed devices are
-    RANKED by the parametric arrival probability instead of raw p42."""
-
-    def __init__(self, mode='override', w=0.3):
-        self.mode = mode
-        self.w = w
-
-    def fit(self, train):
-        self.base = SklearnRUL('logbag').fit(train)
-        self.lag = LagModel().fit(train)
-        return self
-
-    def predict(self, rows):
-        out = self.base.predict(rows)
-        out['rank_score'] = out['p42']
-        crossed = (rows['days_below_245'].notna() & (rows['days_below_245'] >= 0)
-                   & (rows['v_corr_last'] < 2.50)).values
-        if crossed.any():
-            A = self.lag.arrival(rows.loc[crossed, 'days_below_245'].values)
-            if self.mode == 'override':
-                out.loc[crossed, 'rank_score'] = A
-            else:
-                out.loc[crossed, 'rank_score'] = out.loc[crossed, 'p42'] + self.w * A
-        return out
-
-
-class LogBagArr:
-    """X3b: stacked arrival probability as a logistic FEATURE - the soft
-    version of LAH; the logistic recalibrates its weight, so the count channel
-    cannot explode."""
-
-    def fit(self, train):
-        self.lag = LagModel().fit(train)
-        t = train.copy()
-        t['arr42'] = self._arr(t)
-        self.inner = LogBagFeats(['arr42']).fit(t)
-        return self
-
-    def _arr(self, rows):
-        a = rows['days_below_245'].values.astype(float)
-        crossed = np.isfinite(a) & (a >= 0)
-        out = np.zeros(len(rows))
-        out[crossed] = self.lag.arrival(a[crossed])
-        return out
-
-    def predict(self, rows):
-        r = rows.copy()
-        r['arr42'] = self._arr(r)
-        return self.inner.predict(r)
-
-
-X27_MODELS = {
-    'lagrank_ovr': lambda: LagRank('override'),
-    'lagrank_b03': lambda: LagRank('blend', 0.3),
-    'twostage': TwoStage,
-    'ts_anchor': TwoStageAnchor,
-    'logbag_db245': lambda: LogBagFeats(['days_below_245']),
-    'logbag_arr': LogBagArr,
-    'logbag_dark': lambda: LogBagFeats(['dark14']),
-}
-
-
-def model_factory(name):
-    if name in X27_MODELS:
-        return X27_MODELS[name]
-    return mc.resolve_model(name)
-
-
-# --------------------------------------------------------------- X5 planner
-
-def _day_batteries(day_of, groups, loc):
-    """Emission-order day->batteries map: within a day sort by building, room
-    (assemble()'s order, which is what the evaluator prices)."""
-    out = {}
-    for g, d in day_of.items():
-        out.setdefault(d, []).extend(groups[g])
-    for d in out:
-        out[d].sort(key=lambda b: (loc.at[b, 'building'], loc.at[b, 'room']))
-    return out
-
-
-def plan_adrop(pred, included, locations, start_time, day_q,
-               travel_costs=None, settings=None):
-    """Champion tsroute inclusion + day search, then add/drop refinement:
-    adds priced at the marginal EXACT mobility of joining an existing visit
-    day; drops symmetric. Capped at 20 accepted moves."""
-    # champion inclusion (copy of plan_tsroute's chosen-set logic, no floor)
-    psum = pred['p42'].sum()
-    n = int(round(0.8 * psum))
-    order_col = 'rank_score' if 'rank_score' in pred.columns else 'p42'
-    chosen = pred[order_col].nlargest(n).index
-    curves, park = cost_curves(pred, locations, start_time)
-    end = locations.set_index('battery')['end_time'].dt.tz_localize(None)
-    e_far = ((end.reindex(pred.index) - start_time).dt.total_seconds() / 86400
-             + 30.0).values
-    fp_price = mc.EARLY * (e_far - WINDOW_DAYS)
-    be = pred.index[(e_far >= WINDOW_DAYS)
-                    & (curves.min(axis=1) < 0.8 * park)
-                    & (fp_price <= 25.0)]
-    chosen = chosen.union(be)
-    if len(chosen) == 0:
-        return assemble(pd.Series(dtype='datetime64[ns]'), locations, start_time)
-
-    curves_df = pd.DataFrame(curves, index=pred.index)
-    park_s = pd.Series(park, index=pred.index)
-    bldg_all = locations.set_index('battery')['building']
-    room_all = locations.set_index('battery')['room']
-    loc = locations.set_index('battery')
-
-    sub = curves_df.loc[chosen]
-    bucket = sub.values.argmin(axis=1) // 7
-    key = bldg_all.reindex(chosen).astype(str) + '~' + pd.Series(
-        bucket, index=chosen).astype(str)
-    groups = {g: sorted(m.index, key=lambda b: room_all.get(b, ''))
-              for g, m in sub.groupby(key)}
-    travel_hours = travel_costs.set_index(['from', 'to'])['hours'].unstack()
-
-    day_of = optimize_days(groups, sub, locations, travel_hours,
-                           settings, start_time)
-
-    def mob(day_bats):
-        return mobility_cost(day_bats, locations, travel_hours, settings,
-                             start_time)
-
-    base_bats = _day_batteries(day_of, groups, loc)
-    base_mob = mob(base_bats)
-    in_plan = {b for bs in groups.values() for b in bs}
-    # adds can only ever pay if park >= best timing cost (mobility delta >= 0);
-    # allow a small slack for same-day-building complementarity
-    curve_min = curves_df.min(axis=1)
-    pool = [b for b in pred.index
-            if b not in in_plan and np.isfinite(park_s[b])
-            and park_s[b] > curve_min[b] - 1.0]
-    moves = 0
-    improved = True
-    while improved and moves < 20:
-        improved = False
-        # ---- adds: cheapest (timing + marginal mobility - park) candidate
-        visit_days = sorted(set(day_of.values()))
-        best = None
-        for b in pool:
-            if b in in_plan:
-                continue
-            by_timing = sorted(visit_days, key=lambda d: curves_df.at[b, d])[:4]
-            cand_days = by_timing + [int(curves_df.loc[b].values.argmin())]
-            for d in cand_days:
-                trial = {k: list(v) for k, v in base_bats.items()}
-                trial.setdefault(d, []).append(b)
-                trial[d].sort(key=lambda x: (loc.at[x, 'building'],
-                                             loc.at[x, 'room']))
-                delta = (curves_df.at[b, d] + mob(trial) - base_mob
-                         - park_s[b])
-                if delta < -1e-9 and (best is None or delta < best[0]):
-                    best = (delta, b, d, trial)
-        if best is not None:
-            _, b, d, trial = best
-            groups[f'add~{b}'] = [b]
-            day_of[f'add~{b}'] = d
-            in_plan.add(b)
-            base_bats, base_mob = trial, mob(trial)
-            moves += 1
-            improved = True
-        # ---- drops
-        best = None
-        for g, bs in list(groups.items()):
-            for b in bs:
-                trial = {k: [x for x in v if x != b]
-                         for k, v in base_bats.items()}
-                trial = {k: v for k, v in trial.items() if v}
-                delta = (park_s[b] + mob(trial) - base_mob
-                         - curves_df.at[b, day_of[g]])
-                if delta < -1e-9 and (best is None or delta < best[0]):
-                    best = (delta, g, b, trial)
-        if best is not None:
-            _, g, b, trial = best
-            groups[g] = [x for x in groups[g] if x != b]
-            if not groups[g]:
-                del groups[g], day_of[g]
-            in_plan.discard(b)
-            base_bats, base_mob = trial, mob(trial)
-            moves += 1
-            improved = True
-
-    offs = {b: d for d, bs in base_bats.items() for b in bs}
-    offs = pd.Series(offs, dtype=float)
-    days = start_time + pd.to_timedelta(offs.values, unit='D')
-    return assemble(pd.Series(days, index=offs.index), locations, start_time)
-
-
-# ------------------------------------------------------------------- runner
-
-def set_champion_knobs(cfg):
-    plan_tsroute.alpha = cfg.get('p_star', 0.8)
-    plan_tsroute.p_be = 0.0
-    plan_tsroute.floor_day = 0
-    plan_tsroute.p_conf = 0.5
-    plan_tsroute.ec_margin = cfg.get('ec_margin', 0.8)
-    plan_tsroute.ec_pmax = cfg.get('ec_pmax', 25.0)
-    plan_tsroute.split_days = cfg.get('split_days', 7)
-    plan_tsroute.floor_scale = 0.0
-    plan_tsroute.recent_deaths = 0.0
-
-
-def walk(table, ctx, origins, configs, restarts=0, iters=20, save_plans=None):
-    optimize_days.restarts = restarts
-    optimize_days.iters = iters
-    optimize_days.constructive = False
-    optimize_days.nn_order = False
-    fitted, records, plans = {}, [], {}
-    for idx in origins:
-        T = ctx[idx]['start_time']
-        rows = table[table['origin_idx'] == idx].set_index('device_id')
-        for cfg in configs:
-            m = cfg['model']
-            if (m, idx) not in fitted:
-                fitted[(m, idx)] = model_factory(m)().fit(training_pairs(table, T))
-            pred = fitted[(m, idx)].predict(rows)
-            set_champion_knobs(cfg)
-            planner = cfg.get('planner', 'tsroute')
-            kwargs = dict(travel_costs=ctx[idx]['travel_costs'],
-                          settings=ctx[idx]['settings'])
-            fn = plan_adrop if planner == 'adrop' else plan_tsroute
-            plan = fn(pred, None, ctx[idx]['locs'], T, cfg.get('day_q', 0.05),
-                      **kwargs)
-            _, _, overall = evaluate_plan(
-                plan, ctx[idx]['locs'], ctx[idx]['travel_costs'],
-                ctx[idx]['settings'], eol_times=ctx[idx]['eol'],
-                start_time=T, verbose=0)
-            records.append(dict(overall, origin=idx, **cfg))
-            if save_plans is not None:
-                plans[(cfg['name'], idx)] = plan
-        line = ' '.join(f"{r['name']}={r['total_cost']:.0f}"
-                        for r in records if r['origin'] == idx)
-        print(f's_{idx}: {line}', flush=True)
-    if save_plans is not None:
-        with open(save_plans, 'wb') as f:
-            pickle.dump(plans, f)
-    return pd.DataFrame(records)
-
-
-# ---------------------------------------------------------------- X6 noise
-
-def stage_noise(table, ctx, origins, draws=150, seed=0):
-    """Re-score the saved champion plans under resampled label processes.
-    Crossing dates are real (physics); each draw resamples every crossed
-    device's lag from the FULL-train lognormal+cure fit, truncated per
-    scenario so devices unlabeled at the origin stay unlabeled (consistency
-    with the plan having been made). Output: distribution of the panel mean."""
-    with open(OUT / 'plans_champion.pkl', 'rb') as f:
-        plans = pickle.load(f)
-    lag = LagModel().fit(table)
-    print(f'lag fit: mu={lag.mu:.3f} sigma={lag.sigma:.3f} pi={lag.pi:.3f} '
-          f'events={lag.n_events}', flush=True)
-
-    # real crossing date per device: from the latest origin row that has one
-    latest = (table.dropna(subset=['days_below_245'])
-              .sort_values('origin').groupby('device_id').tail(1))
-    cross = (latest['origin'] - pd.to_timedelta(latest['days_below_245'], unit='D'))
-    cross.index = latest['device_id'].values
-
-    rng = np.random.RandomState(seed)
-    names = sorted({k[0] for k in plans})
-    means = {n: [] for n in names}
-    for k in range(draws):
-        u_cure = pd.Series(rng.rand(len(cross)), index=cross.index)
-        u_lag = pd.Series(rng.rand(len(cross)), index=cross.index)
-        per_origin = {n: [] for n in names}
-        for idx in origins:
-            T = ctx[idx]['start_time']
-            eol_real = ctx[idx]['eol']
-            synth = pd.Series(pd.NaT, index=eol_real.index)
-            for dev in eol_real.index:
-                if dev not in cross.index:
-                    continue
-                c = cross[dev]
-                a = max((T - c).days, 0) if c <= T else 0
-                # truncation: no label by origin => condition on L > a
-                Sa = 1.0 - lag.F(a) if a > 0 else 1.0
-                p_cure = lag.pi / (lag.pi + (1 - lag.pi) * Sa)
-                if u_cure[dev] < p_cure:
-                    continue  # ghost in this draw
-                # inverse-CDF sample of L | L > a
-                q = lag.F(a) + u_lag[dev] * (1.0 - lag.F(a))
-                L = float(np.exp(lag.mu + lag.sigma * norm.ppf(min(q, 1 - 1e-9))))
-                synth[dev] = (c + pd.Timedelta(days=round(L))).normalize()
-            for n in names:
-                plan = plans[(n, idx)]
-                _, _, overall = evaluate_plan(
-                    plan, ctx[idx]['locs'], ctx[idx]['travel_costs'],
-                    ctx[idx]['settings'], eol_times=synth, start_time=T,
-                    verbose=0)
-                per_origin[n].append(overall['total_cost'])
-        for n in names:
-            means[n].append(np.mean(per_origin[n]))
-        if (k + 1) % 10 == 0:
-            print(f'draw {k+1}/{draws}: ' +
-                  ' '.join(f'{n}={np.mean(means[n]):.1f}+-{np.std(means[n]):.1f}'
-                           for n in names), flush=True)
-    res = pd.DataFrame(means)
-    res.to_csv(OUT / 'noise_panel_means.csv', index=False)
-    print('\n=== label-process noise on the panel mean ===')
-    print(res.describe().round(1).to_string())
-    if len(names) == 2:
-        d = res[names[1]] - res[names[0]]
-        print(f'\npaired delta {names[1]}-{names[0]}: mean {d.mean():.1f} '
-              f'sd {d.std():.1f} p5 {d.quantile(.05):.1f} p95 {d.quantile(.95):.1f}')
-    return res
-
-
-# --------------------------------------------------------------- X7 policy
-
-def stage_policy(table, ctx):
-    """3-parameter direct rule: include iff tau_projected <= theta and
-    staleness <= cliff; swap day = frac * tau_projected; building-batched.
-    Tuned on odd panel origins, scored on even (and full)."""
-    results = []
-    origins = list(ALL_ORIGINS)
-    for theta in (10, 20, 30, 45, 60):
-        for cliff in (14, 30, 10000):
-            for frac in (0.0, 0.3, 0.6):
-                per = {}
-                for idx in origins:
-                    T = ctx[idx]['start_time']
-                    rows = table[table['origin_idx'] == idx].set_index('device_id')
-                    tau = rows['tau_projected']
-                    inc = tau.notna() & (tau <= theta) & (rows['staleness'] <= cliff)
-                    offs = (frac * tau[inc].clip(lower=0)).clip(0, WINDOW_DAYS).round()
-                    if len(offs):
-                        plan = _batch_days(offs, ctx[idx]['locs'], T)
-                    else:
-                        plan = assemble(pd.Series(dtype='datetime64[ns]'),
-                                        ctx[idx]['locs'], T)
-                    _, _, overall = evaluate_plan(
-                        plan, ctx[idx]['locs'], ctx[idx]['travel_costs'],
-                        ctx[idx]['settings'], eol_times=ctx[idx]['eol'],
-                        start_time=T, verbose=0)
-                    per[idx] = overall['total_cost']
-                odd = np.mean([v for k, v in per.items() if k % 2 == 1])
-                even = np.mean([v for k, v in per.items() if k % 2 == 0])
-                results.append(dict(theta=theta, cliff=cliff, frac=frac,
-                                    odd=odd, even=even,
-                                    full=np.mean(list(per.values()))))
-                print(f'theta={theta} cliff={cliff} frac={frac}: '
-                      f'odd={odd:.1f} even={even:.1f}', flush=True)
-    df = pd.DataFrame(results)
-    df.to_csv(OUT / 'policy_grid.csv', index=False)
-    best = df.sort_values('odd').iloc[0]
-    print('\nbest on odd origins:', dict(best))
-    print(f"its even-origin (holdout) score: {best['even']:.1f}, full {best['full']:.1f}")
-    return df
-
-
-# ------------------------------------------------------------------- stages
-
-def main():
-    stage = sys.argv[1] if len(sys.argv) > 1 else 'smoke'
-    module_md5s()
-    t0 = time.time()
-    dataset_path = Path(os.environ.get('BATTERYSWAP_DATASET_PATH', 'dataset'))
-    table = load_table(dataset_path)
-    ctx = load_context(dataset_path)
-    print(f'loaded in {time.time()-t0:.0f}s', flush=True)
-
-    def cfgs(*names, planner='tsroute'):
-        return [dict(CHAMPION, model=n, planner=planner,
-                     name=(n if planner == 'tsroute' else f'{n}+{planner}'))
-                for n in names]
-
-    if stage == 'smoke':
-        df = walk(table, ctx, [8, 9], cfgs('logbag', 'logbag_dark'))
-        ref = pd.read_csv('research/baselines/tier1_champion.csv')
-        for idx in (8, 9):
-            mine = df[(df.origin == idx) & (df.model == 'logbag')]['total_cost'].iloc[0]
-            theirs = ref[ref.origin == idx]['total_cost'].iloc[0]
-            print(f'parity s_{idx}: mine {mine:.3f} baseline {theirs:.3f} '
-                  f'{"OK" if abs(mine-theirs) < 0.01 else "MISMATCH"}')
-    elif stage == 'panel':
-        configs = cfgs('logbag', 'lagrank_ovr', 'lagrank_b03', 'twostage',
-                       'ts_anchor', 'logbag_db245', 'logbag_arr', 'logbag_dark')
-        df = walk(table, ctx, list(ALL_ORIGINS), configs,
-                  save_plans=OUT / 'plans_panel.pkl')
-        df.to_csv(OUT / 'panel_results.csv', index=False)
-        summarize(df, 'X27 PANEL s_8..s_41 (tier-1 budget r0/i20)')
-        selection_stability(df)
-    elif stage == 'adrop':
-        configs = cfgs('logbag') + cfgs('logbag', planner='adrop')
-        df = walk(table, ctx, list(ALL_ORIGINS), configs)
-        df.to_csv(OUT / 'adrop_results.csv', index=False)
-        summarize(df, 'X27 ADROP s_8..s_41 (tier-1 budget)')
-    elif stage == 'noise':
-        configs = cfgs('logbag')
-        if not (OUT / 'plans_champion.pkl').exists():
-            walk(table, ctx, list(ALL_ORIGINS), configs,
-                 save_plans=OUT / 'plans_champion.pkl')
-        stage_noise(table, ctx, list(ALL_ORIGINS))
-    elif stage == 'policy':
-        stage_policy(table, ctx)
-    elif stage == 'full':
-        import json
-        names = json.loads(Path(OUT / 'finalists.json').read_text())
-        df = walk(table, ctx, list(ALL_ORIGINS), cfgs(*names),
-                  restarts=2, iters=60)
-        df.to_csv(OUT / 'full_results.csv', index=False)
-        summarize(df, 'X27 FULL s_8..s_41 (r2/i60)')
-        selection_stability(df)
-    print(f'stage {stage} done in {time.time()-t0:.0f}s', flush=True)
-
-
-if __name__ == '__main__':
-    main()

exp_x28.py

@@ -1,631 +0,0 @@
-"""
-Phase 28 experiment runner: candidates ported from the public Attention-Heads
-competitor repo (huggingface.co/Attention-Heads/BatterySwapAI2026-attention-heads,
-cloned 2026-08-19; rank 5, 1798 public). New file - frozen modules are imported,
-never edited; parallel-session protocol: md5-assert on import.
-
-What their SOLUTION.md establishes (measured on their fork, worth testing here):
-  - The official EOL comes from a temp-gated daily median + 7d rolling median,
-    so a real Li-MnO2 knee shows up in HOURLY mins / hours-below-threshold
-    counts days before the smoothed daily series moves. Misses of that shape
-    were ~30% of their false negatives; they fixed them with a planner-side
-    probability floor gated on an hourly "cliff" mask, not with a feature.
-  - Global prob_scale ~2.2 on the CDF used for include/park (timing kept raw)
-    minimised their cost (1.0 -> 2187, 2.2 -> ~2005, 3.0 -> 2068); per-battery
-    variant: knees full scale, "lingerers" (2.48-2.62 V, no hourly cliff)
-    capped low. Maps to our count/rank channels, NOT to timing curves.
-  - Miss-queue park pricing: the evaluator force-visits missed batteries one
-    per day, so the k-th miss is k extra days late; expected-queue term
-    p * 0.5 * max(E[misses]-1, 0) * LATE. (Distinct from phase-21's exact
-    emergency-day test, which moved the anchor for ALL parks; this term grows
-    with expected miss count only.)
-  - Peer offsets: device voltage gap vs building/room median normalises
-    unit-to-unit offsets.
-
-Candidates (panel = tier-1 budget r0/i20; deployment-facing ones ALSO go
-through the static foreign-fit holdout regime, per phases 23-26 - the panel's
-rolling native-fleet regime no longer matches what ships):
-  H1 logbag_hrly[2] - hourly knee features in the WHO classifier
-  H2 cliff_floor    - planner-facing p42 floor on hourly-cliff rows (bypasses
-                      the classifier; phase-27 showed crossing-flavoured
-                      FEATURES dilute - logbag_db245 lost 50)
-  H3 cliff_rank/count - per-battery appetite: rank_score/p42_sum scaled 2.2 on
-                      cliff rows only (count channel stays raw elsewhere)
-  H4 linger downweight - two-sided variant of H3 (lingerers 0.7)
-  H5 tsq            - miss-queue park penalty inside the ec top-up channel
-  H6 logbag_peer    - bldg/room voltage-gap features
-
-Stages (argv[1]): facts | smoke | tier0 | panel | holdout
-"""
-
-import hashlib
-import os
-import sys
-import time
-import warnings
-from pathlib import Path
-
-import numpy as np
-import pandas as pd
-
-from batteryswap_public.utils import load_dataset
-from batteryswap_public.evaluate import evaluate_plan
-
-import batteryswap_example.models_cv as mc
-from batteryswap_example.models_cv import (
-    FEATURES, SklearnRUL, LogBagFeats, training_pairs, load_context,
-    cost_curves, optimize_days, mobility_cost, plan_tsroute, assemble,
-    summarize, selection_stability, ALL_ORIGINS, EARLY, LATE)
-from batteryswap_example.features import (
-    load as load_features, WINDOW_DAYS, MIN_TEMP, MAX_TEMP)
-
-OUT = Path('/tmp/x28')
-OUT.mkdir(exist_ok=True)
-
-# frozen-module pins (parallel-session protocol; peers may commit while runs
-# are live - a silent checkout swap must hard-fail, not no-op)
-PINS = {
-    'batteryswap_example/models_cv.py': '7ead54afd3814aad4523c1405596f27b',
-    'batteryswap_example/features.py': 'b7437c5fe10b4f9842256a20ba8aa506',
-}
-
-CHAMPION = dict(model='logbag', planner='tsroute', p_star=0.8, day_q=0.05,
-                ec_margin=0.8, ec_pmax=25.0, split_days=7)
-ALPHA_STATIC = 1.5          # deployment appetite on the static held-out path
-HOLDOUT_ORIGINS = list(range(0, 8)) + [12, 16, 20, 24, 28, 32, 36, 40]
-
-# hourly cliff mask defaults; facts-stage grid picks these (fractions, not
-# absolute hour counts, so the rule is cadence-invariant)
-CLIFF = dict(frac_thr=0.35, v_thr=2.46, min_n=20)
-
-
-def module_md5s():
-    for mod, want in PINS.items():
-        h = hashlib.md5(Path(mod).read_bytes()).hexdigest()
-        assert h == want, f'FROZEN-FILE DRIFT: {mod} is {h}, pinned {want}'
-        print(f'md5 {mod} {h} OK', flush=True)
-    for attr in ('alpha', 'ec_margin', 'ec_pmax', 'split_days', 'floor_scale',
-                 'recent_deaths', 'p_be', 'floor_day', 'p_conf'):
-        assert hasattr(plan_tsroute, attr), attr
-
-
-# ------------------------------------------------------- x28 feature layer
-
-def build_x28(dataset_path=Path('dataset'), table=None):
-    """Hourly/unsmoothed trailing features per (device, origin) row.
-
-    All derived from raw readings <= origin: leak-free by construction (same
-    argument as features.py). Cached; keyed by the origin_idx set.
-    """
-    cache = OUT / 'x28_hourly.parquet'
-    if table is None:
-        table = load_features(dataset_path)
-    origins = table[['origin_idx', 'origin']].drop_duplicates()
-    if cache.exists():
-        x = pd.read_parquet(cache)
-        if set(origins['origin_idx']) <= set(x['origin_idx'].unique()):
-            return x
-    _, timeseries, _, _ = load_dataset(dataset_path / 'train')
-    df = timeseries.reset_index()
-    df['device_id'] = df['device_id'].astype(str)
-    df['day'] = df['end_time'].dt.normalize()
-    v = df['voltage']
-    df['b250'] = (v < 2.50).astype(np.int32)
-    df['b245'] = (v < 2.45).astype(np.int32)
-    df['b242'] = (v < 2.42).astype(np.int32)
-    agg = df.groupby(['device_id', 'day'], observed=True).agg(
-        umin=('voltage', 'min'), n=('voltage', 'size'),
-        n250=('b250', 'sum'), n245=('b245', 'sum'), n242=('b242', 'sum'))
-    gated = df[(df['temperature'] > MIN_TEMP) & (df['temperature'] < MAX_TEMP)]
-    gmin = gated.groupby(['device_id', 'day'], observed=True)['voltage'] \
-        .min().rename('gmin')
-    A = agg.join(gmin, how='left')
-
-    days = A.index.get_level_values('day')
-    cal = pd.date_range(days.min(),
-                        max(days.max(), origins['origin'].max().normalize()),
-                        freq='D')
-    P = {c: A[c].unstack(0).reindex(cal) for c in A.columns}
-    n7 = P['n'].fillna(0.0).rolling(7, min_periods=1).sum()
-    feats = {'tele_n7': n7}
-    for k in ('n250', 'n245', 'n242'):
-        s7 = P[k].fillna(0.0).rolling(7, min_periods=1).sum()
-        feats[f'frac{k[1:]}_7'] = s7 / n7.clip(lower=1.0)
-    feats['umin_m3'] = P['umin'].rolling(3, min_periods=1).min()
-    feats['umin_m7'] = P['umin'].rolling(7, min_periods=1).min()
-    gm7 = P['gmin'].rolling(7, min_periods=1).min()
-    feats['gmin_m7'] = gm7
-    feats['gmin_d7'] = gm7 - gm7.shift(7)
-
-    parts = []
-    for oi, ot in origins.itertuples(index=False):
-        d = pd.Timestamp(ot).normalize()
-        sub = pd.DataFrame({k: F.loc[d] for k, F in feats.items()})
-        sub.index.name = 'device_id'
-        sub['origin_idx'] = oi
-        parts.append(sub.reset_index())
-    x = pd.concat(parts, ignore_index=True)
-    x.to_parquet(cache)
-    return x
-
-
-def peer_features(table, locations):
-    """Voltage gap vs fresh (staleness<=14) building/room peers at the origin."""
-    loc = locations.set_index('battery')
-    t = table[['device_id', 'origin_idx', 'v_last', 'staleness']].copy()
-    t['bldg'] = t['device_id'].map(loc['building'])
-    t['room'] = t['device_id'].map(loc['room'])
-    fresh = t[t['staleness'] <= 14]
-    for key, pref in (('bldg', 'bldg'), ('room', 'room')):
-        g = fresh.groupby(['origin_idx', key])['v_last']
-        med = g.median().where(g.size() >= 3)
-        mi = pd.MultiIndex.from_frame(t[['origin_idx', key]])
-        t[f'{pref}_v_gap'] = t['v_last'].values - med.reindex(mi).values
-    return t[['device_id', 'origin_idx', 'bldg_v_gap', 'room_v_gap']]
-
-
-def load_table(dataset_path=Path('dataset')):
-    table = load_features(dataset_path)
-    table['v_proj'] = table['v_last'] + table['slope_lastk'] * table['staleness']
-    x28 = build_x28(dataset_path, table)
-    table = table.merge(x28, on=['device_id', 'origin_idx'], how='left')
-    locations, _, _, _ = load_dataset(dataset_path / 'train')
-    table = table.merge(peer_features(table, locations),
-                        on=['device_id', 'origin_idx'], how='left')
-    return table
-
-
-# --------------------------------------------------------------- wrappers
-
-def cliff_mask(rows, frac_thr=None, v_thr=None, min_n=None):
-    p = dict(CLIFF)
-    if frac_thr is not None:
-        p['frac_thr'] = frac_thr
-    if v_thr is not None:
-        p['v_thr'] = v_thr
-    if min_n is not None:
-        p['min_n'] = min_n
-    frac = rows['frac245_7'].fillna(0.0)
-    n = rows['tele_n7'].fillna(0.0)
-    v = rows['v_corr_last'].fillna(rows['v_last']).fillna(3.0)
-    return ((frac >= p['frac_thr']) & (n >= p['min_n'])
-            & (v < p['v_thr'])).values
-
-
-_LOGBAG_CACHE = {}
-
-
-def shared_logbag(train):
-    """One champion fit per training set (wrappers share it across configs)."""
-    key = (len(train), str(train['origin'].max()), int(train['dies42'].sum()))
-    if key not in _LOGBAG_CACHE:
-        _LOGBAG_CACHE[key] = SklearnRUL('logbag').fit(train)
-    return _LOGBAG_CACHE[key]
-
-
-class CliffFloor:
-    """H2: floor p42 on hourly-cliff rows. Planner-side override - moves both
-    the count channel (sum p42) and the ranking, without touching the
-    classifier (phase 27: crossing-flavoured features dilute the fit)."""
-
-    def __init__(self, floor=0.25, **kw):
-        self.floor = float(floor)
-        self.kw = kw
-
-    def fit(self, train):
-        self.base = shared_logbag(train)
-        return self
-
-    def predict(self, rows):
-        out = self.base.predict(rows)
-        m = cliff_mask(rows, **self.kw)
-        if m.any():
-            out.loc[m, 'p42'] = np.maximum(out.loc[m, 'p42'].values, self.floor)
-        return out
-
-
-class CliffScale:
-    """H3/H4: per-battery appetite. rank_score = p42 * s and (count mode)
-    p42_sum = sum(clip(p42 * s, 0, 1)); the raw p42 column is untouched, so
-    cost_curves timing stays honest (their scale_mode='park' lesson)."""
-
-    def __init__(self, scale=2.2, linger=None, count=True, **kw):
-        self.scale = float(scale)
-        self.linger = linger
-        self.count = bool(count)
-        self.kw = kw
-
-    def fit(self, train):
-        self.base = shared_logbag(train)
-        return self
-
-    def predict(self, rows):
-        out = self.base.predict(rows)
-        s = np.ones(len(rows))
-        m = cliff_mask(rows, **self.kw)
-        s[m] = self.scale
-        if self.linger is not None:
-            v = rows['v_corr_last'].fillna(rows['v_last']).fillna(3.0)
-            frac = rows['frac245_7'].fillna(0.0)
-            lg = ((v >= 2.47) & (v <= 2.60) & (frac < 0.05)
-                  & (rows['staleness'].fillna(99) <= 14)).values & ~m
-            s[lg] = float(self.linger)
-        scaled = np.clip(out['p42'].values * s, 0.0, 1.0)
-        out['rank_score'] = scaled
-        if self.count:
-            out['p42_sum'] = scaled.sum()
-        return out
-
-
-class CliffFloorPark:
-    """H2b: park-mode floor - floors the count/rank channels on cliff rows but
-    leaves the p42 column (and hence cost_curves timing) raw. AH measured this
-    split (scale_mode='park') as strictly better than scaling everything."""
-
-    def __init__(self, floor=0.50, **kw):
-        self.floor = float(floor)
-        self.kw = kw
-
-    def fit(self, train):
-        self.base = shared_logbag(train)
-        return self
-
-    def predict(self, rows):
-        out = self.base.predict(rows)
-        m = cliff_mask(rows, **self.kw)
-        boosted = out['p42'].values.copy()
-        boosted[m] = np.maximum(boosted[m], self.floor)
-        out['rank_score'] = boosted
-        out['p42_sum'] = boosted.sum()
-        return out
-
-
-X28_MODELS = {
-    'hrly': lambda: LogBagFeats(['frac245_7', 'umin_m7', 'gmin_d7']),
-    'hrly2': lambda: LogBagFeats(['frac250_7', 'frac245_7', 'frac242_7',
-                                  'umin_m7', 'gmin_m7', 'gmin_d7', 'tele_n7']),
-    'peer': lambda: LogBagFeats(['bldg_v_gap', 'room_v_gap']),
-    'cliff_floor25': lambda: CliffFloor(0.25),
-    'cliff_floor50': lambda: CliffFloor(0.50),
-    'cliff_rank22': lambda: CliffScale(2.2, count=False),
-    'cliff_count22': lambda: CliffScale(2.2, count=True),
-    'cliff_count_lg': lambda: CliffScale(2.2, linger=0.7, count=True),
-    'cliff_floor50p': lambda: CliffFloorPark(0.50),
-}
-
-
-def model_factory(name):
-    if name in X28_MODELS:
-        return X28_MODELS[name]
-    return mc.resolve_model(name)
-
-
-# ------------------------------------------------------------ H5 planner
-
-def plan_tsq(pred, included, locations, start_time, day_q,
-             travel_costs=None, settings=None):
-    """plan_tsroute + miss-queue park penalty: expected misses are force-
-    visited one per day, so the average one waits (E[misses]-1)/2 extra days
-    at the late rate. Raises park for every candidate the break-even top-up
-    prices, proportionally to how crowded the miss queue already is.
-    Knobs are read from plan_tsroute (set_champion_knobs wires them there)."""
-    P = plan_tsroute
-    psum = pred['p42_sum'].iloc[0] if 'p42_sum' in pred else pred['p42'].sum()
-    if P.floor_scale > 0:
-        psum = max(psum, P.floor_scale * P.recent_deaths)
-    n = int(round(P.alpha * psum))
-    order_col = 'rank_score' if 'rank_score' in pred.columns else 'p42'
-    chosen = pred[order_col].nlargest(n).index
-    if P.p_be > 0:
-        chosen = chosen.union(pred.index[pred['p42'] >= P.p_be])
-    curves, park = cost_curves(pred, locations, start_time)
-
-    p42 = pred['p42'].values
-    parked = ~pred.index.isin(chosen)
-    n_exp = float(p42[parked].sum())
-    park = park + p42 * 0.5 * max(n_exp - 1.0, 0.0) * LATE * plan_tsq.queue_mult
-
-    if P.ec_margin > 0:
-        end = locations.set_index('battery')['end_time'].dt.tz_localize(None)
-        e_far = ((end.reindex(pred.index) - start_time).dt.total_seconds()
-                 / 86400 + 30.0).values
-        fp_price = EARLY * (e_far - WINDOW_DAYS)
-        be = pred.index[(e_far >= WINDOW_DAYS)
-                        & (curves.min(axis=1) < P.ec_margin * park)
-                        & (fp_price <= P.ec_pmax)]
-        chosen = chosen.union(be)
-    if len(chosen) == 0:
-        return assemble(pd.Series(dtype='datetime64[ns]'), locations, start_time)
-
-    if P.floor_day > 0:
-        uncertain = (pred['p42'] < P.p_conf).values
-        curves[np.ix_(uncertain, np.arange(P.floor_day))] = np.inf
-    curves = pd.DataFrame(curves, index=pred.index).loc[chosen]
-    bldg = locations.set_index('battery')['building'].reindex(chosen)
-    room = locations.set_index('battery')['room']
-    key = bldg
-    if P.split_days > 0:
-        bucket = curves.values.argmin(axis=1) // int(P.split_days)
-        key = bldg.astype(str) + '~' + pd.Series(bucket, index=curves.index).astype(str)
-    groups = {g: sorted(m.index, key=lambda b: room.get(b, ''))
-              for g, m in curves.groupby(key)}
-    travel_hours = travel_costs.set_index(['from', 'to'])['hours'].unstack()
-    day_of = optimize_days(groups, curves, locations, travel_hours,
-                           settings, start_time)
-    offs = key.map(day_of).astype(float)
-    days = start_time + pd.to_timedelta(offs.values, unit='D')
-    return assemble(pd.Series(days, index=offs.index), locations, start_time)
-
-
-plan_tsq.queue_mult = 1.0
-
-
-# ------------------------------------------------------------------ runner
-
-LABEL_COLS = ['eol', 'dies42', 'days_to_eol', 'window_closed']
-
-
-def redact(rows):
-    return rows.drop(columns=[c for c in LABEL_COLS if c in rows.columns])
-
-
-def set_champion_knobs(cfg):
-    plan_tsroute.alpha = cfg.get('p_star', 0.8)
-    plan_tsroute.p_be = 0.0
-    plan_tsroute.floor_day = 0
-    plan_tsroute.p_conf = 0.5
-    plan_tsroute.ec_margin = cfg.get('ec_margin', 0.8)
-    plan_tsroute.ec_pmax = cfg.get('ec_pmax', 25.0)
-    plan_tsroute.split_days = cfg.get('split_days', 7)
-    plan_tsroute.floor_scale = 0.0
-    plan_tsroute.recent_deaths = 0.0
-    plan_tsq.queue_mult = cfg.get('queue_mult', 1.0)
-
-
-def walk(table, ctx, origins, configs, restarts=0, iters=20):
-    optimize_days.restarts = restarts
-    optimize_days.iters = iters
-    optimize_days.constructive = False
-    optimize_days.nn_order = False
-    fitted, records = {}, []
-    for idx in origins:
-        T = ctx[idx]['start_time']
-        rows = table[table['origin_idx'] == idx].set_index('device_id')
-        for cfg in configs:
-            m = cfg['model']
-            if (m, idx) not in fitted:
-                fitted[(m, idx)] = model_factory(m)().fit(training_pairs(table, T))
-            pred = fitted[(m, idx)].predict(redact(rows))
-            set_champion_knobs(cfg)
-            fn = plan_tsq if cfg.get('planner') == 'tsq' else plan_tsroute
-            plan = fn(pred, None, ctx[idx]['locs'], T, cfg.get('day_q', 0.05),
-                      travel_costs=ctx[idx]['travel_costs'],
-                      settings=ctx[idx]['settings'])
-            _, _, overall = evaluate_plan(
-                plan, ctx[idx]['locs'], ctx[idx]['travel_costs'],
-                ctx[idx]['settings'], eol_times=ctx[idx]['eol'],
-                start_time=T, verbose=0)
-            records.append(dict(overall, origin=idx, **cfg))
-        line = ' '.join(f"{r['name']}={r['total_cost']:.0f}"
-                        for r in records if r['origin'] == idx)
-        print(f's_{idx}: {line}', flush=True)
-    return pd.DataFrame(records)
-
-
-# ------------------------------------------------------------------- facts
-
-def stage_facts(table):
-    """Does the AH smoothing-lag claim hold on OUR data, and at what
-    thresholds is the cliff mask precise enough to gate appetite on?"""
-    real = table[table['origin_idx'] < 100]
-    panel = real[real['origin_idx'].isin(list(ALL_ORIGINS))]
-    print(f"rows: {len(panel)} panel, deaths {panel['dies42'].sum()}")
-    print(f"tele_n7 (readings/7d) median {panel['tele_n7'].median():.0f} "
-          f"p90 {panel['tele_n7'].quantile(.9):.0f}")
-
-    dying = panel[panel['dies42'] == 1]
-    alive = panel[panel['dies42'] == 0]
-    # the cohort the champion is blind to: dying but not yet crossed on the
-    # smoothed corrected daily median (days_below_245 unset at origin)
-    blind = dying[~(dying['days_below_245'] >= 0)]
-    print(f"dying rows {len(dying)}, of which pre-crossing (champion-blind) "
-          f"{len(blind)}")
-
-    print(f"\n{'frac':>5} {'v_thr':>5} | {'P(cliff|die)':>12} "
-          f"{'P(cliff|blind-die)':>18} {'alive flagged/origin':>20}")
-    for frac in (0.2, 0.35, 0.5, 0.65):
-        for v_thr in (2.44, 2.46, 2.50):
-            kw = dict(frac_thr=frac, v_thr=v_thr)
-            md = cliff_mask(dying, **kw).mean() if len(dying) else 0
-            mb = cliff_mask(blind, **kw).mean() if len(blind) else 0
-            fa = cliff_mask(alive, **kw).sum() / max(alive['origin_idx'].nunique(), 1)
-            print(f'{frac:5.2f} {v_thr:5.2f} | {md:12.2f} {mb:18.2f} {fa:20.2f}')
-
-    # per-death view at the chosen default: does the hourly min cross before
-    # the smoothed series at the origins where it matters?
-    m = cliff_mask(dying)
-    got = dying[m]
-    print(f"\ndefault mask {CLIFF}: catches {len(got)}/{len(dying)} dying rows; "
-          f"of the champion-blind {cliff_mask(blind).sum()}/{len(blind)}")
-    if len(got):
-        print("caught dying rows by p42-free stats: v_corr median "
-              f"{got['v_corr_last'].median():.3f}, staleness median "
-              f"{got['staleness'].median():.0f}d")
-
-
-# ------------------------------------------------------------------ tier 0
-
-def stage_tier0(table):
-    from sklearn.metrics import roc_auc_score
-    import json
-    base = json.loads(Path('research/baselines/tier0.json').read_text())
-    names = ['hrly', 'hrly2', 'peer', 'cliff_floor25', 'cliff_count22']
-    print('baseline :', {k: round(v, 3) for k, v in base.items()})
-    for name in names:
-        out = []
-        for idx in [10, 16, 22, 28, 34, 40]:
-            rows = table[table['origin_idx'] == idx].set_index('device_id')
-            T = rows['origin'].iloc[0]
-            t0 = time.time()
-            model = model_factory(name)().fit(training_pairs(table, T))
-            fit_s = time.time() - t0
-            p = model.predict(redact(rows))['p42']
-            y = rows['dies42']
-            crossed = rows['days_below_245'].notna() & (rows['days_below_245'] >= 0)
-            cy, cp = y[crossed], p[crossed]
-            dc = crossed & (y == 1)
-            out.append(dict(
-                fit_s=fit_s,
-                auc_all=roc_auc_score(y, p) if 0 < y.sum() < len(y) else np.nan,
-                auc_crossed=(roc_auc_score(cy, cp)
-                             if 0 < cy.sum() < len(cy) else np.nan),
-                count_err=abs(float(p.sum()) - int(y.sum())),
-                dying_crossed_mean_p=float(p[dc].mean()) if dc.any() else np.nan))
-        df = pd.DataFrame(out)
-        s = df.mean().to_dict()
-        gate = ((s['auc_crossed'] >= base['auc_crossed'] + 0.02
-                 or s['dying_crossed_mean_p'] >= 1.2 * base['dying_crossed_mean_p'])
-                and s['count_err'] <= 1.1 * base['count_err']
-                and df['fit_s'].max() <= 90)
-        print(f"{name:16s} auc_all {s['auc_all']:.3f} auc_crossed "
-              f"{s['auc_crossed']:.3f} count_err {s['count_err']:.2f} "
-              f"dying_p {s['dying_crossed_mean_p']:.3f} "
-              f"-> {'PROMOTE' if gate else 'reject (screen)'}", flush=True)
-
-
-# ----------------------------------------------------------------- holdout
-
-def stage_holdout(table, ctx, dataset_path=Path('dataset')):
-    """Static foreign-fit holdout (phases 24b/26 pattern): buildings split
-    into fleet A (packaged pool) / fleet B (target); ONE fit per model on A's
-    closed windows; plan and score B-only scenarios at 16 origins with the
-    deployment appetite ALPHA_STATIC. This is the regime public/private see."""
-    locations, _, _, _ = load_dataset(dataset_path / 'train')
-    bldg_of = locations.set_index('battery')['building']
-    buildings = sorted(bldg_of.unique())
-
-    configs = [dict(CHAMPION, p_star=ALPHA_STATIC, model=m, name=m)
-               for m in ('logbag', 'cliff_floor25', 'cliff_floor50',
-                         'cliff_count22', 'cliff_count_lg')]
-    configs.append(dict(CHAMPION, p_star=ALPHA_STATIC, model='logbag',
-                        planner='tsq', name='logbag+tsq'))
-
-    records = []
-    for seed, flip in ((0, False), (0, True), (1, False)):
-        rng = np.random.RandomState(seed)
-        half = rng.permutation(len(buildings))[:len(buildings) // 2]
-        A_bldg = {buildings[i] for i in half}
-        if flip:
-            A_bldg = set(buildings) - A_bldg
-        A_dev = set(bldg_of.index[bldg_of.isin(A_bldg)].astype(str))
-        pool = table[table['window_closed'] & table['device_id'].isin(A_dev)]
-        print(f"seed {seed} flip {flip}: pool {len(pool)} pairs, "
-              f"{pool[pool['dies42'] == 1]['device_id'].nunique()} dead devices",
-              flush=True)
-        _LOGBAG_CACHE.clear()
-        fitted = {c['model']: model_factory(c['model'])().fit(pool)
-                  for c in configs}
-        optimize_days.restarts = 0
-        optimize_days.iters = 20
-        for idx in HOLDOUT_ORIGINS:
-            T = ctx[idx]['start_time']
-            locs = ctx[idx]['locs']
-            locsB = locs[~locs['battery'].astype(str).isin(A_dev)].reset_index(drop=True)
-            eolB = ctx[idx]['eol']
-            eolB = eolB[eolB.index.isin(set(locsB['battery']))]
-            rows = table[(table['origin_idx'] == idx)
-                         & table['device_id'].isin(set(locsB['battery'].astype(str)))
-                         ].set_index('device_id')
-            for cfg in configs:
-                pred = fitted[cfg['model']].predict(redact(rows))
-                set_champion_knobs(cfg)
-                fn = plan_tsq if cfg.get('planner') == 'tsq' else plan_tsroute
-                plan = fn(pred, None, locsB, T, cfg.get('day_q', 0.05),
-                          travel_costs=ctx[idx]['travel_costs'],
-                          settings=ctx[idx]['settings'])
-                _, _, overall = evaluate_plan(
-                    plan, locsB, ctx[idx]['travel_costs'], ctx[idx]['settings'],
-                    eol_times=eolB, start_time=T, verbose=0)
-                records.append(dict(overall, origin=idx, seed=seed, flip=flip,
-                                    **cfg))
-            line = ' '.join(
-                f"{r['name']}={r['total_cost']:.0f}" for r in records
-                if r['origin'] == idx and r['seed'] == seed and r['flip'] == flip)
-            print(f'  s_{idx}: {line}', flush=True)
-    df = pd.DataFrame(records)
-    df.to_csv(OUT / 'holdout_results.csv', index=False)
-    print('\n=== X28 STATIC HOLDOUT (alpha=1.5, 16 origins x 3 fleet splits) ===')
-    g = df.groupby('name')
-    tab = g['total_cost'].mean().sort_values().to_frame('mean')
-    for c in ('late_swap', 'early_swap'):
-        tab[c] = g[c].mean()
-    ops = ['travel', 'overtime', 'daily_limit', 'weekly_limit',
-           'battery_swap', 'building_change', 'room_change']
-    tab['ops'] = g[ops].mean().sum(axis=1)
-    print(tab.round(1).to_string())
-    base = df[df['name'] == 'logbag'].set_index(['seed', 'flip', 'origin'])['total_cost']
-    for name in df['name'].unique():
-        if name == 'logbag':
-            continue
-        c = df[df['name'] == name].set_index(['seed', 'flip', 'origin'])['total_cost']
-        d = (c - base).dropna()
-        print(f"{name:16s} delta mean {d.mean():+8.1f}  "
-              f"wins {(d < -0.5).sum()}/{len(d)}  worst {d.max():+.1f}")
-    return df
-
-
-# ------------------------------------------------------------------- main
-
-def main():
-    warnings.filterwarnings('ignore')
-    stage = sys.argv[1] if len(sys.argv) > 1 else 'smoke'
-    module_md5s()
-    t0 = time.time()
-    dataset_path = Path(os.environ.get('BATTERYSWAP_DATASET_PATH', 'dataset'))
-    table = load_table(dataset_path)
-    print(f'table loaded in {time.time()-t0:.0f}s', flush=True)
-
-    def cfgs(*names, planner='tsroute'):
-        return [dict(CHAMPION, model=n, planner=planner,
-                     name=(n if planner == 'tsroute' else f'{n}+{planner}'))
-                for n in names]
-
-    if stage == 'facts':
-        stage_facts(table)
-        return
-    if stage == 'tier0':
-        stage_tier0(table)
-        return
-
-    ctx = load_context(dataset_path)
-    print(f'ctx loaded in {time.time()-t0:.0f}s', flush=True)
-
-    if stage == 'smoke':
-        df = walk(table, ctx, [8, 9], cfgs('logbag', 'cliff_floor25')
-                  + cfgs('logbag', planner='tsq'))
-        ref = pd.read_csv('research/baselines/tier1_champion.csv')
-        for idx in (8, 9):
-            mine = df[(df.origin == idx) & (df.name == 'logbag')]['total_cost'].iloc[0]
-            theirs = ref[ref.origin == idx]['total_cost'].iloc[0]
-            print(f'parity s_{idx}: mine {mine:.3f} baseline {theirs:.3f} '
-                  f'{"OK" if abs(mine - theirs) < 0.01 else "MISMATCH"}')
-    elif stage == 'panel':
-        # hrly/hrly2/peer rejected at the tier-0 screen (2026-08-19): crossed
-        # AUC unmoved - our un-rolled daily medians never had the smoothing
-        # lag their hourly features fix. Cliff/scale/tsq candidates work the
-        # rank/count/park channels, which tier-0 cannot see - panel + holdout
-        # adjudicate them directly.
-        configs = cfgs('logbag', 'cliff_floor25', 'cliff_floor50',
-                       'cliff_rank22', 'cliff_count22', 'cliff_count_lg') \
-            + cfgs('logbag', planner='tsq')
-        df = walk(table, ctx, list(ALL_ORIGINS), configs)
-        df.to_csv(OUT / 'panel_results.csv', index=False)
-        summarize(df, 'X28 PANEL s_8..s_41 (tier-1 budget r0/i20)')
-        selection_stability(df)
-    elif stage == 'holdout':
-        stage_holdout(table, ctx, dataset_path)
-    print(f'stage {stage} done in {time.time()-t0:.0f}s', flush=True)
-
-
-if __name__ == '__main__':
-    main()

exp_x29_new_signal_sweep.py

@@ -1,214 +0,0 @@
-"""Phase-49 sweep: four new ordering-signal families, all REJECTED.
-
-Usage: venv/bin/python exp_x29_new_signal_sweep.py [long|tau|tavg]
-
-  long  A. LONGITUDINAL cross-origin features - the per-device history ACROSS
-           the 168-origin snapshot grid (~3-4d apart), untouched by any prior
-           phase. mom28/mom56 = fleet-rank change vs 28/56d ago; wolf = count
-           of snapshots >=42d ago in the bottom decile, survived to T.
-        B. INTERACTIONS - within-origin z-score products (the champion is
-           linear with zero interaction terms).
-  tau   C. Descent-band tau_projected re-rank at fixed N (the phase-43 knn
-           screen's "open deployment-regime candidate", never tested as a
-           pure reorder).
-  tavg  D. Temporal score ensembling - rank-average the current fit with the
-           previous 1-2 origins' fits applied to current rows (fit-noise
-           smoothing; effective n ~80 deaths).
-
-Verdicts (fixed-N deaths caught, paired per origin, bootstrap 95% CI):
-  A: wolf -24 [-42,-7] native N=12, -14 [-27,-1] gate; momwolf/allnew worse;
-     mom -5 [-13,+3] native, significantly negative gate N=16/20. The raw
-     signal EXISTS (bottom-decile chronics die 20.7% vs 15.5%; dying devices'
-     rank slides -0.025 median in 28d) but the 30d-window slope/drop features
-     already price it in - the rank version adds noise, not information.
-  B: inter flat everywhere (+2/-2 native, +0/-1 gate) - no exploitable
-     interaction structure; the phase-45 linear ceiling again.
-  C: native -45 of 454 at N=12 (7W/29L) - decisive. Gate mildly positive
-     (+3..+11), the same protocol-specificity signature as v_last in phase
-     41: the weekly refit already extracts the timer. Do not trust the gate's
-     sign against a native loss.
-  D: dead flat (tavg2 -3 at N=12, 0W/3L; tavg3 -4, 1W/5L) - adjacent 7d-apart
-     refits produce near-identical orderings. Third leg of the
-     variance-vs-information triangulation (with phase 44's flat C sweep and
-     logbag==logistic): the marginal-rank defect is MISSING INFORMATION, not
-     estimation noise. Any future ordering candidate must bring a NEW signal,
-     not a better estimate of the current ones.
-"""
-import sys
-
-import numpy as np
-import pandas as pd
-
-from batteryswap_example import cv
-from batteryswap_example import models_cv as mc
-
-NS = (8, 12, 16, 20, 30)
-
-
-# ---------------------------------------------------------------- A + B ----
-
-def _ensure_longitudinal(t):
-    if 'mom28' in t.columns:
-        return
-    snap = (t[['device_id', 'origin', 'r_vlast']]
-            .sort_values('origin').reset_index(drop=True))
-    snap['danger'] = (snap['r_vlast'] < 0.10).astype(float)
-    snap = snap.sort_values(['device_id', 'origin'])
-    snap['cum_danger'] = snap.groupby('device_id')['danger'].cumsum()
-    snap = snap.sort_values('origin')
-    base = t[['device_id', 'origin']].copy()
-    base['_row'] = np.arange(len(base))
-
-    def asof(offset_days, col, tol):
-        left = base.sort_values('origin').copy()
-        left['key'] = left['origin'] - pd.Timedelta(days=offset_days)
-        left = left.sort_values('key')
-        m = pd.merge_asof(left, snap[['device_id', 'origin', col]]
-                          .rename(columns={'origin': 'snap_origin'})
-                          .sort_values('snap_origin'),
-                          left_on='key', right_on='snap_origin',
-                          by='device_id', direction='backward',
-                          tolerance=pd.Timedelta(days=tol))
-        return m.sort_values('_row')[col].values
-
-    r28 = asof(28, 'r_vlast', 10)
-    r56 = asof(56, 'r_vlast', 10)
-    cur = t['r_vlast'].values
-    t['mom28'] = np.where(np.isnan(r28), 0.0, cur - r28)
-    t['mom56'] = np.where(np.isnan(r56), 0.0, cur - r56)
-    cd = asof(42, 'cum_danger', 400)
-    t['wolf'] = np.where(np.isnan(cd), 0.0, cd)
-
-
-def _ensure_inter(t):
-    if 'ix_v_slope' in t.columns:
-        return
-
-    def z(col):
-        g = t.groupby('origin_idx')[col]
-        return (t[col] - g.transform('mean')) / g.transform('std').replace(0, 1)
-
-    zv, zs = z('v_last'), z('slope_30')
-    zm7, za = z('v_min_7'), z('accel')
-    zvs = z('vstd_30')
-    t['ix_v_slope'] = (zv * zs).clip(-30, 30).fillna(0)
-    t['ix_v_vstd'] = (zv * zvs).clip(-30, 30).fillna(0)
-    t['ix_m7_accel'] = (zm7 * za).clip(-30, 30).fillna(0)
-
-
-LONG_VARIANTS = {
-    'mom': lambda t: cv.BASE_FEATURES + ['mom28', 'mom56'],
-    'wolf': lambda t: cv.BASE_FEATURES + ['wolf'],
-    'momwolf': lambda t: cv.BASE_FEATURES + ['mom28', 'mom56', 'wolf'],
-    'inter': lambda t: cv.BASE_FEATURES + ['ix_v_slope', 'ix_v_vstd',
-                                           'ix_m7_accel'],
-    'allnew': lambda t: cv.BASE_FEATURES + ['mom28', 'mom56', 'wolf',
-                                            'ix_v_slope', 'ix_v_vstd',
-                                            'ix_m7_accel'],
-}
-
-
-def run_long():
-    cv.VARIANTS.update(LONG_VARIANTS)
-    # ablate() slices training_pairs BEFORE the variant loop, so the columns
-    # must exist at load time (same latent limitation bites the zone variant)
-    orig = cv.load_table
-
-    def load_plus(dataset_path=cv.Path('dataset')):
-        t = orig(dataset_path)
-        _ensure_longitudinal(t)
-        _ensure_inter(t)
-        return t
-
-    cv.load_table = load_plus
-    var = ('base',) + tuple(LONG_VARIANTS)
-    df, ns = cv.ablate(variants=var, protocol='all48')
-    df.to_csv('/tmp/x29_long_all48.csv', index=False)
-    cv.ablation_report(df, ns, 'native all48 / longitudinal + interactions')
-    df2, ns2 = cv.ablate(variants=var, protocol='static')
-    df2.to_csv('/tmp/x29_long_static.csv', index=False)
-    cv.ablation_report(df2, ns2, 'static gate / longitudinal + interactions')
-
-
-# -------------------------------------------------------------------- C ----
-
-def _tau_transforms(rows, p42):
-    s = pd.DataFrame({'p42': p42})
-    s['rp'] = s['p42'].rank(pct=True)
-    tau = rows['tau_projected'].reindex(s.index)
-    vc = rows['v_corr_last'].reindex(s.index)
-    rt = (-tau).rank(pct=True)
-    band = (vc < 2.55) & tau.notna()
-    out = {'base': s['p42']}
-    b = s['rp'].copy()
-    b[band] = 0.5 * s['rp'][band] + 0.5 * rt[band]
-    out['blend_b'] = b
-    a = s['rp'].copy()
-    m = tau.notna()
-    a[m] = 0.5 * s['rp'][m] + 0.5 * rt[m]
-    out['blend_all'] = a
-    p = s['rp'].copy()
-    p[band] = np.maximum(s['rp'][band], rt[band])
-    out['promote'] = p
-    return out
-
-
-def _native_scores(score_fn):
-    """Shared walk: fit champion per origin, emit _order_stats per variant."""
-    t = cv.load_table()
-    ctx = mc.load_context(cv.Path('dataset'))
-    mc.FEATURES = list(cv.BASE_FEATURES)
-    recs, models = [], {}
-    for idx in range(48):
-        c = ctx[idx]
-        T = c['start_time']
-        tr = mc.training_pairs(t, T)
-        if tr['dies42'].sum() < 3:
-            continue
-        models[idx] = mc.SklearnRUL(cv.CHAMPION['model']).fit(tr)
-        rows = (t[t['origin_idx'] == idx].set_index('device_id')
-                .reindex(c['locs']['battery']).dropna(subset=['v_last']))
-        eol = c['eol'].reindex(rows.index)
-        deaths = set(eol.index[(eol > T)
-                               & (eol <= T + pd.Timedelta(days=42))])
-        if not deaths:
-            continue
-        for name, score in score_fn(models, idx, rows).items():
-            recs.append(dict(cv._order_stats(score, deaths, NS),
-                             origin=idx, variant=name, seed=0))
-        print(f's_{idx} done', file=sys.stderr, flush=True)
-    return pd.DataFrame(recs)
-
-
-def run_tau():
-    def fn(models, idx, rows):
-        p = models[idx].predict(rows)['p42']
-        return _tau_transforms(rows, p)
-
-    df = _native_scores(fn)
-    df.to_csv('/tmp/x29_tau_all48.csv', index=False)
-    cv.ablation_report(df, NS, 'native all48 / tau descent-band re-rank')
-
-
-# -------------------------------------------------------------------- D ----
-
-def run_tavg():
-    def fn(models, idx, rows):
-        p_now = models[idx].predict(rows)['p42']
-        r_now = p_now.rank(pct=True)
-        scores = {'base': p_now}
-        prev = [models[j].predict(rows)['p42'].rank(pct=True)
-                for j in (idx - 1, idx - 2) if j in models]
-        if len(prev) >= 1:
-            scores['tavg2'] = 0.6 * r_now + 0.4 * prev[0]
-        if len(prev) >= 2:
-            scores['tavg3'] = 0.5 * r_now + 0.3 * prev[0] + 0.2 * prev[1]
-        return scores
-
-    df = _native_scores(fn)
-    df.to_csv('/tmp/x29_tavg_all48.csv', index=False)
-    cv.ablation_report(df, NS, 'native all48 / temporal score ensembling')
-
-
-if __name__ == '__main__':
-    {'long': run_long, 'tau': run_tau, 'tavg': run_tavg}[sys.argv[1]]()

exp_x30_dynamics_forecast.py

@@ -1,259 +0,0 @@
-"""Fleet-learned conditional dynamics + MC crossing forecast (the forecast axis).
-
-Usage: venv/bin/python exp_x30_dynamics_forecast.py [screen|sweep|harness|fuel]
-
-CONSTRUCTION. One-step (7d) voltage forecaster - HGBR on state (level, recent
-slope, v_min_30, vstd_30, accel, below_frac, staleness, n_30, tau, age) -
-trained on ~70k UNLABELED device-week transitions from the 168-origin grid,
-rolled out 6 weeks with empirical heteroscedastic residuals (HGBR sigma model
-+ standardized-residual pool, so plunge tails survive the sampling).
-p_dyn = P(path crosses 2.40 V corrected within 42d).
-
-Threads the phase-49 standing wall by construction: auxiliary supervision
-(not the ~80 death labels), fleet-pooled NONLINEAR conditional drift (not
-WienerFHT's noisy per-device linear drift, phase 9), evaluated at fixed N so
-calibration never enters. Strictly backward-looking: dynamics AND residual
-pool refit per origin on transitions ending <= T.
-
-GHOST GUARD (cert-first public lesson, phase 48): devices at or below 2.40
-keep the base rank/probability untouched. NOT a local no-op: train has 342
-below-threshold rows at scenario origins of which 236 are never-labeled (the
-QUARANTINE comment's 26 ghost trajectories) - and public has more.
-
-EVIDENCE (fixed-N, paired per origin, bootstrap 95% CI, ghost-guarded):
-- native: +11 at N=16 (13W/6L), +13 [+5,+22] at N=20 (12W/1L)
-- gate:   +24 [+11,+38] at N=8, +23 [+11,+36] at N=12 (28W/7L)
-- weight plateau: all of w=0.1..0.5 positive in 39/40 cells (both instruments)
-- tail diagnostic: of 104 champion misses at ranks 13-30 (the phase-44 gap
-  population), p_dyn promotes 30 into its own top-12
-- fuel: descent-band (2.40-2.60) AUC 0.8005 vs v_last 0.8024 on 891
-  positives, Spearman(p_dyn, v_last) -0.683 - real independent variation
-- runtime: fit 2.7s + rollout 1.3s at 2x fleet vs 1800s/split budget
-
-COST HARNESS (the verdict instrument): three post-hoc deliveries, count
-pinned in all three, monotone improvement, never crosses zero (1661.7 scale):
-  rank_score channel        1744.5 (+83)  inclusion disagrees with placement
-  global p42 quantile-blend 1717.1 (+55)  every device's placement perturbed
-  band-limited (ranks 9-30) 1690.8 (+29)  catches AT PARITY, +13 late timing
-Post-mortem: post-hoc blends produce beliefs inconsistent with the machinery
-that prices count and timing; promoted deaths' rank gains never become landed
-swaps because the day model schedules modest-p42 devices late and they die
-first. The non-post-hoc delivery (p_dyn as a 14th feature inside the label
-model + day quantiles, so everything moves together) is the open arm.
-"""
-import sys
-import time
-
-import numpy as np
-import pandas as pd
-from sklearn.ensemble import HistGradientBoostingRegressor
-
-from batteryswap_example import cv
-from batteryswap_example import models_cv as mc
-
-THR = 2.40
-STEPS, PATHS = 6, 200
-NS = (8, 12, 16, 20, 30)
-DYN_FROZEN = ['v_min_30', 'vstd_30', 'accel', 'below_frac', 'staleness',
-              'n_30', 'tau_c', 'age_days']
-RNG = np.random.default_rng(7)
-
-
-def make_transitions(t):
-    cols = ['device_id', 'origin', 'v_corr_last', 'slope_lastk'] + [
-        c for c in DYN_FROZEN if c != 'tau_c'] + ['tau_projected']
-    s = t[cols].dropna(subset=['v_corr_last']).copy()
-    s['tau_c'] = s['tau_projected'].clip(-400, 400)
-    s = s.sort_values('origin')
-    end = s[['device_id', 'origin', 'v_corr_last']].rename(
-        columns={'origin': 'o_end', 'v_corr_last': 'v_end'})
-    s2 = s.copy()
-    s2['key'] = s2['origin'] + pd.Timedelta(days=7)
-    m = pd.merge_asof(s2.sort_values('key'), end.sort_values('o_end'),
-                      left_on='key', right_on='o_end', by='device_id',
-                      direction='nearest', tolerance=pd.Timedelta(days=2))
-    m = m.dropna(subset=['v_end'])
-    m = m[m['o_end'] > m['origin']]
-    gap = (m['o_end'] - m['origin']).dt.total_seconds() / 86400
-    m['dv'] = (m['v_end'] - m['v_corr_last']) * 7.0 / gap
-    return m
-
-
-def fit_dynamics(trans):
-    X = np.column_stack([trans['v_corr_last'].values,
-                         trans['slope_lastk'].values] +
-                        [trans[c].values for c in DYN_FROZEN])
-    # HGBR's binner crashes on all-NaN/constant columns (tau_c is all-NaN
-    # before the first descent appears) - mask them per fit
-    mask = np.array([len(np.unique(X[:, j][~np.isnan(X[:, j])])) >= 2
-                     for j in range(X.shape[1])])
-    X = X[:, mask]
-    y = trans['dv'].values
-    mu = HistGradientBoostingRegressor(max_iter=150, max_depth=4,
-                                       random_state=0).fit(X, y)
-    r = y - mu.predict(X)
-    sg = HistGradientBoostingRegressor(max_iter=100, max_depth=4,
-                                       random_state=0,
-                                       loss='absolute_error').fit(
-        X, np.log(np.abs(r) + 1e-4))
-    sig = np.exp(sg.predict(X))
-    z = np.clip(r / np.maximum(sig, 1e-4), -12, 12)
-    return mu, sg, z, mask
-
-
-def rollout(rows, mu, sg, zpool, mask):
-    rows = rows.copy()
-    rows['tau_c'] = rows['tau_projected'].clip(-400, 400)
-    n = len(rows)
-    v = np.repeat(rows['v_corr_last'].values, PATHS)
-    s7 = np.repeat(rows['slope_lastk'].fillna(0).values, PATHS)
-    fro = np.repeat(rows[DYN_FROZEN].values, PATHS, axis=0)
-    dead = v <= THR
-    for _ in range(STEPS):
-        alive = ~dead
-        if not alive.any():
-            break
-        X = np.column_stack([v[alive], s7[alive], fro[alive]])[:, mask]
-        m = mu.predict(X)
-        s = np.exp(sg.predict(X))
-        z = RNG.choice(zpool, size=alive.sum())
-        dv = m + s * z
-        v[alive] = v[alive] + dv
-        s7[alive] = dv / 7.0
-        dead = dead | (v <= THR)
-    return dead.reshape(n, PATHS).mean(axis=1)
-
-
-def native_pdyn(t, trans, ctx):
-    """Rolling per-origin dynamics scores for the 48 scenario origins."""
-    out = {}
-    for idx in range(48):
-        T = ctx[idx]['start_time']
-        rows = (t[t['origin_idx'] == idx].set_index('device_id')
-                .reindex(ctx[idx]['locs']['battery'])
-                .dropna(subset=['v_last']))
-        mu, sg, zpool, mask = fit_dynamics(trans[trans['o_end'] <= T])
-        out[idx] = pd.Series(rollout(rows, mu, sg, zpool, mask),
-                             index=rows.index)
-        print(f'dyn s_{idx}', file=sys.stderr, flush=True)
-    return out
-
-
-def guarded_blend(p42, dyn, vcorr, w):
-    rp = p42.rank(pct=True)
-    rd = dyn.reindex(rp.index).rank(pct=True)
-    b = (1 - w) * rp + w * rd
-    below = (vcorr.reindex(rp.index) <= THR).fillna(False)
-    b[below] = rp[below]
-    return b
-
-
-def run_screen():
-    t = cv.load_table()
-    trans = make_transitions(t)
-    ctx = mc.load_context(cv.Path('dataset'))
-    mc.FEATURES = list(cv.BASE_FEATURES)
-    pdyn = native_pdyn(t, trans, ctx)
-    pd.concat([pd.DataFrame({'device_id': s.index, 'origin_idx': i,
-                             'p_dyn': s.values})
-               for i, s in pdyn.items()]).to_parquet(
-        '/tmp/pdyn_native.parquet', index=False)
-    recs, diag = [], []
-    for idx in range(48):
-        T = ctx[idx]['start_time']
-        tr = mc.training_pairs(t, T)
-        if tr['dies42'].sum() < 3:
-            continue
-        rows = (t[t['origin_idx'] == idx].set_index('device_id')
-                .reindex(ctx[idx]['locs']['battery'])
-                .dropna(subset=['v_last']))
-        eol = ctx[idx]['eol'].reindex(rows.index)
-        deaths = set(eol.index[(eol > T)
-                               & (eol <= T + pd.Timedelta(days=42))])
-        if not deaths:
-            continue
-        p = mc.SklearnRUL(cv.CHAMPION['model']).fit(tr).predict(rows)['p42']
-        scores = {'base': p, 'dyn_alone': pdyn[idx],
-                  'blend37': guarded_blend(p, pdyn[idx],
-                                           rows['v_corr_last'], 0.3)}
-        for name, score in scores.items():
-            recs.append(dict(cv._order_stats(score, deaths, NS),
-                             origin=idx, variant=name, seed=0))
-        order_p = p.sort_values(ascending=False)
-        rank_p = {b: r for r, b in enumerate(order_p.index, start=1)}
-        order_d = pdyn[idx].sort_values(ascending=False)
-        rank_d = {b: r for r, b in enumerate(order_d.index, start=1)}
-        diag += [{'origin': idx, 'rank_p42': rank_p[b],
-                  'rank_dyn': rank_d.get(b, 999)}
-                 for b in deaths if b in rank_p and 13 <= rank_p[b] <= 30]
-    df = pd.DataFrame(recs)
-    df.to_csv('/tmp/cv_dyn_all48.csv', index=False)
-    cv.ablation_report(df, NS, 'native all48 / dynamics MC crossing')
-    d = pd.DataFrame(diag)
-    print(f'\nTail diagnostic (champion misses at 13-30, n={len(d)}): '
-          f'median rank p42 {d["rank_p42"].median()} -> dyn '
-          f'{d["rank_dyn"].median()}; promoted into top-12: '
-          f'{int((d["rank_dyn"] <= 12).sum())}')
-
-
-def run_harness():
-    """Cost harness with one of the three post-hoc deliveries.
-
-    argv[2]: rank_score | pblend | band (default band, the best of the three).
-    All pin the count via p42_sum; all ghost-guard below-THR devices.
-    """
-    delivery = sys.argv[2] if len(sys.argv) > 2 else 'band'
-    t = cv.load_table()
-    ctx = mc.load_context(cv.Path('dataset'))
-    mc.FEATURES = list(cv.BASE_FEATURES)
-    trans = make_transitions(t)
-    pdyn = native_pdyn(t, trans, ctx)
-    vlow = {i: t[t['origin_idx'] == i].set_index('device_id')['v_corr_last']
-            for i in range(48)}
-    orig_predict = mc.SklearnRUL.predict
-
-    def patched(self, rows):
-        out = orig_predict(self, rows)
-        if 'origin_idx' not in rows.columns:
-            return out
-        idx = int(rows['origin_idx'].iloc[0])
-        if idx not in pdyn:
-            return out
-        rd = pdyn[idx].reindex(out.index)
-        if rd.notna().mean() <= 0.9:
-            return out
-        p42 = out['p42']
-        below = (vlow[idx].reindex(out.index) <= THR).fillna(False)
-        if delivery == 'rank_score':
-            rp = p42.rank(pct=True)
-            rs = 0.7 * rp + 0.3 * rd.rank(pct=True)
-            rs[below] = rp[below]
-            out['rank_score'] = rs
-            return out
-        out['p42_sum'] = float(p42.sum())
-        rdr = rd.rank(pct=True)
-        vals = np.sort(p42.values)
-        dyn_cal = pd.Series(
-            np.quantile(vals, rdr.clip(0, 1).fillna(0.5)), index=out.index)
-        p_new = p42.copy()
-        if delivery == 'pblend':
-            p_new = 0.7 * p42 + 0.3 * dyn_cal
-        else:
-            base_rank = p42.rank(ascending=False, method='first')
-            band = (base_rank >= 9) & (base_rank <= 30)
-            p_new[band] = 0.7 * p42[band] + 0.3 * dyn_cal[band]
-        p_new[below] = p42[below]
-        p_new[rd.isna()] = p42[rd.isna()]
-        out['p42'] = p_new
-        return out
-
-    mc.SklearnRUL.predict = patched
-    df = mc.walk_forward(t, ctx, range(48), [dict(cv.CHAMPION)])
-    mc.SklearnRUL.predict = orig_predict
-    df.to_csv(f'/tmp/cv_all48_dyn_{delivery}.csv', index=False)
-    cv.scorecard(df, f'all48 / dynamics delivery={delivery} (1661.7 scale)')
-
-
-if __name__ == '__main__':
-    mode = sys.argv[1] if len(sys.argv) > 1 else 'screen'
-    {'screen': run_screen, 'harness': run_harness}[mode]()

research/baselines/sat_champion.csv

@@ -1,35 +0,0 @@
-early_swap,room_change,travel,daily_limit,weekly_limit,late_swap,overtime,battery_swap,building_change,total_cost,origin,model,planner,p_star,day_q,ec_margin,ec_pmax,split_days,restarts,iters
-247.0,7.0,30.651666666666667,100.0,200.0,920.0,87.48633333333332,4.0,11.0,1607.138,8,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-974.5,7.0,22.201833333333337,100.0,100.0,300.0,54.921,4.5,11.0,1574.1228333333331,9,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-878.5,6.5,23.006,200.0,100.0,70.0,80.95666666666668,4.5,11.0,1374.4626666666668,10,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-559.0,6.5,30.13566666666667,100.0,100.0,430.0,90.86866666666666,4.0,12.0,1332.5043333333333,11,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-760.5,9.0,27.572666666666663,100.0,200.0,1590.0,67.52133333333333,4.75,14.0,2773.344,12,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-838.0,8.0,48.00216666666667,300.0,200.0,1670.0,141.83366666666672,5.25,15.0,3226.085833333333,13,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-608.5,6.5,29.65716666666666,200.0,100.0,2010.0,85.03033333333332,4.25,12.0,3055.9375,14,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-408.0,8.0,23.285,200.0,200.0,3170.0,62.32166666666667,4.25,15.0,4090.856666666667,15,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-276.5,7.5,36.275666666666666,300.0,300.0,2570.0,92.98100000000001,4.25,11.0,3598.506666666667,16,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-305.0,7.0,22.120166666666666,100.0,200.0,1630.0,52.26,4.5,10.0,2330.880166666667,17,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-531.0,7.5,26.513666666666673,100.0,200.0,1820.0,68.977,4.25,12.0,2770.2406666666666,18,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-590.5,6.0,22.385,100.0,200.0,1130.0,54.614,3.5,10.0,2116.9990000000003,19,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-228.5,5.0,22.90316666666667,100.0,200.0,1050.0,55.512,2.5,9.0,1673.4151666666667,20,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-347.5,6.5,118.16833333333332,800.0,300.0,1400.0,345.92866666666663,3.25,11.0,3332.347,21,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-534.5,6.0,36.557166666666674,400.0,200.0,720.0,108.80499999999999,3.5,10.0,2019.3621666666666,22,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-540.5,6.0,105.01983333333334,500.0,400.0,550.0,291.10633333333334,3.75,10.0,2406.376166666667,23,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-532.0,6.0,37.39416666666666,500.0,300.0,750.0,116.34033333333332,3.75,9.0,2254.4845,24,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-588.0,6.5,40.301166666666674,500.0,300.0,1670.0,123.66333333333333,4.0,10.0,3242.4645,25,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-448.0,5.5,29.703666666666667,200.0,300.0,2060.0,84.374,3.5,8.0,3139.0776666666666,26,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-490.0,5.5,7.145333333333333,100.0,100.0,1400.0,16.353333333333335,3.5,10.0,2132.4986666666664,27,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-298.0,4.5,24.977,100.0,100.0,790.0,51.730999999999995,3.0,8.0,1380.2079999999999,28,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-209.0,4.5,24.928333333333335,100.0,100.0,440.0,50.654999999999994,2.5,8.0,939.5833333333334,29,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-309.5,5.0,25.614666666666665,200.0,200.0,310.0,85.22933333333333,2.5,6.0,1143.844,30,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-257.0,5.5,44.781666666666666,200.0,200.0,110.0,129.325,3.0,9.0,958.6066666666666,31,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-307.5,7.5,57.19266666666666,500.0,300.0,620.0,150.16833333333335,3.75,12.0,1958.111,32,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-176.0,5.0,61.70300000000001,300.0,300.0,280.0,142.93533333333335,2.75,10.0,1278.3883333333333,33,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-208.5,5.0,71.39816666666667,500.0,300.0,800.0,192.601,3.25,10.0,2090.7491666666665,34,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-180.5,6.0,72.70333333333333,500.0,300.0,840.0,187.08933333333331,3.25,11.0,2100.5426666666667,35,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-254.5,6.5,71.042,500.0,300.0,1210.0,181.26666666666668,3.5,9.0,2535.808666666667,36,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-345.0,6.0,50.5915,300.0,300.0,780.0,127.45500000000001,3.25,10.0,1922.2965,37,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-220.0,4.5,53.50816666666667,300.0,200.0,770.0,136.04466666666667,2.5,7.0,1693.5528333333332,38,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-521.0,9.0,32.00716666666667,100.0,100.0,380.0,79.01566666666668,4.75,12.0,1237.7728333333334,39,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-536.5,9.5,35.863666666666674,100.0,200.0,220.0,88.156,5.25,13.0,1208.2696666666666,40,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-644.0,8.0,37.8015,200.0,100.0,300.0,108.05699999999999,6.0,10.0,1413.8585,41,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60

research/baselines/tier0.json

@@ -1,8 +0,0 @@
-{
- "auc_all": 0.930567841842267,
- "auc_crossed": 0.8021138140264079,
- "count_err": 5.702596246648099,
- "crossed_count_err": 2.170831477801592,
- "dying_crossed_mean_p": 0.45313844808967824,
- "max_fit_s": 2.1
-}

research/baselines/tier1_champion.csv

@@ -1,35 +0,0 @@
-building_change,weekly_limit,early_swap,room_change,late_swap,overtime,daily_limit,travel,battery_swap,total_cost,origin,planner,p_star,day_q,ec_margin,ec_pmax,split_days,model,restarts,iters
-12.0,100.0,237.5,7.5,1060.0,24.84533333333333,0.0,32.25083333333333,4.0,1478.0961666666667,8,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,0.0,978.0,7.0,300.0,23.703000000000003,0.0,22.964000000000002,4.5,1347.1670000000001,9,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,0.0,878.0,7.0,20.0,18.956999999999994,0.0,32.12383333333334,4.5,972.5808333333334,10,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,0.0,557.5,6.5,390.0,28.291999999999998,0.0,31.271,4.0,1029.563,11,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-14.0,0.0,756.0,9.0,1590.0,27.19366666666667,0.0,28.146333333333327,4.75,2429.09,12,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-15.0,100.0,832.5,8.0,1690.0,48.894333333333336,0.0,48.63400000000001,5.25,2748.278333333333,13,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,100.0,608.0,6.0,2010.0,25.173999999999996,0.0,28.472999999999995,4.25,2793.897,14,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-15.0,100.0,404.5,8.0,3170.0,24.104000000000003,0.0,22.700833333333332,4.25,3748.5548333333336,15,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,100.0,276.0,7.5,2570.0,58.925333333333334,100.0,36.326,4.25,3164.001333333333,16,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,298.5,7.0,1640.0,28.730333333333327,0.0,22.291666666666668,4.5,2011.0220000000002,17,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,100.0,529.5,7.5,1850.0,26.260666666666665,0.0,27.17516666666667,4.25,2556.685833333333,18,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,588.0,6.0,1130.0,26.426666666666666,0.0,23.63266666666667,3.5,1787.5593333333331,19,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,0.0,228.5,5.0,1020.0,27.958000000000002,0.0,25.213,2.5,1318.1709999999998,20,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,100.0,335.5,6.5,1370.0,250.0213333333333,500.0,118.16833333333332,3.25,2694.4396666666667,21,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,0.0,532.5,6.0,740.0,40.284333333333336,0.0,37.73900000000001,3.5,1371.0233333333333,22,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,200.0,532.0,6.0,550.0,231.81866666666667,400.0,104.126,3.75,2036.6946666666668,23,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,518.5,6.0,750.0,47.44799999999999,0.0,39.35933333333334,3.75,1375.0573333333332,24,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,582.5,6.5,1680.0,42.67266666666667,0.0,43.27016666666667,4.0,2368.9428333333335,25,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,0.0,435.5,6.0,2020.0,35.633333333333326,0.0,29.739333333333338,3.5,2539.3726666666666,26,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,0.0,493.5,5.5,1400.0,0.0,0.0,6.9111666666666665,3.5,1918.4111666666668,27,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-7.0,100.0,299.5,4.5,800.0,28.88133333333333,0.0,24.505166666666668,3.0,1267.3864999999998,28,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-8.0,0.0,210.0,4.5,440.0,29.748999999999995,0.0,24.656166666666664,2.5,719.4051666666667,29,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-7.0,0.0,313.0,5.0,310.0,40.477666666666664,0.0,26.6025,2.5,704.5801666666666,30,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,100.0,257.5,5.5,110.0,69.965,0.0,46.48566666666666,3.0,601.4506666666666,31,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,0.0,305.5,7.5,620.0,79.22399999999999,0.0,57.020833333333336,3.75,1084.9948333333334,32,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,200.0,168.5,5.0,270.0,125.87233333333333,200.0,69.63366666666667,2.75,1051.7559999999999,33,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,100.0,203.5,5.5,800.0,116.42533333333333,100.0,72.75016666666667,3.25,1412.4255,34,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,100.0,178.0,6.0,840.0,121.959,100.0,72.62383333333334,3.25,1433.8328333333334,35,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,100.0,257.0,6.5,1220.0,118.35299999999998,100.0,70.9795,3.5,1885.3325,36,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,100.0,342.0,6.0,730.0,100.79233333333335,100.0,58.06166666666667,3.25,1451.104,37,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-7.0,100.0,212.0,4.5,780.0,101.69466666666666,100.0,59.76166666666666,2.5,1367.4563333333333,38,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,100.0,504.0,8.5,380.0,61.19,0.0,42.419000000000004,4.75,1112.859,39,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,545.0,9.0,230.0,63.55433333333333,0.0,41.792500000000004,5.25,904.5968333333333,40,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,0.0,622.5,8.5,270.0,59.859333333333325,0.0,48.086,6.0,1026.9453333333333,41,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20

research/baselines/tier2_champion.csv

@@ -1,35 +0,0 @@
-early_swap,travel,weekly_limit,late_swap,building_change,daily_limit,overtime,room_change,battery_swap,total_cost,origin,model,planner,p_star,day_q,ec_margin,ec_pmax,split_days,restarts,iters
-246.0,30.686666666666667,0.0,950.0,11.0,0.0,20.22566666666667,7.0,4.0,1268.9123333333334,8,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-982.5,22.80516666666667,0.0,300.0,11.0,0.0,23.703000000000003,7.0,4.5,1351.5081666666665,9,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-878.0,32.12383333333334,0.0,20.0,12.0,0.0,18.956999999999994,7.0,4.5,972.5808333333334,10,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-562.0,32.39566666666667,0.0,390.0,12.0,0.0,25.437000000000005,6.5,4.0,1032.332666666667,11,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-756.0,28.146333333333327,0.0,1590.0,14.0,0.0,27.19366666666667,9.0,4.75,2429.09,12,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-832.5,48.63400000000001,100.0,1690.0,15.0,0.0,48.894333333333336,8.0,5.25,2748.278333333333,13,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-613.5,28.516666666666666,100.0,2010.0,12.0,0.0,25.173999999999996,6.0,4.25,2799.440666666667,14,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-410.5,22.700833333333332,100.0,3170.0,15.0,0.0,24.104000000000003,8.0,4.25,3754.5548333333336,15,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-276.0,36.326,100.0,2570.0,11.0,100.0,58.925333333333334,7.5,4.25,3164.001333333333,16,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-300.0,22.291666666666668,0.0,1640.0,10.0,0.0,28.730333333333327,7.0,4.5,2012.522,17,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-531.0,28.262000000000004,0.0,1820.0,12.0,0.0,26.399,7.0,4.25,2428.911,18,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-588.0,23.63266666666667,0.0,1130.0,10.0,0.0,26.426666666666662,6.0,3.5,1787.5593333333331,19,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-229.0,25.056833333333337,0.0,1010.0,9.0,0.0,27.958,5.0,2.5,1308.5148333333334,20,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-343.0,133.68166666666667,100.0,1410.0,11.0,400.0,280.50266666666664,6.5,3.25,2687.9343333333336,21,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-532.5,37.73900000000001,0.0,740.0,11.0,0.0,40.284333333333336,6.0,3.5,1371.0233333333333,22,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-532.0,104.126,200.0,550.0,9.0,400.0,231.81866666666667,6.0,3.75,2036.6946666666668,23,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-521.0,41.3665,0.0,750.0,10.0,0.0,42.93599999999999,6.0,3.75,1375.0525,24,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-585.5,41.27883333333333,0.0,1670.0,10.0,0.0,42.67266666666667,6.5,4.0,2359.9515,25,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-431.5,29.635333333333335,0.0,2020.0,8.0,0.0,35.633333333333326,5.5,3.5,2533.768666666667,26,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-493.0,6.8445,0.0,1400.0,9.0,0.0,0.0,5.5,3.5,1917.8445,27,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-299.5,24.505166666666668,100.0,800.0,7.0,0.0,28.88133333333333,4.5,3.0,1267.3864999999998,28,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-210.0,24.656166666666664,0.0,440.0,8.0,0.0,29.748999999999995,4.5,2.5,719.4051666666667,29,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-315.5,26.473833333333335,0.0,310.0,6.0,0.0,40.09,5.0,2.5,705.5638333333334,30,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-266.0,44.59016666666666,0.0,110.0,9.0,0.0,69.765,5.5,3.0,507.8551666666667,31,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-306.5,56.99399999999999,0.0,620.0,12.0,0.0,78.45933333333333,7.5,3.75,1085.2033333333334,32,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-168.5,69.63366666666667,100.0,300.0,10.0,200.0,125.46300000000002,5.0,2.75,981.3466666666668,33,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-203.5,72.75016666666667,100.0,800.0,11.0,100.0,116.42533333333331,5.5,3.25,1412.4255,34,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-178.0,72.62383333333334,100.0,840.0,12.0,100.0,121.959,6.0,3.25,1433.8328333333334,35,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-257.0,70.9795,100.0,1220.0,9.0,100.0,118.35299999999998,6.5,3.5,1885.3325,36,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-346.0,58.467833333333346,100.0,740.0,10.0,100.0,98.337,6.0,3.25,1462.0548333333334,37,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-212.0,59.76166666666666,100.0,780.0,7.0,100.0,101.69466666666666,4.5,2.5,1367.4563333333333,38,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-503.0,42.244166666666665,0.0,390.0,12.0,0.0,61.56933333333333,8.5,4.75,1022.0635,39,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-548.0,42.10983333333334,0.0,200.0,10.0,0.0,61.123000000000005,9.0,5.25,875.4828333333335,40,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60
-622.5,48.086,0.0,270.0,12.0,0.0,59.859333333333325,8.5,6.0,1026.945333333333,41,logbag,tsroute,0.8,0.05,0.8,25.0,7,2,60

research/frozen.json

@@ -1,12 +0,0 @@
-{
- "research/run_experiment.py": "2e9931f9f5aa6fb4627980f86318441f4865fa00c13ba2579f50d85ffead5615",
- "research/program.md": "e87d9b80ca60cd4222664a63c77bd9bd7ce9cdc630f6eb617b32a9f3a779b81b",
- "research/run_sat.py": "4a5c09ffc605dca4a1051307cc2ffecc72bbaae26187e0272866c91750c179d5",
- "research/program_opt.md": "87c598abc3732fa4e1d07533f5783fcaeb228b64bd4f578645806de9ba68843a",
- "batteryswap_example/models_cv.py": "67565183fec483edc064a23f2c9a9436218b1b8df00113d01c2584b604ad2726",
- "batteryswap_example/features.py": "4e9719559c8575bd72ef8b2875600b9ab8845cb0da11adc4814b080e66a274da",
- "batteryswap_example/aux_features.py": "4195728fdb40118d7969ce769fe0c5a67ad9d157e212bee16f0d3d01b8814d98",
- "batteryswap_example/fht_features.py": "6272f40f50b83d6aac00ae1ce85c17d1f6b1eb4d3fdd5817869bbc0c7c6fb3fb",
- "batteryswap_public/evaluate.py": "686554ce05980e4eb9259c6dba0dad2f5bd12be04628cd707863c2624273f900",
- "batteryswap_public/utils.py": "03959461fc7c36c28a1095908f4a2300c81e275ded8d6fb463d41859e2f8fbb3"
-}

research/hvac_device_profiles.csv

@@ -1,462 +0,0 @@
-building_id,room_id,device_id,t_mean,t_min,t_max,t_p05,t_p95,diurnal_wd,diurnal_we,seasonal,device_class
-b_15bc00efc223,r_89c3bc35e31a,d_27ce1ac0a554,11.720566749572754,-13.0,29.0,-2.883333444595337,22.0,9.091339111328123,2.131082057952881,14.248479843139648,AHU on weekday schedule (unheated fan room / intake)
-b_15bc00efc223,r_89c3bc35e31a,d_75ee4d242a8c,12.842096328735352,-12.5,34.050846099853516,0.0,25.09847488403317,10.627193450927734,2.3819475173950195,17.16628074645996,AHU on weekday schedule (unheated fan room / intake)
-b_15bc00efc223,r_89c3bc35e31a,d_d8db71884f87,10.458189010620115,-14.0,31.06779670715332,-4.0,21.6782335281372,9.738447189331056,2.1403679847717285,15.521805763244627,AHU on weekday schedule (unheated fan room / intake)
-b_15bc00efc223,r_8fa484a8c87f,d_4643496a7525,12.23665714263916,-15.0,38.53992462158203,-2.0,26.0,12.61494255065918,4.180089950561523,17.319780349731445,AHU on weekday schedule (unheated fan room / intake)
-b_15bc00efc223,r_8fa484a8c87f,d_d13630641f39,12.77605438232422,-14.0,33.0,-2.0,24.0,11.45786952972412,3.6446962356567383,15.160465240478516,AHU on weekday schedule (unheated fan room / intake)
-b_2b69c4007582,r_268fb9378735,d_58fb8fc8d049,19.874509811401367,11.0,39.41428756713867,14.82456111907959,27.630857467651367,3.30810546875,0.2761955261230469,6.893109321594238,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_268fb9378735,d_603a3854ad64,17.00457763671875,-1.7951806783676147,33.587501525878906,7.0,25.050957012176507,4.612733840942383,3.990692138671875,12.649039268493652,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_268fb9378735,d_797774d63773,19.173418045043945,2.0,45.102272033691406,12.0,28.81405782699585,5.670654296875,4.515874862670898,8.941559791564941,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_268fb9378735,d_7a18144dbc8c,19.69457244873047,12.910256385803224,31.213115692138672,16.58451747894287,23.949490547180176,3.823644638061523,0.5068645477294922,4.065673828125,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_268fb9378735,d_a8798c4b779b,21.409412384033203,12.0,30.620689392089844,17.0,26.0,4.681884765625,4.193572998046875,4.289604187011719,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_268fb9378735,d_f28992902f9a,16.366439819335938,-1.034883737564087,28.0,6.7797980308532715,24.0,3.710890769958496,0.2669582366943359,11.774910926818848,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_7c97db8498cc,d_0039522d1a72,21.728832244873047,14.708860397338867,35.7662353515625,18.880434036254883,26.36619758605957,5.473882675170898,1.299386978149414,3.3742523193359375,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_072f1ae046d5,18.39676856994629,-1.6172839403152466,35.47541046142578,7.0,26.826377868652344,7.528735160827637,1.430694580078125,12.655166625976562,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_1be9309abd26,21.79002571105957,15.077777862548828,30.438356399536133,19.0,25.17478637695312,4.281316757202148,3.727474212646485,2.893413543701172,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_7c97db8498cc,d_24c9ba0637d3,21.113731384277344,9.0,35.0,16.15779609680176,26.5,5.940044403076172,1.1014957427978516,4.591632843017578,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_3b313557033e,15.829476356506348,-5.0,31.630434036254883,4.363362216949463,25.0,5.764679908752441,1.7860851287841797,14.611957550048828,AHU on weekday schedule (unheated fan room / intake)
-b_2b69c4007582,r_7c97db8498cc,d_49c35453655e,20.204553604125977,14.0,30.53333282470703,16.533118057250977,24.711915969848633,4.66297721862793,5.397190093994141,3.4974803924560547,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_7c97db8498cc,d_5dda17594a4d,17.910062789916992,0.0833333358168602,36.0,9.0,25.0,2.6259937286376953,1.3257598876953125,11.269867897033691,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_7c97db8498cc,d_65da0b29465f,19.711929321289062,7.0,34.0,13.0,25.66666603088379,1.5318489074707031,0.5750999450683594,7.943972587585449,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_7c97db8498cc,d_68681751508d,21.143003463745117,11.0,34.907691955566406,17.0,25.818599128723143,4.844944000244141,0.9557819366455078,4.167694091796875,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_6c39de1f466c,16.531587600708008,-2.0,32.81666564941406,5.574166703224183,25.316667556762695,5.910214424133301,5.900474548339844,13.01028823852539,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_7c97db8498cc,d_6e85484c361a,23.803627014160156,11.0,45.0,14.0,37.74217529296874,13.685470581054688,14.162721633911133,3.8824024200439458,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_7c97db8498cc,d_85cdd412f0d4,17.07576560974121,-1.0,32.49367141723633,8.0,24.586520004272455,4.734308242797852,4.223555564880371,11.37317943572998,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_7c97db8498cc,d_969010a27c20,21.664613723754883,12.847457885742188,34.0,18.27118682861328,25.649999618530273,4.799837112426758,0.7630577087402344,1.6525192260742188,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_ac9596782280,15.884255409240724,-5.031745910644531,31.89655113220215,4.296491050720216,25.0,6.558197975158691,1.5858564376831057,14.51486587524414,AHU on weekday schedule (unheated fan room / intake)
-b_2b69c4007582,r_7c97db8498cc,d_ae681061ad09,17.62384033203125,-5.0,30.66666603088379,6.0,25.36666679382324,5.340941429138184,1.2925643920898438,13.532517433166504,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_af3181f5a335,19.64582633972168,6.7833333015441895,43.0,13.0,25.57203493118282,1.4632720947265625,0.4249496459960937,8.91523265838623,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_7c97db8498cc,d_b081a28de119,21.22010803222656,14.0,31.016666412353516,18.0,25.758296680450435,4.354696273803711,5.225385665893555,3.5012130737304688,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_7c97db8498cc,d_b7941ce176c5,16.856136322021484,-3.0,30.53191566467285,5.093461537361145,25.0,6.824064254760742,1.1701068878173828,12.97052764892578,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_c6e9d12f54f4,17.909610748291016,-4.0,31.491525650024418,6.726487159729004,26.0,5.37246036529541,1.2804603576660156,13.569674491882324,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_ce92343b3466,21.106523513793945,13.939023971557615,34.21126937866211,17.88504238128662,25.5332462310791,5.158334732055664,1.1838130950927734,3.5236988067626953,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_e1e4faede955,20.479612350463867,8.0,34.0,15.0,25.706005764007568,5.9165802001953125,1.1784934997558594,4.676502227783203,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_e4f4d37f7a22,22.157878875732425,13.449999809265137,34.69841384887695,19.017735481262207,26.180964088439932,4.721805572509766,0.8218498229980469,1.6852436065673828,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_e8316ec4007e,17.34235382080078,-1.96610164642334,33.95000076293945,6.071115493774414,26.408533573150635,6.376008033752441,6.660612106323242,13.132269859313965,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_7c97db8498cc,d_e84c22b129bd,20.129281997680664,8.3125,44.0,14.0,25.337434577941895,1.9573307037353516,0.4941272735595703,7.4806928634643555,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_7c97db8498cc,d_f2c19747d515,26.250886917114254,8.0,52.025001525878906,14.52448649406433,38.76811008453367,14.61431884765625,17.21548843383789,1.3105735778808594,electrical/server room (constant heat load)
-b_2b69c4007582,r_7c97db8498cc,d_f4ac02d710b7,16.873836517333984,-3.0,29.936708450317383,5.160534143447876,25.0,6.161290168762207,1.0518054962158203,13.052984237670898,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_fdd2405211e0,20.5537052154541,10.608108520507812,33.70000076293945,16.120716381073,25.0,4.845315933227539,0.9679718017578124,4.239168167114258,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_7c97db8498cc,d_ff81017c5b6a,16.335556030273438,-3.0,26.799999237060547,4.983333349227905,23.0,5.090707778930664,0.9453201293945312,12.486043930053713,AHU on weekday schedule (unheated fan room / intake)
-b_2b69c4007582,r_a86a82a16ca0,d_04677e838244,16.19078254699707,-4.730769157409668,33.966102600097656,4.036151885986328,25.13333320617676,5.447981834411621,4.205502510070801,13.684288024902344,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_2b69c4007582,r_a86a82a16ca0,d_0d8ba51ca08a,21.30316925048828,14.0,29.0,17.649999618530273,27.0,4.034461975097656,0.3396644592285156,3.7073287963867183,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_a86a82a16ca0,d_1304e3aa21dd,22.244142532348633,6.766666889190674,38.0,12.649999618530272,31.0,12.559588432312012,11.193439483642578,2.270526885986328,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_a86a82a16ca0,d_2ccf5c6f5047,16.768871307373047,-4.604650974273682,31.0,4.0,25.0,5.596611976623535,0.6347846984863281,12.858224868774414,AHU on weekday schedule (unheated fan room / intake)
-b_2b69c4007582,r_a86a82a16ca0,d_4752f4c9d899,21.115386962890625,15.0,29.5,17.0,25.0,5.917627334594727,4.495311737060547,1.074441909790039,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_a86a82a16ca0,d_7f2679539212,19.21288299560547,-5.0,34.349998474121094,8.0,26.23432207107544,5.038547515869141,3.669586181640625,11.829445838928224,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_a86a82a16ca0,d_7f7b3efb75a6,19.692941665649418,6.137930870056152,28.483333587646484,16.53419542312622,24.0,3.8449878692626953,3.383831024169922,3.2857818603515625,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_a86a82a16ca0,d_89852baf20d6,20.97535514831543,14.0,29.0,16.241378784179688,25.0,6.401071548461914,4.888246536254883,0.8987407684326172,AHU on weekday schedule (tempered plant room)
-b_2b69c4007582,r_a86a82a16ca0,d_a0c99683a3b7,19.911218643188477,8.0,29.03333282470703,17.0,24.0,3.596405029296875,3.3305892944335938,3.041492462158203,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_a86a82a16ca0,d_a431623f4b31,16.3203125,-4.625,34.19230651855469,4.346235394477844,25.40169477462769,5.1338396072387695,3.906153678894043,13.83450984954834,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_2b69c4007582,r_a86a82a16ca0,d_c9ef403b69ca,17.62718963623047,1.2978723049163818,42.060001373291016,9.0,28.640570831298824,2.920373916625977,2.160490036010742,11.10678768157959,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_a86a82a16ca0,d_cd5d32257970,22.402862548828125,7.699999809265137,38.0,13.0,31.0,12.14903163909912,10.903261184692385,2.4935855865478516,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_a86a82a16ca0,d_d9b48866e76e,19.152080535888672,7.0,45.0,11.075471878051758,31.25353765487671,5.150566101074219,4.091098785400391,10.764589309692385,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_a86a82a16ca0,d_dac65fa276da,19.28464126586914,-5.0,34.25,9.0,25.987500190734863,5.062030792236328,4.008840560913086,11.0407075881958,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_a86a82a16ca0,d_df4946867bd5,17.456384658813477,-3.81818175315857,34.79661178588867,4.363323020935058,26.271495056152343,6.416512489318848,4.154000282287598,14.52467441558838,AHU on weekday schedule (unheated fan room / intake)
-b_2b69c4007582,r_a86a82a16ca0,d_e64e6520dadb,17.509336471557617,-3.775861978530884,34.38333511352539,4.845517158508303,26.237287521362305,6.122804641723633,3.76456069946289,14.672659873962402,AHU on weekday schedule (unheated fan room / intake)
-b_2b69c4007582,r_a86a82a16ca0,d_fdc281eb6fdf,18.0334415435791,0.0,38.8983039855957,7.0,30.11864471435547,3.0995445251464844,2.193675994873047,10.851329803466797,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_e1bac19501a4,d_30e86d81563a,19.876813888549805,13.0,27.0,17.0,23.0,0.4484500885009765,0.4008045196533203,2.4114208221435547,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_e1bac19501a4,d_4d06bb13499d,11.339960098266602,-12.0,43.91666793823242,-3.0,26.36666679382324,9.69847297668457,8.910895347595215,18.731786727905273,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_2b69c4007582,r_e1bac19501a4,d_766f6b91ee47,19.744712829589844,12.0,26.0,16.0,23.0,0.3954029083251953,0.4462223052978515,3.1367721557617188,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_e1bac19501a4,d_7ecd63c5752d,16.77283477783203,-5.0,35.5,3.0,26.0,1.873234748840332,1.6789484024047852,11.076727867126465,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_2b69c4007582,r_e1bac19501a4,d_89bca7e18496,27.78583717346191,3.866666555404663,46.900001525878906,11.333333015441896,41.0,1.974702835083008,2.4412364959716797,16.311519622802734,hot plant (boiler / DH substation / hot pipes)
-b_2b69c4007582,r_e1bac19501a4,d_8d4ede45eea7,20.5036678314209,-5.0,37.0,6.0,30.865965843200662,1.3480453491210938,1.1654186248779297,5.075130462646484,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_e1bac19501a4,d_cbf684280da2,19.373607635498047,13.0,27.0,16.0,22.0,0.4299373626708984,0.4266014099121094,2.81441879272461,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_e1bac19501a4,d_cf89420c1953,19.097993850708008,-5.0,40.0,5.0,29.0,1.4870452880859375,1.2361202239990234,5.180994987487793,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_e1bac19501a4,d_cff8f40023ea,11.686605453491213,-10.199999809265137,35.25,-0.233333334326744,24.233333587646484,5.496964454650879,4.808553695678711,16.38241958618164,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_2b69c4007582,r_e1bac19501a4,d_ec9b52a6ae09,19.879165649414062,-5.0,41.0,6.0,30.0,1.2445831298828125,1.2273292541503906,4.193321228027344,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_e1bac19501a4,d_eea64468d01f,12.177592277526855,-9.883333206176758,38.50847625732422,0.0,24.933332443237305,5.771266937255859,5.087883949279785,16.453916549682617,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_2b69c4007582,r_e1bac19501a4,d_f47dd85c5ba2,20.110994338989254,13.466666221618652,27.0,17.0,23.0,0.4235305786132812,0.3518104553222656,2.561511993408203,tempered indoor space (comfort HVAC / radiators)
-b_2b69c4007582,r_e1bac19501a4,d_f51e2ab8527e,27.76450538635254,8.866666793823242,46.01666641235352,15.724598026275636,39.0,1.320087432861328,2.14498519897461,14.083057403564451,unheated/semi-outdoor space (vent-only)
-b_2b69c4007582,r_e1bac19501a4,d_ff4f5ffc61ff,11.639287948608398,-11.550000190734863,40.06666564941406,-2.0,25.53333282470703,7.429887771606445,6.69234561920166,18.039852142333984,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_4217a3b6e58b,r_84ab2c9a82b8,d_77f3f0494d56,18.937320709228516,6.406779766082764,30.694915771484375,12.0,25.0,2.034788131713867,1.2020111083984375,7.84197998046875,unheated/semi-outdoor space (vent-only)
-b_4217a3b6e58b,r_a36e0de3c5c2,d_1c46b0502100,19.423198699951172,12.260000228881836,28.0,15.0,24.0,2.282491683959961,1.1491031646728516,5.7493133544921875,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_a36e0de3c5c2,d_231eb3170e25,18.21867561340332,4.767857074737549,31.0,10.0,25.969444751739506,4.012916564941406,2.4137649536132812,11.109143257141112,AHU on weekday schedule (tempered plant room)
-b_4217a3b6e58b,r_a36e0de3c5c2,d_2d492b0b7492,16.646615982055664,-2.0,33.79999923706055,7.0,26.0,8.030694961547852,5.317658424377441,11.771852493286133,AHU on weekday schedule (tempered plant room)
-b_4217a3b6e58b,r_a36e0de3c5c2,d_921338dbff08,23.027982711791992,15.0,32.95000076293945,18.049999237060547,27.566667556762695,3.5971946716308594,1.5657215118408203,-2.4993228912353516,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_a36e0de3c5c2,d_97b61ab060a6,19.216384887695312,11.306122779846191,27.66666603088379,14.0,23.0,2.6343536376953125,0.7745170593261719,5.483874320983887,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_a36e0de3c5c2,d_980c3c138384,20.807689666748047,10.0,32.06666564941406,15.0,26.0,3.6019420623779297,1.8915328979492188,5.327747344970703,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_a36e0de3c5c2,d_999dd113343c,20.039228439331055,12.0,34.44827651977539,16.0,24.2566665649414,6.469852447509766,3.4929122924804688,1.2295989990234375,AHU on weekday schedule (tempered plant room)
-b_4217a3b6e58b,r_a36e0de3c5c2,d_ed1309692d81,19.829374313354492,11.0,27.933332443237305,15.0,24.0,3.96272087097168,2.3071517944335938,5.111183166503906,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_bd4c57fb510d,d_06f1081b126d,20.56335067749023,13.0,31.233333587646484,17.0,25.88333320617676,7.556924819946289,4.097537994384766,0.8559055328369141,AHU on weekday schedule (tempered plant room)
-b_4217a3b6e58b,r_bd4c57fb510d,d_0ae76f0be3e4,19.847396850585938,13.0,27.172412872314453,16.036429309844973,23.0,2.9157962799072266,1.8681507110595703,3.627853393554687,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_bd4c57fb510d,d_479be9ca4f22,23.89474868774414,15.0,34.0,19.0,28.0,1.8832416534423828,0.8908348083496094,-2.6956043243408203,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_bd4c57fb510d,d_60d1aa2cdc3a,19.570173263549805,12.0,26.25,16.0,22.733333587646484,2.584993362426758,1.596364974975586,3.767766952514648,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_bd4c57fb510d,d_7532136f7c0a,18.207664489746097,4.0,32.25,12.0,24.0,4.2708282470703125,2.7895946502685547,7.815701484680176,unheated/semi-outdoor space (vent-only)
-b_4217a3b6e58b,r_bd4c57fb510d,d_7bdc39a8d750,19.443267822265625,12.0,27.766666412353516,16.816667556762695,22.86440658569336,4.299596786499023,3.4528942108154297,-0.115549087524414,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_bd4c57fb510d,d_866c14727d8f,19.294578552246097,12.0,26.0,16.0,22.983051300048828,1.5224266052246094,1.3294792175292969,4.076026916503906,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_bd4c57fb510d,d_ccd3a65228a9,21.899639129638672,15.0,42.06666564941406,19.0,24.600000381469727,1.861907958984375,0.8272457122802734,-1.1082839965820312,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_bd4c57fb510d,d_d11f0f3781af,25.219560623168945,11.0,48.54237365722656,16.0,38.383020782470695,17.509279251098633,8.388250350952148,-2.237651824951172,electrical/server room (constant heat load)
-b_4217a3b6e58b,r_bd4c57fb510d,d_ea9025c83c80,19.224292755126953,0.0,35.58333206176758,10.0,28.898305892944336,7.248908996582031,7.381744384765625,12.197389602661133,unheated/semi-outdoor space (vent-only)
-b_4217a3b6e58b,r_bd4c57fb510d,d_f95037e38d17,18.902177810668945,-1.0,39.694915771484375,9.0,29.0,11.804664611816406,6.4600114822387695,12.882052421569824,AHU on weekday schedule (tempered plant room)
-b_4217a3b6e58b,r_cdb37c5edd19,d_360efda95d1e,21.13947868347168,14.0,27.183332443237305,17.318277740478514,24.183332443237305,0.6779556274414062,0.3721809387207031,3.7910213470458984,tempered indoor space (comfort HVAC / radiators)
-b_4217a3b6e58b,r_cdb37c5edd19,d_4b58677ce660,21.86496353149414,10.133333206176758,37.03333282470703,15.371666717529298,28.91166610717773,9.635303497314451,7.75323486328125,5.71905517578125,AHU on weekday schedule (tempered plant room)
-b_4217a3b6e58b,r_cdb37c5edd19,d_8753828dd859,18.708280563354492,7.550000190734863,29.683332443237305,13.616666793823242,24.0,7.596772193908691,6.774287223815918,1.8848953247070312,tempered indoor space (comfort HVAC / radiators)
-b_53df020ad893,r_4d95939c1d71,d_5aa382f54871,21.67461585998535,6.083333492279053,33.66666793823242,13.0,27.0,0.5143699645996094,0.4483757019042969,6.379417419433594,unheated/semi-outdoor space (vent-only)
-b_53df020ad893,r_4d95939c1d71,d_773cab150c55,20.831632614135746,9.0,30.0,14.0,26.0,0.7396697998046875,0.7207412719726562,4.756689071655273,tempered indoor space (comfort HVAC / radiators)
-b_53df020ad893,r_4d95939c1d71,d_f95f29911228,20.743249893188477,9.0,30.0,14.0,26.0,0.7050285339355469,0.6418628692626953,4.738216400146484,tempered indoor space (comfort HVAC / radiators)
-b_53df020ad893,r_9b5c4b767d88,d_500e0082592c,14.949610710144045,-13.716666221618652,33.61666488647461,0.7145238220691684,25.69300880432127,3.947136878967285,3.0363693237304688,15.756927490234377,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_53df020ad893,r_9b5c4b767d88,d_b5b678a3f79f,18.4337215423584,6.666666507720947,31.20339012145996,13.666666984558104,23.315245628356923,1.748676300048828,1.6980609893798828,5.037962913513184,tempered indoor space (comfort HVAC / radiators)
-b_53df020ad893,r_9b5c4b767d88,d_e248f88b4536,27.54089736938477,1.0,57.0,18.0,38.87691402435303,3.963661193847656,1.5410003662109375,0.5263137817382812,electrical/server room (constant heat load)
-b_55038c6689a0,r_ddd012cdb553,d_3a09948220e0,18.350473403930664,1.0,35.69945526123047,9.854684925079344,26.55321941375732,4.718257904052734,2.6798152923583984,10.67188835144043,AHU on weekday schedule (tempered plant room)
-b_55038c6689a0,r_ddd012cdb553,d_440dd2bc82e5,22.21662712097168,17.758621215820312,34.54999923706055,19.762712478637695,26.399999618530273,5.256837844848633,4.011192321777344,0.7636032104492188,tempered indoor space (comfort HVAC / radiators)
-b_55038c6689a0,r_ddd012cdb553,d_44fc5c1bcf38,21.33912467956543,11.117647171020508,32.01960754394531,18.69892120361328,24.52941131591797,4.810388565063477,1.1132164001464844,0.7736034393310547,AHU on weekday schedule (tempered plant room)
-b_55038c6689a0,r_ddd012cdb553,d_52deb7b58d70,17.369216918945312,0.0,33.193180084228516,9.0,25.0,6.324104309082031,0.964975357055664,9.744051933288574,AHU on weekday schedule (tempered plant room)
-b_55038c6689a0,r_ddd012cdb553,d_7d8343427c5f,17.576967239379883,0.0,32.51515197753906,9.0,24.309987449645984,5.447837829589844,0.4157676696777344,8.806678771972656,AHU on weekday schedule (tempered plant room)
-b_55038c6689a0,r_ddd012cdb553,d_9bda8de77e4f,18.800039291381836,8.0,39.45000076293945,13.0,30.0,3.294715881347656,2.743837356567383,8.308440208435059,unheated/semi-outdoor space (vent-only)
-b_55038c6689a0,r_ddd012cdb553,d_a7c64361308e,23.87238311767578,11.0,48.0,15.0,36.83363533020019,12.290653228759766,6.324550628662109,-0.4448394775390625,AHU on weekday schedule (tempered plant room)
-b_55038c6689a0,r_ddd012cdb553,d_ada4f028bcd1,24.17230987548828,11.0,51.375999450683594,14.972982501983644,40.0,17.074941635131836,0.6233005523681641,1.2377071380615234,hot plant (boiler / DH substation / hot pipes)
-b_55038c6689a0,r_ddd012cdb553,d_c7742a7b81aa,21.55841827392578,14.683453559875488,31.254545211791992,18.934851455688477,25.45180521011352,6.43426513671875,0.3978271484375,1.888071060180664,AHU on weekday schedule (tempered plant room)
-b_55038c6689a0,r_ddd012cdb553,d_f1f5cc249ed6,23.86528778076172,11.0,49.02000045776367,15.0,39.0,15.253204345703123,0.4571762084960937,0.8038215637207031,AHU on weekday schedule (tempered plant room)
-b_60c6d1796ab4,r_b4c6b2670bd8,d_00c7d7c93295,20.594017028808594,14.0,29.0,17.0,24.88333320617676,0.445953369140625,0.3964729309082031,5.433786392211914,tempered indoor space (comfort HVAC / radiators)
-b_60c6d1796ab4,r_b4c6b2670bd8,d_0aade8a828a7,19.451894760131836,11.0,52.0,15.0,27.0,1.076864242553711,1.298837661743164,5.8356781005859375,tempered indoor space (comfort HVAC / radiators)
-b_60c6d1796ab4,r_b4c6b2670bd8,d_84bde2e0b708,16.265453338623047,-5.0,36.0,5.0,29.0,6.030130386352539,6.236741065979004,16.02835464477539,unheated/semi-outdoor space (vent-only)
-b_60c6d1796ab4,r_b4c6b2670bd8,d_a781dd8864b0,30.169496536254883,24.0,36.0,27.0,34.0,1.066385269165039,1.043134689331055,1.859477996826172,electrical/server room (constant heat load)
-b_60c6d1796ab4,r_e301f99df217,d_abfc3817b110,20.13431167602539,8.0,29.0,17.0,24.450000762939453,1.2161598205566406,1.1746940612792969,5.6898956298828125,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5bd522ea865e,d_35724f3a1f80,13.586320877075195,-10.593220710754396,32.75,3.0,20.62659645080567,6.641728401184082,2.2269954681396484,9.112591743469238,AHU on weekday schedule (unheated fan room / intake)
-b_62e09ab003f8,r_5bd522ea865e,d_4a83f49a6956,13.594980239868164,-8.433333396911621,29.91666603088379,3.0,22.0,4.292301177978516,3.793912887573242,12.677648544311523,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_62e09ab003f8,r_5bd522ea865e,d_67db8b5aec3b,13.57537841796875,-8.508474349975586,29.96666717529297,3.0,22.13559341430664,4.175118446350098,3.70977783203125,12.595836639404297,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_62e09ab003f8,r_5bd522ea865e,d_6882bb5bcffe,17.739303588867188,9.050000190734863,27.600000381469727,13.0,22.0,3.62820816040039,2.764802932739258,3.692172050476074,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5bd522ea865e,d_80ea9e8cb3df,16.84536361694336,11.0,22.0,14.0,19.549236679077133,1.554279327392578,1.3564720153808594,3.054903984069824,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5bd522ea865e,d_d97de528ad3a,17.422466278076172,9.0,27.53333282470703,13.0,21.332204055786107,3.393988609313965,2.913022041320801,3.552003860473633,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_0a09ec0f5d62,28.255964279174805,22.0,37.0,23.016948699951172,34.03631267547604,1.4583168029785156,0.6676883697509766,-6.323455810546875,heating equipment (winter duty)
-b_62e09ab003f8,r_5c8cfab78cad,d_16c8e77c07a0,27.046560287475582,24.0,30.0,25.0,29.0,1.0753059387207031,0.4803009033203125,-0.8524856567382812,electrical/server room (constant heat load)
-b_62e09ab003f8,r_5c8cfab78cad,d_20b07bbc3e8f,20.38301467895508,17.0,23.0,19.0,22.0,0.6654853820800781,0.221414566040039,-0.36859130859375,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_235be54ccf3c,26.220849990844727,21.0,31.0,23.0,30.0,1.107320785522461,0.5640220642089844,-4.616489410400391,heating equipment (winter duty)
-b_62e09ab003f8,r_5c8cfab78cad,d_357ab55c0b35,21.70428466796875,18.0,25.0,20.0,23.600000381469727,0.5241260528564453,0.4405517578125,-0.970855712890625,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_451150680718,25.555648803710938,20.0,29.200000762939453,21.0,28.0,0.5966339111328125,0.2556018829345703,-2.273284912109375,electrical/server room (constant heat load)
-b_62e09ab003f8,r_5c8cfab78cad,d_5a5b7fb9fb17,23.573909759521484,20.0,28.0,21.0,26.0,2.022573471069336,1.224687576293945,-0.9249839782714844,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_5b5652ea14dd,24.13760757446289,16.0,34.0,18.0,32.88333511352539,1.0689983367919922,0.6530742645263672,10.933151245117188,unheated/semi-outdoor space (vent-only)
-b_62e09ab003f8,r_5c8cfab78cad,d_7a70ccaf7e9c,24.422826766967773,18.0,33.0,20.0,31.0,1.0395660400390625,0.5047283172607422,-7.337440490722656,heating equipment (winter duty)
-b_62e09ab003f8,r_5c8cfab78cad,d_802c98a69ce8,22.688383102416992,19.516666412353516,28.0,21.0,26.0,0.3873424530029297,0.3247184753417969,0.2225494384765625,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_87852d5a2385,14.91299057006836,-10.983333587646484,31.0,2.0,22.0,8.955095291137695,5.288843154907227,10.03170394897461,AHU on weekday schedule (unheated fan room / intake)
-b_62e09ab003f8,r_5c8cfab78cad,d_9388d3123fcd,22.7893009185791,18.0,26.86666679382324,20.0,25.100000381469727,0.8427677154541016,0.35693359375,-0.4856243133544922,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_9b227597b574,22.520244598388672,17.0,28.0,20.0,26.0,1.2320499420166016,0.5065021514892578,-1.403158187866211,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_9e9a121da2bd,24.869291305541992,21.0,28.0,22.96666717529297,27.0,0.8894786834716797,0.4456634521484375,-2.6304168701171875,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_a5ffa7db0b8f,16.875120162963867,-4.833333492279053,31.683332443237305,4.896551609039307,24.0,7.1247711181640625,5.062893867492676,10.142809867858888,AHU on weekday schedule (unheated fan room / intake)
-b_62e09ab003f8,r_5c8cfab78cad,d_ae955e62c36f,18.8437442779541,-4.333333492279053,27.0,15.0,22.450000762939453,2.9273033142089844,1.6199798583984375,2.833667755126953,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_b12e5238f227,17.33588981628418,-6.0,31.933332443237305,4.25,24.0,6.784093856811523,3.73137092590332,11.345625877380373,AHU on weekday schedule (unheated fan room / intake)
-b_62e09ab003f8,r_5c8cfab78cad,d_b284008acadb,18.704158782958984,-3.900000095367432,27.0,15.0,22.111474990844723,3.0389671325683594,1.806863784790039,2.8221969604492188,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_bd747a44f474,23.343931198120117,18.0,28.0,21.0,26.0,0.7814712524414062,0.2474384307861328,-0.8805923461914062,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_bfd3f9589f83,20.90424156188965,5.639344215393066,29.0,17.616666793823242,24.0,2.38692855834961,2.2510223388671875,2.366313934326172,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_c0d57a7ee8f6,17.467266082763672,-5.5333333015441895,32.0,4.7230171203613285,24.0,6.639972686767578,3.61187744140625,11.27530002593994,AHU on weekday schedule (unheated fan room / intake)
-b_62e09ab003f8,r_5c8cfab78cad,d_ccd6076a4a3c,22.558130264282227,19.0,27.0,20.0,25.0,0.8229179382324219,0.28546142578125,-1.9271087646484375,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_cf33d60bc183,27.44664192199707,20.0,35.0,22.0,33.0,1.4104385375976562,0.6713066101074219,-4.845611572265625,heating equipment (winter duty)
-b_62e09ab003f8,r_5c8cfab78cad,d_d38a9bd56419,17.24452018737793,-6.366666793823242,32.31666564941406,6.833333492279053,23.450000762939453,4.435503005981445,2.975379943847656,8.45080280303955,unheated/semi-outdoor space (vent-only)
-b_62e09ab003f8,r_5c8cfab78cad,d_e76cf26c8d1c,24.659770965576172,20.03333282470703,28.816667556762695,22.0,27.037117767333996,1.1552391052246094,0.5562591552734375,-2.576698303222656,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_e87ebf12f990,24.985868453979492,22.0,28.0,23.0,27.0,0.8974380493164062,0.3183765411376953,-0.5029811859130859,tempered indoor space (comfort HVAC / radiators)
-b_62e09ab003f8,r_5c8cfab78cad,d_ee65c2e953f9,26.067882537841797,20.0,29.0,22.064167308807377,28.0,0.7149791717529297,0.306985855102539,-1.9914627075195312,electrical/server room (constant heat load)
-b_6869b0f15a78,r_0facb54e227f,d_80fdf8e269cf,35.920936584472656,23.0,53.0,24.0,50.0,2.8122406005859375,2.9198570251464844,-17.08929443359375,hot plant (boiler / DH substation / hot pipes)
-b_6869b0f15a78,r_0facb54e227f,d_fb23e7531fc3,31.5136661529541,23.0,39.69333267211914,24.0,38.0,0.7947158813476562,0.9622020721435548,-7.047491073608398,heating equipment (winter duty)
-b_6869b0f15a78,r_974c30d91109,d_0af9f5cde20b,34.45762634277344,22.0,45.0,26.072738552093504,42.0,0.8762359619140625,0.7302322387695312,9.859308242797852,hot plant (boiler / DH substation / hot pipes)
-b_6869b0f15a78,r_974c30d91109,d_3768c72e253d,36.46186828613281,23.670886993408203,47.32558059692383,29.0,43.0,0.8093948364257812,0.4441261291503906,9.256406784057615,hot plant (boiler / DH substation / hot pipes)
-b_6869b0f15a78,r_974c30d91109,d_39a0633e29a3,36.06916427612305,22.20339012145996,48.11249923706055,28.0,43.75,0.8043212890625,0.5951919555664062,10.465158462524414,hot plant (boiler / DH substation / hot pipes)
-b_6869b0f15a78,r_974c30d91109,d_70e835d18878,35.09449005126953,20.008546829223633,45.79464340209961,27.529350566864014,42.0,0.8510055541992188,0.5680313110351562,9.324811935424805,hot plant (boiler / DH substation / hot pipes)
-b_6869b0f15a78,r_974c30d91109,d_a87a43566961,37.645530700683594,24.0,50.0,29.49775676727295,45.48213310241699,0.8321876525878906,0.60614013671875,10.661127090454102,hot plant (boiler / DH substation / hot pipes)
-b_6869b0f15a78,r_974c30d91109,d_fea7ade52d34,33.47358703613281,21.0,73.88235473632812,25.69672784805298,41.0,0.89825439453125,0.6545181274414062,9.722963333129885,hot plant (boiler / DH substation / hot pipes)
-b_6869b0f15a78,r_ff96a3eef647,d_3049b6345348,13.09604549407959,-11.0,31.915254592895508,0.0,25.0,3.7295188903808594,3.1356658935546875,17.442358016967773,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_6869b0f15a78,r_ff96a3eef647,d_71ac9b979bea,13.950658798217772,-0.9882352948188782,32.30526351928711,4.0,22.233333587646484,3.896574020385742,3.6276779174804688,10.605496406555176,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_6869b0f15a78,r_ff96a3eef647,d_ee4f601d5f37,15.74240016937256,3.0,33.584415435791016,7.324166631698613,23.431665897369367,2.6630144119262695,2.44063949584961,9.709576606750488,unheated/semi-outdoor space (vent-only)
-b_7aac45259c8a,r_ab8e8ce3ee18,d_5dc80411b804,17.852872848510742,10.449999809265137,43.70000076293945,12.0,28.200000762939453,0.4639720916748047,0.529693603515625,3.1146907806396484,tempered indoor space (comfort HVAC / radiators)
-b_7aac45259c8a,r_ab8e8ce3ee18,d_6a6b1b21a680,21.63105583190918,13.0,40.98333358764648,15.0,31.96666717529297,0.8631305694580078,0.6458835601806641,4.362485885620117,tempered indoor space (comfort HVAC / radiators)
-b_7aac45259c8a,r_ab8e8ce3ee18,d_6c52c6e90375,22.66771697998047,12.0,46.0,14.0,34.45159091949462,0.8360233306884766,0.5923824310302734,7.360691070556641,unheated/semi-outdoor space (vent-only)
-b_7aac45259c8a,r_ab8e8ce3ee18,d_aac728f354d4,18.84382438659668,11.0,44.04999923706055,13.0,29.86666679382324,0.5596694946289062,0.4409084320068359,4.497394561767578,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_02e6b266abf6,d_859902cf5bc9,19.78389549255371,-6.589743614196777,30.534883499145508,16.84403438568115,22.081396102905273,3.2535152435302734,1.601032257080078,0.5478401184082031,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_02e6b266abf6,d_8993ef06e18e,14.299779891967772,-10.114753723144531,43.0,3.0,29.50999984741209,5.026612281799316,4.571268081665039,13.91147804260254,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_02e6b266abf6,d_bccc83ab0c2d,13.9711275100708,-11.0,49.36666488647461,-0.5214830309152599,30.99237308502196,7.79100513458252,6.9495134353637695,21.85967254638672,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_02e6b266abf6,d_c133de37fcbb,20.054906845092773,-5.830188751220703,30.06410217285156,16.554509353637695,22.92914314270019,3.379873275756836,1.5754127502441406,0.6361522674560547,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_02e6b266abf6,d_da4b7d2e7153,14.376791954040527,-11.0,49.517242431640625,-0.0828551925718784,31.102241897583006,7.798102378845215,6.893368721008301,22.09796142578125,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_0d59c2f44f40,d_101df492b4e2,21.371721267700195,14.15463924407959,40.45000076293945,17.0,28.0,7.360239028930664,4.786396026611328,3.3897571563720703,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_13c98b15c384,20.91629409790039,15.0,38.522727966308594,17.0,27.0,7.258022308349609,5.159387588500977,2.432558059692383,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_17577be96a87,22.549650192260746,13.41891860961914,37.98333358764648,17.0,28.636883544921872,7.931127548217773,4.490238189697266,-0.1003322601318359,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_1e9752415535,23.53884506225586,15.568181991577148,38.150001525878906,18.477420234680174,30.0,6.696722030639648,3.794485092163086,1.507272720336914,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_309f0e63be64,16.026607513427734,-5.866666793823242,35.06593322753906,2.88356500864029,24.466997909545892,4.500430107116699,5.186928749084473,14.609079360961914,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_0d59c2f44f40,d_3fcf7c5e0255,19.222980499267575,-3.0493826866149902,37.75,5.972079563140869,27.984880828857417,2.9231739044189453,1.9328155517578125,13.931962966918944,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_0d59c2f44f40,d_402d39007db3,19.624454498291016,-5.492537498474121,38.41666793823242,6.0,28.33239269256591,6.174186706542969,3.4113006591796875,13.24675178527832,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_447f45b687da,19.85405921936035,-5.035714149475098,42.86666488647461,6.0,30.492063522338867,4.711273193359375,2.478239059448242,15.8733491897583,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_4ade02e87b5e,16.94444465637207,-2.0,36.882354736328125,7.53317928314209,23.505494689941425,1.9462318420410156,1.6650466918945312,8.144815444946289,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_0d59c2f44f40,d_4d3899e239f2,17.87832260131836,-4.0,54.01351165771485,6.538240361213685,29.169499969482416,3.0688343048095703,2.242349624633789,9.850542068481444,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_0d59c2f44f40,d_68a0cc054179,18.25982093811035,-3.0,36.13043594360352,6.0,26.434247970581037,4.094265937805176,1.7232818603515625,14.045140266418455,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_6ff29d2db939,19.543485641479492,-2.4383561611175537,40.33333206176758,7.0,28.608849430084227,5.097173690795898,3.212806701660156,13.967514991760254,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_70d09dad9888,21.363142013549805,15.0,35.0,17.0,26.0,4.254789352416992,3.0737838745117188,0.4323787689208984,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_0d59c2f44f40,d_71cdbca5457a,21.284738540649418,14.0,30.776470184326172,17.62419538497925,25.0,3.1133899688720703,4.665798187255859,2.7495555877685547,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_0d59c2f44f40,d_7475202e8bf1,18.32288932800293,-4.0,35.0,6.0,25.5781512260437,3.3518123626708984,2.323516845703125,12.547614097595217,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_0d59c2f44f40,d_7aa40cb4e3a2,18.06245231628418,-0.8169013857841492,54.0,7.429906368255615,30.770641326904297,2.551063537597656,1.7989177703857422,9.451640129089355,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_0d59c2f44f40,d_7d80d3aa458c,18.908411026000977,-3.0,35.25,7.0,26.4575662612915,2.717344284057617,2.1606311798095703,12.883415222167969,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_0d59c2f44f40,d_8151bdb896a9,17.783039093017578,5.804123878479004,36.0,10.0,23.0,1.9476661682128904,1.0873756408691406,6.730679512023926,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_0d59c2f44f40,d_8be3a8f54b99,20.64142417907715,16.0,33.74576187133789,17.0,26.529493331909173,3.267618179321289,2.31024169921875,2.875020980834961,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_0d59c2f44f40,d_91e5f056f524,21.834245681762695,15.83505153656006,41.25,19.009561157226564,27.0,4.922435760498047,4.774076461791992,2.946237564086914,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_0d59c2f44f40,d_98c64844c9e9,18.99027442932129,-5.011494159698486,37.18181991577149,6.0,26.035319137573243,3.5771350860595703,2.14801025390625,12.605046272277832,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_0d59c2f44f40,d_9d5c212e0cd7,21.025617599487305,15.758620262145996,35.272727966308594,17.0,25.0,3.531530380249023,3.3649158477783203,1.0700645446777344,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_0d59c2f44f40,d_a3913efef129,18.540637969970703,-4.460674285888672,38.272727966308594,5.0,27.472210311889643,5.248493194580078,3.211679458618164,13.796539306640623,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_b48711795e0f,21.88749313354492,15.81521701812744,39.838382720947266,18.0,30.0,7.244701385498047,5.1038970947265625,3.9063453674316406,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_bb54aa9bdf92,19.997699737548828,15.0,31.926828384399418,16.0,25.0,3.2090988159179688,2.481039047241211,2.29498291015625,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_0d59c2f44f40,d_c31dbd35bdd9,19.255481719970703,-4.0,38.70588302612305,5.0,28.577060699462887,6.936413764953613,6.428609848022461,15.507346153259276,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_0d59c2f44f40,d_d5f20cf449f2,21.09160614013672,-2.9733333587646484,39.78333282470703,9.0,30.012855815887452,6.189027786254883,3.02310562133789,13.356749534606934,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_0d59c2f44f40,d_d70bb89bb900,16.663455963134766,-4.878378391265869,35.20779037475586,3.861562466621399,25.0,3.961881637573242,4.549726486206055,14.432123184204102,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_0d59c2f44f40,d_e175be1edebb,22.001605987548828,15.833333015441896,33.8070182800293,19.518686294555664,26.03144989013671,3.6531848907470703,2.003175735473633,1.1162738800048828,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_0d59c2f44f40,d_e6d590357383,21.56213760375977,15.0,31.140350341796875,19.0,25.01748275756836,3.1380748748779297,1.7658615112304688,0.8402786254882812,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_0d59c2f44f40,d_e7092c8db709,18.08206558227539,-5.649122714996338,42.3636360168457,3.1839548587799076,27.754634857177727,7.126055717468262,6.724370956420898,15.87726593017578,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_0d59c2f44f40,d_f47e395465b8,21.307401657104492,14.0,34.010101318359375,18.0,25.06282787322997,3.434053421020508,3.3739166259765625,1.246185302734375,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_0d59c2f44f40,d_fce9ffa6fd96,17.153100967407227,0.0,36.0,6.032167911529542,22.37337074279785,2.260282516479492,1.7597675323486328,7.905817985534668,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_27b0cfa20d53,d_6143a8090a07,14.194241523742676,-12.0,32.16666793823242,0.016666667535901,26.0,8.424931526184082,6.735470771789551,16.895414352416992,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_27b0cfa20d53,d_e629c6dd1a0d,19.080291748046875,-0.2641509473323822,35.034481048583984,15.0,25.516666412353516,2.879636764526367,3.617259979248047,3.923269271850586,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_27b0cfa20d53,d_f21bcef9d3b9,19.389375686645508,-2.207547187805176,34.655174255371094,15.0,25.066667556762695,3.621551513671875,3.9069385528564458,3.133838653564453,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_27b0cfa20d53,d_f8ce7614da97,14.08227252960205,-3.650000095367432,30.0,4.633163213729858,22.299999237060547,3.634002685546875,4.30047607421875,10.141989707946776,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_27b0cfa20d53,d_fa590b5fab73,14.49704933166504,-12.0,32.70000076293945,0.7465425342321397,26.972083759307843,8.512107849121094,6.818678855895996,16.92833709716797,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_2823849bded1,d_0dee527f5202,21.860286712646484,7.0,37.0,17.16666603088379,28.0,5.537754058837891,5.437091827392578,4.296884536743164,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_2823849bded1,d_3ff7d358691f,21.19178581237793,3.0,29.549999237060547,15.95333309173584,25.0,6.034908294677734,3.943077087402344,2.5269737243652344,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_2823849bded1,d_62320366b27d,16.52559471130371,-6.0,44.0,2.0,25.9666389465332,3.116004943847656,3.0075016021728516,12.66014575958252,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_2823849bded1,d_840fe9661fd8,15.745258331298828,-3.7666666507720943,31.0,9.0,21.0,3.4519758224487305,3.51055908203125,7.006026268005371,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_2823849bded1,d_92572f97c6d7,15.888691902160645,-2.066666603088379,32.099998474121094,5.683965587615967,25.183332443237305,2.2596378326416016,1.5125808715820312,13.84849739074707,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_2823849bded1,d_c56d12a6de45,16.166357040405273,-0.8333333134651184,37.73333358764648,5.966129899024963,26.38333320617676,4.789773941040039,4.428606033325195,13.271716117858888,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_2823849bded1,d_cba338edaedc,20.82585906982422,2.4137930870056152,29.13333320617676,15.583333015441896,24.11666679382324,5.632305145263672,3.589773178100586,2.868337631225586,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_2823849bded1,d_e20d74c76618,15.598577499389648,-2.683333396911621,33.21666717529297,5.0,25.76646871566772,2.0416526794433594,1.471552848815918,14.247384071350098,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_5b4caff693e8,d_1b2793814262,15.507293701171877,2.0,24.0,7.0,21.0,1.2368383407592771,1.165117263793945,11.542733192443848,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_5b4caff693e8,d_75c911f9d64e,14.840559005737305,-5.762712001800537,25.0,5.0,22.0,0.9765472412109376,0.7302799224853516,13.475296020507812,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_5b4caff693e8,d_fb1085fa2812,15.983396530151367,1.0,32.0,6.0,25.57416744232176,3.909366607666016,3.5294647216796875,11.200604438781738,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_786737d33818,d_1e4300c6339e,15.89241886138916,-11.0,32.81666564941406,3.048653793334961,26.33333396911621,10.77454948425293,10.12161350250244,13.39768409729004,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_786737d33818,d_259ad0f929c8,21.137157440185547,10.0,88.89473724365234,15.949999809265137,26.766666412353516,6.365262985229492,6.0817108154296875,-0.7830486297607422,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_786737d33818,d_793ab8940394,20.454301834106445,9.666666984558104,89.17857360839844,15.179360961914062,25.85074119567871,6.311361312866211,6.269325256347656,0.5884304046630859,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_786737d33818,d_e65abe350e27,16.26299476623535,-11.133333206176758,33.25,3.0414654731750494,27.0,11.520197868347168,10.961297988891602,13.499512672424316,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_9b5477e4058b,d_50c42cdc9743,17.105297088623047,-1.0,32.0,10.0,24.252183055877683,3.3076562881469727,2.8416357040405273,8.077247619628906,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9b5477e4058b,d_6f75806d4198,14.77148723602295,-13.285714149475098,36.0,0.0161640080157669,25.675369071960446,5.859504699707031,3.6131324768066406,16.122652053833008,AHU on weekday schedule (unheated fan room / intake)
-b_955300d5f090,r_9b5477e4058b,d_e48beb704ed4,14.939997673034668,-3.133333444595337,29.0,4.762712001800537,22.000833320617662,3.6949234008789062,1.6317415237426758,10.359997749328612,AHU on weekday schedule (unheated fan room / intake)
-b_955300d5f090,r_9cb7e6657012,d_2c260c3da0e0,23.54893684387207,16.044944763183594,46.0,18.0,33.05128192901611,0.8121719360351562,0.7606697082519531,9.088043212890623,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9cb7e6657012,d_37d5d18f8127,23.99941635131836,16.43037986755371,41.3220329284668,18.0,32.489739990234376,0.7864398956298828,0.8619861602783203,11.05500602722168,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9cb7e6657012,d_5d79d8c6ba7a,24.156219482421875,16.03676414489746,44.534481048583984,18.0,34.0,0.9320755004882812,0.7715034484863281,11.309234619140623,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9cb7e6657012,d_690ce74ac6ee,21.033578872680664,14.770270347595217,34.0,18.0,26.200000762939453,2.3986282348632812,1.7931594848632812,3.5955810546875,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_9cb7e6657012,d_6ae01eac2a9a,23.89593505859375,16.437955856323242,44.561405181884766,18.0,32.0,0.8447246551513672,0.9799232482910156,10.56585693359375,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9cb7e6657012,d_b2e6d2cdef68,21.67174530029297,15.962963104248049,35.0,18.0,27.0,1.5911827087402344,1.4264640808105469,4.819587707519531,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_9cb7e6657012,d_b8f4a36cf8ad,23.065603256225582,16.023256301879883,47.2337646484375,18.0,31.0,1.3232040405273438,0.968881607055664,9.917320251464844,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9cb7e6657012,d_ee0408c81af7,22.43134880065918,15.939759254455566,43.38554382324219,18.0,29.0,0.976776123046875,0.7695636749267578,7.848335266113281,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9ea33a360e32,d_035878198859,19.775768280029297,0.1754385977983474,73.78333282470703,11.322033882141112,34.33920593261718,3.329742431640625,2.228555679321289,9.925748825073242,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9ea33a360e32,d_0a280b2ba796,18.146303176879883,-5.948276042938232,33.98305130004883,4.0,28.0,5.278630256652832,4.038911819458008,17.914203643798828,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_9ea33a360e32,d_2af7891ce12d,17.945131301879883,-6.0,38.135135650634766,5.349565124511719,27.0,3.8415355682373047,2.3621768951416016,15.522449493408203,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9ea33a360e32,d_3f537962b028,20.773006439208984,14.333333015441896,40.53845977783203,16.0,28.411723136901855,6.748210906982422,3.631040573120117,3.9478931427001953,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_467ff6c54ed3,19.577777862548828,-5.666666507720947,39.6363639831543,6.0,28.96666717529297,5.66722297668457,2.526317596435547,16.551342010498047,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_53d61e22fef6,19.521652221679688,-3.4909090995788574,41.411766052246094,5.795000195503236,32.0,6.24140739440918,3.1188735961914062,19.006311416625977,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_5e9be577006a,22.733795166015625,4.890909194946289,36.349998474121094,15.383333206176758,29.389047241210925,8.370512008666992,6.929115295410156,7.091846466064453,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9ea33a360e32,d_659282cb1d0f,21.74028205871582,14.163636207580566,38.490909576416016,18.0,28.0,6.944791793823242,3.654220581054687,3.27996826171875,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_7072e61a8468,18.529930114746097,-4.068965435028076,41.38888931274414,5.0,29.402714157104487,6.189640045166016,2.7487754821777344,18.15654945373535,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_936802563817,17.292282104492188,0.0,44.36666488647461,7.382051277160644,29.10881214141846,4.254245758056641,2.308521270751953,11.82038402557373,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_9a22d49dbcf6,21.522523880004883,9.404254913330078,40.47618865966797,15.0,30.90243911743164,10.324527740478516,6.612579345703125,5.905794143676758,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_acc762a9c9d3,19.230894088745117,7.0,44.32558059692383,12.54457745552063,27.83075695037839,3.577198028564453,2.116893768310547,8.032905578613281,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9ea33a360e32,d_b31f97e26753,20.2166690826416,-4.706896781921387,40.880001068115234,6.99083321094513,30.220338821411133,5.647136688232422,2.4142017364501953,17.187854766845703,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_bc4b9b3717b0,21.887712478637695,14.535714149475098,38.38461685180664,18.0,28.32543153762816,6.619911193847656,3.4927501678466797,3.270833969116211,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_c7ecb6857b9d,18.85988426208496,-4.0,39.33333206176758,7.0,29.0,3.60330581665039,2.024921417236328,15.857961654663086,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9ea33a360e32,d_d2041339c655,22.758947372436523,10.0,43.0,15.0,33.26296195983887,10.960155487060549,6.53453254699707,6.543399810791016,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_9ea33a360e32,d_d4075135f991,21.779937744140625,2.596153736114502,34.16666793823242,15.0,27.899999618530273,7.180454254150391,6.109100341796875,6.537387847900391,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_9ea33a360e32,d_ebfebcd31227,18.68112564086914,-5.896551609039307,34.0,4.0,28.337006187438952,5.675716400146484,4.384565353393555,17.93153190612793,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_9ea33a360e32,d_eeefc1b8d774,20.88640594482422,14.449999809265137,40.91999816894531,16.0,28.67796516418457,7.056573867797852,3.7227096557617183,3.9923095703125,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_baa7dd603600,d_3b8706965be3,9.309328079223633,-11.0,39.099998474121094,0.0,20.83206281661987,6.599045753479004,4.845264911651611,21.68914031982422,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_baa7dd603600,d_98c67d44ad4d,13.451656341552734,-4.960784435272217,33.233333587646484,3.0,28.277491092681885,8.380760192871094,6.741645812988281,10.923495292663574,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_baa7dd603600,d_eb8be752b2ff,16.42601776123047,4.0,26.566667556762695,10.0,21.0,7.060050964355469,5.244081497192383,6.086867332458496,AHU on weekday schedule (tempered plant room)
-b_955300d5f090,r_ca9045f4fef2,d_1dd3ee06e4fd,14.477652549743652,-9.154929161071776,33.465518951416016,2.0429991722106937,26.0,9.517077445983888,9.31047534942627,15.101716995239258,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_ca9045f4fef2,d_c909d1e8fed3,20.50885772705078,3.5833332538604736,33.0,12.0,26.483214569091796,8.778468132019043,8.490010261535645,3.3897972106933594,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_ca9045f4fef2,d_d3cad477403e,20.206546783447266,4.0,33.0,12.0,26.16666603088379,8.445963859558105,8.160444259643555,3.5489635467529297,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_ca9045f4fef2,d_f587ec3fd537,14.9800443649292,-9.0,37.0,2.630909156799317,26.96731834411621,10.388449668884276,10.061945915222168,14.96333122253418,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_cb32bb120909,d_3d3af490dbc7,29.572975158691406,19.0,47.0,20.0,35.0,1.0594406127929688,0.8980426788330078,-8.60342788696289,heating equipment (winter duty)
-b_955300d5f090,r_cb32bb120909,d_9f1eb55e773d,29.42007827758789,19.0,45.0,20.0246488571167,35.0,0.9302539825439452,0.6247062683105469,-8.309709548950195,heating equipment (winter duty)
-b_955300d5f090,r_e156e26e7583,d_751ac25d2ed3,18.82941436767578,12.0,28.79660987854004,15.84783926010132,24.0,2.175159454345703,1.7288970947265625,4.710456848144531,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_e156e26e7583,d_9020559b3e37,17.115028381347656,-7.650000095367432,33.779659271240234,6.0,25.508757019042967,2.5612640380859375,2.657482147216797,10.712075233459473,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_e156e26e7583,d_9dd770fbab17,23.96672248840332,9.0,48.0,14.0,33.11772823333739,6.48255729675293,7.041770935058594,7.479555130004883,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_e292e2e62ec3,d_35f1f7767260,17.916046142578125,-1.0,35.0,7.0,27.0,1.1252479553222656,1.077301025390625,12.50465488433838,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_e292e2e62ec3,d_3f55a3193a24,21.98560905456543,-1.0,39.0,10.0,33.54999923706055,1.4113216400146484,1.5696773529052734,12.255239486694336,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_e292e2e62ec3,d_84c4f7bf7552,21.37899398803711,0.0,38.900001525878906,10.0,33.0,1.6220073699951172,1.4327125549316406,11.56665325164795,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_e292e2e62ec3,d_bb40a477d811,17.39714813232422,-2.0,29.0,7.0,26.0,1.196489334106445,1.1771221160888672,12.64758586883545,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_e292e2e62ec3,d_f6f64b939eaf,23.65085792541504,6.0,33.0,16.0,30.0,0.7664508819580078,0.7618446350097656,1.6248188018798828,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_e2a69ae237a4,d_97f334f0b4f5,15.025590896606444,-6.0,44.0,2.0,29.225128936767582,6.893613815307617,5.795497894287109,17.633169174194336,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_e2a69ae237a4,d_a12d4c724c23,15.074217796325684,-2.0,37.375,6.0,26.813077163696292,3.372255325317383,4.9812517166137695,11.243616104125977,unheated/semi-outdoor space (vent-only)
-b_955300d5f090,r_e2a69ae237a4,d_a95d5d10df5a,18.38113594055176,5.0,33.0,14.0,21.266666412353516,4.572484970092773,4.559695243835449,1.5266342163085938,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_e2a69ae237a4,d_a98a55fd1547,14.633928298950195,-6.650000095367432,42.75,1.938079071044923,29.11735305786133,6.624569892883301,5.582104682922363,17.813465118408203,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_e2a69ae237a4,d_dc8f05c852c6,18.13214111328125,5.081967353820801,33.0,14.0,21.06762733459473,4.016910552978516,4.389474868774414,1.3621673583984375,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_f2d7d812a5f0,d_758fc488c883,22.713642120361328,5.0,38.0,14.028333568572998,28.16666603088379,8.203170776367188,8.26121711730957,1.715078353881836,tempered indoor space (comfort HVAC / radiators)
-b_955300d5f090,r_f2d7d812a5f0,d_fc045d3599a2,15.490020751953123,-7.733333110809326,33.830509185791016,2.606610083580018,28.36367769241332,9.76571273803711,9.448280334472656,16.143718719482422,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_f7088836b3c8,d_78ab487abc88,15.352253913879396,-7.583333492279053,32.0,4.3275861740112305,23.0,8.661450386047363,8.287834167480469,9.468268394470217,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_f7088836b3c8,d_803b5158dbd7,25.22698402404785,11.948275566101074,47.344825744628906,18.31352596282959,32.147935104370106,8.393651962280273,7.872804641723633,2.4993858337402344,electrical/server room (constant heat load)
-b_955300d5f090,r_f7088836b3c8,d_a12a10261e9b,13.831034660339355,-13.0,40.5,0.0,26.0,8.49013614654541,7.9010772705078125,17.36005210876465,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_f7088836b3c8,d_beb4340cd12a,14.31838035583496,-13.0,41.06666564941406,0.0,27.0,8.788797378540039,8.195165634155273,17.233488082885742,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_955300d5f090,r_f7088836b3c8,d_d4a0918c2099,23.702730178833008,11.533333778381348,46.43333435058594,18.0,30.91666603088379,6.655710220336914,6.08978271484375,3.5265045166015625,tempered indoor space (comfort HVAC / radiators)
-b_9a5a45d6e498,r_101aa0c8e52c,d_2405b9e7bd2f,21.36368179321289,11.283019065856934,40.45098114013672,17.0,26.89825382232666,5.334722518920898,0.509429931640625,4.408226013183594,AHU on weekday schedule (tempered plant room)
-b_9a5a45d6e498,r_101aa0c8e52c,d_3734123e6d90,17.571552276611328,-2.7272727489471436,36.0476188659668,5.015114212036133,26.0,9.520512580871582,0.3691425323486328,9.736645698547363,AHU on weekday schedule (tempered plant room)
-b_9a5a45d6e498,r_101aa0c8e52c,d_5039a842fbdd,15.079570770263672,-5.0,61.793479919433594,4.0,29.583877182006816,7.625663757324219,2.732311248779297,16.218624114990234,AHU on weekday schedule (unheated fan room / intake)
-b_9a5a45d6e498,r_22232f8e4b91,d_1b79ece9f065,14.13456916809082,-4.479166507720947,27.0,5.0,21.0,2.8308029174804688,1.9019050598144531,9.779414176940918,unheated/semi-outdoor space (vent-only)
-b_9a5a45d6e498,r_22232f8e4b91,d_3e0292ad88b6,13.409470558166504,-13.0,33.712642669677734,0.9295114874839792,24.0,5.995222091674805,3.29679012298584,16.126699447631836,AHU on weekday schedule (unheated fan room / intake)
-b_9a5a45d6e498,r_22232f8e4b91,d_87604c88656b,18.75518798828125,-1.0,34.45098114013672,10.0,24.644559574127197,7.012384414672852,1.3654842376708984,5.4041900634765625,AHU on weekday schedule (tempered plant room)
-b_9a5a45d6e498,r_6359aa838297,d_045289757308,13.218707084655762,-17.246753692626953,37.74074172973633,-3.0510436415672304,26.377812194824223,4.028426170349121,3.3405990600585938,19.98171043395996,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_9a5a45d6e498,r_6359aa838297,d_102b0defe0cc,19.909095764160156,9.0,38.81666564941406,13.0,25.693127822875976,8.086503982543945,6.497640609741211,5.452365875244141,AHU on weekday schedule (tempered plant room)
-b_9a5a45d6e498,r_6359aa838297,d_1420f5ff758c,12.725709915161133,-17.0,37.33620834350586,-2.0,25.03560972213745,4.857306480407715,4.229633331298828,18.14447975158691,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_9a5a45d6e498,r_6359aa838297,d_3b189e96404c,12.840081214904783,-17.0,37.83185958862305,-3.0,25.314983844757077,4.422924995422363,3.811573028564453,18.87548065185547,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_9a5a45d6e498,r_6359aa838297,d_3efa498ef3d6,12.656044006347656,-17.390243530273438,37.45454406738281,-3.1790329694747927,26.0,4.280600547790527,3.622556686401367,18.524799346923828,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_9a5a45d6e498,r_6359aa838297,d_5a32d5bc2799,11.568306922912598,-13.0,40.17204284667969,-0.6635416805744175,26.0,10.468531608581545,4.355809688568115,17.44038963317871,AHU on weekday schedule (unheated fan room / intake)
-b_9a5a45d6e498,r_6359aa838297,d_6b475a7a7d3d,18.45942497253418,4.0,35.0,10.948534202575685,24.026770496368407,8.205302238464355,6.270359992980957,4.528957366943359,AHU on weekday schedule (tempered plant room)
-b_9a5a45d6e498,r_6359aa838297,d_834971e10647,12.949817657470703,-17.0,39.32500076293945,-1.9667376160621644,24.951085948944083,4.994367599487305,4.439969062805176,17.96832847595215,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_9a5a45d6e498,r_6359aa838297,d_9b73308fafc4,13.993330001831056,-16.0,38.59756088256836,-1.511604976654051,26.752970123291018,4.608491897583008,3.8583860397338863,19.15096664428711,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_9a5a45d6e498,r_6359aa838297,d_9ba303a913ad,15.19163990020752,-8.0,40.959999084472656,3.54895156621933,26.152413272857665,10.260004997253418,4.243503570556641,11.188955307006836,AHU on weekday schedule (unheated fan room / intake)
-b_9a5a45d6e498,r_6359aa838297,d_c1672e150e5b,13.498056411743164,-8.0,48.0,3.0,26.13569498062133,9.113289833068848,3.7946720123291016,14.123367309570312,AHU on weekday schedule (unheated fan room / intake)
-b_9a5a45d6e498,r_6359aa838297,d_d678d9f27b31,13.502012252807615,-16.88157844543457,39.2931022644043,-2.0,25.71390895843505,5.426761627197266,5.064448356628418,19.447612762451172,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_9a5a45d6e498,r_6359aa838297,d_de8ae6808bd5,14.033610343933104,-16.0,42.75308609008789,-1.219355869293213,27.308040428161608,5.421100616455078,4.759790420532227,19.24839973449707,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_9a5a45d6e498,r_6359aa838297,d_e8416d741c77,17.455127716064453,3.0,36.0,8.0,24.0,10.090961456298828,7.508760452270508,6.506142616271973,AHU on weekday schedule (tempered plant room)
-b_9a5a45d6e498,r_6359aa838297,d_f6439f15661d,17.307157516479492,2.0,36.400001525878906,8.0,23.600000381469727,9.487736701965332,7.036087989807129,6.5806427001953125,AHU on weekday schedule (tempered plant room)
-b_9a5a45d6e498,r_d1124136aa68,d_37142ede415b,14.50966739654541,-18.0,43.20388412475586,0.0,26.965029621124263,11.26941967010498,4.595731735229492,18.4365291595459,AHU on weekday schedule (unheated fan room / intake)
-b_9a5a45d6e498,r_d1124136aa68,d_d4b4272d5229,16.349300384521484,-15.440677642822266,31.0,2.0,22.613523960113525,9.232687950134276,3.738041877746582,5.773582458496094,AHU on weekday schedule (unheated fan room / intake)
-b_9a5a45d6e498,r_d1124136aa68,d_d993db63a0c7,12.259100914001465,-16.0,35.09090805053711,0.0,24.66458320617676,5.195042610168457,3.4673843383789062,17.191965103149414,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_a0e6376410cb,r_0524a19c4bfd,d_05d8311854b1,17.211360931396484,-4.0,36.0,6.0,27.0,6.359148025512695,4.623726844787598,15.34818172454834,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_0524a19c4bfd,d_1ae1d79ea843,24.30741310119629,19.0,32.408451080322266,21.5950569152832,27.0,0.4388046264648437,0.5721302032470703,-2.0261192321777344,tempered indoor space (comfort HVAC / radiators)
-b_a0e6376410cb,r_0524a19c4bfd,d_9034280235c8,13.514487266540527,-10.0,37.0,2.0,25.932557582855225,4.425678253173828,3.947467803955078,16.527929306030273,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_a0e6376410cb,r_0524a19c4bfd,d_c2ff015918a9,18.12930679321289,-2.0,33.13541793823242,7.550517320632935,25.783238410949707,8.82664680480957,8.263466835021973,9.050766944885254,unheated/semi-outdoor space (vent-only)
-b_a0e6376410cb,r_66f8934ecd8c,d_1cd02a13c100,17.68466567993164,4.0,39.04999923706055,10.7780601978302,24.766666412353516,5.897726058959961,4.599392890930176,5.2610883712768555,tempered indoor space (comfort HVAC / radiators)
-b_a0e6376410cb,r_66f8934ecd8c,d_90995329584e,18.71905517578125,7.0,36.0,12.0,25.41666603088379,6.735957145690918,5.210790634155273,2.5001163482666016,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_66f8934ecd8c,d_cfdd8b69ddd4,13.686174392700195,-9.96610164642334,41.0,3.0,25.66249942779541,9.347047805786133,6.349296569824219,13.808695793151855,AHU on weekday schedule (unheated fan room / intake)
-b_a0e6376410cb,r_66f8934ecd8c,d_e13ae085ad7d,21.25451278686523,8.0,35.0,14.0,28.78333282470703,7.945068359375,5.1810455322265625,-1.9439239501953125,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_66f8934ecd8c,d_e453b68fa165,24.03247833251953,11.0,36.0,17.0,29.0,3.751178741455078,2.820838928222656,0.2722740173339844,tempered indoor space (comfort HVAC / radiators)
-b_a0e6376410cb,r_770fc166ffa0,d_0333b66cd4d2,25.84547996520996,18.0,31.0,24.0,28.0,1.4840049743652344,0.3880977630615234,-0.380645751953125,electrical/server room (constant heat load)
-b_a0e6376410cb,r_770fc166ffa0,d_21e4ad2ae589,22.642362594604492,13.0,34.65542984008789,19.62254524230957,26.68293571472168,5.274499893188477,0.611968994140625,2.6434860229492188,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_770fc166ffa0,d_7792551fb838,19.6143856048584,-6.065573692321777,37.64313888549805,6.0,27.468992233276367,7.728634834289551,1.0728015899658203,11.366632461547852,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_770fc166ffa0,d_80dec6893ea9,28.08795166015625,22.0,34.0,25.0,31.0,0.6066207885742188,0.1561870574951172,-2.811666488647461,electrical/server room (constant heat load)
-b_a0e6376410cb,r_770fc166ffa0,d_9ae8a0552434,20.75593376159668,3.0,39.24827575683594,15.0,25.0,4.099954605102539,1.0035972595214844,4.706878662109375,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_770fc166ffa0,d_d82479011d55,22.70545959472656,13.0,34.6119384765625,19.883419799804688,26.835792160034178,5.015361785888672,0.5916805267333984,2.7637901306152344,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_77a0d3ba0f9c,d_1d0190865d91,13.777812004089355,-8.983333587646484,32.13333511352539,2.2333333492279053,23.766073322296137,2.0966615676879883,1.836775779724121,14.60659408569336,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_a0e6376410cb,r_77a0d3ba0f9c,d_260aced72405,15.556476593017578,-11.0,31.0,2.6350001096725486,25.0,5.772229194641113,6.570120811462402,12.76050853729248,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_a0e6376410cb,r_77a0d3ba0f9c,d_2f41e9aecd71,21.453344345092773,12.0,30.0,17.0,25.0,4.323944091796875,4.417280197143555,-0.5633373260498047,tempered indoor space (comfort HVAC / radiators)
-b_a0e6376410cb,r_77a0d3ba0f9c,d_4406ae97dc09,15.030332565307615,-9.0,32.79999923706055,2.785833299160005,25.899999618530273,2.8709545135498047,3.1256933212280273,16.2866153717041,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_a0e6376410cb,r_77a0d3ba0f9c,d_934319e8a338,17.857908248901367,-11.0,37.0,4.0,27.78333282470703,2.570493698120117,1.517221450805664,15.18009090423584,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_a0e6376410cb,r_77a0d3ba0f9c,d_9a4e6a7828e3,24.95998764038086,16.410715103149414,31.0,21.0,28.91666603088379,1.1246910095214844,0.6696300506591797,-2.2834224700927734,tempered indoor space (comfort HVAC / radiators)
-b_a0e6376410cb,r_77a0d3ba0f9c,d_bf217e522ef2,27.56915283203125,18.0,37.0,23.0,32.6768371582031,2.0933589935302734,0.9145450592041016,2.218053817749023,electrical/server room (constant heat load)
-b_a0e6376410cb,r_77a0d3ba0f9c,d_c39fe6bab820,20.44007110595703,0.0,36.28333282470703,10.30508518218994,29.711864471435547,4.613822937011719,2.759313583374023,10.278379440307615,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_e60f0e5e874f,d_0641d7913b47,18.102535247802734,0.0,47.0,9.0,25.634920120239254,6.704960823059082,4.861433029174805,7.211861610412598,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_e60f0e5e874f,d_2d2c5d1566f8,16.91263198852539,-6.712500095367432,42.66379165649414,6.0,28.158850860595702,7.999507904052734,5.124577522277832,12.458128929138184,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_e60f0e5e874f,d_93852a9fb47a,25.249731063842773,13.44444465637207,44.20000076293945,17.0,32.87116298675537,11.52781105041504,6.05531120300293,0.3563880920410156,electrical/server room (constant heat load)
-b_a0e6376410cb,r_f32d30daa895,d_7bd1921d0e3c,26.036542892456055,14.673469543457031,42.98387145996094,18.0,32.0,7.56397819519043,4.941301345825195,-3.235631942749023,AHU on weekday schedule (tempered plant room)
-b_a0e6376410cb,r_f32d30daa895,d_bd045b2e32a2,13.046881675720217,-8.0,34.0,2.0,22.0,5.374957084655762,4.547656059265137,12.400114059448242,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_a0e6376410cb,r_f32d30daa895,d_eb7b60348254,14.601313591003418,0.0,29.0,7.0,20.784285736083984,4.236031532287598,4.117071151733398,7.934282302856445,unheated/semi-outdoor space (vent-only)
-b_ad4a5872f2db,r_38ffec571713,d_350730de3316,22.84842872619629,19.53424644470215,44.20000076293945,20.52542304992676,26.649999618530273,3.5678043365478516,0.342905044555664,0.2699718475341797,tempered indoor space (comfort HVAC / radiators)
-b_ad4a5872f2db,r_38ffec571713,d_3d26e12378f1,19.078256607055664,-4.0,30.0,7.0,25.0,11.381299018859863,0.5470485687255859,5.475801467895508,AHU on weekday schedule (tempered plant room)
-b_ad4a5872f2db,r_38ffec571713,d_3f59e13bbfaf,17.115131378173828,-6.0,29.0,6.0,23.60646858215332,8.413435935974121,8.112293243408203,8.665217399597168,unheated/semi-outdoor space (vent-only)
-b_ad4a5872f2db,r_38ffec571713,d_575bd0b46798,19.59112548828125,17.0,26.0,18.0,22.0,1.2683601379394531,1.321136474609375,0.8474922180175781,tempered indoor space (comfort HVAC / radiators)
-b_ad4a5872f2db,r_38ffec571713,d_588011609899,22.28626251220703,14.0,35.0,18.0,27.0,6.914957046508789,0.4001960754394531,-0.3002185821533203,AHU on weekday schedule (tempered plant room)
-b_ad4a5872f2db,r_38ffec571713,d_98c5a7156a5e,22.601318359375,17.0,32.24324417114258,19.0,26.29833755493164,3.5189743041992188,2.9585189819335938,0.3984241485595703,tempered indoor space (comfort HVAC / radiators)
-b_ad4a5872f2db,r_38ffec571713,d_a702ad79bffb,17.68029022216797,-11.949999809265137,32.0,3.9151082277297977,23.0,11.48390293121338,0.4640827178955078,5.791308403015137,AHU on weekday schedule (unheated fan room / intake)
-b_ad4a5872f2db,r_38ffec571713,d_a85bae19463d,22.56890296936035,18.0,28.0,22.0,24.0,0.8378753662109375,0.6271171569824219,0.8616867065429688,tempered indoor space (comfort HVAC / radiators)
-b_ad4a5872f2db,r_38ffec571713,d_b56846e1ecf6,19.843677520751953,-6.0,36.08771896362305,8.2380952835083,27.0,4.444658279418945,3.9081058502197266,11.784070014953612,unheated/semi-outdoor space (vent-only)
-b_ad4a5872f2db,r_38ffec571713,d_c9a2ce794b68,20.14611053466797,16.482759475708008,25.0,19.0,22.0,1.431966781616211,0.2634487152099609,0.1689186096191406,tempered indoor space (comfort HVAC / radiators)
-b_ad4a5872f2db,r_38ffec571713,d_d9d695df1683,20.693456649780273,-4.0,35.88888931274414,8.0,26.0,8.711423873901367,0.3611888885498047,7.035398483276367,AHU on weekday schedule (tempered plant room)
-b_ad4a5872f2db,r_38ffec571713,d_ed28ebb86070,20.001340866088867,-5.0,35.06666564941406,7.539611291885377,25.983333587646484,7.296988487243652,0.4662151336669922,7.753772735595703,AHU on weekday schedule (tempered plant room)
-b_ad4a5872f2db,r_38ffec571713,d_f37ed1e83a5f,20.502477645874023,16.299999237060547,25.0,19.0,22.0,0.8345890045166016,0.3593463897705078,-0.6620082855224609,tempered indoor space (comfort HVAC / radiators)
-b_ad4a5872f2db,r_99bd12b98a34,d_124f1e85339f,30.41178321838379,21.11111068725586,49.0,24.0,38.731666946411096,2.4797840118408203,2.122516632080078,-9.478578567504885,heating equipment (winter duty)
-b_ad4a5872f2db,r_99bd12b98a34,d_357f39737767,22.844989776611328,18.25423812866211,25.0,21.0,24.0,0.5116119384765625,0.435373306274414,1.6271133422851562,tempered indoor space (comfort HVAC / radiators)
-b_ad4a5872f2db,r_99bd12b98a34,d_6e78c63aeb1e,22.908416748046875,18.33333396911621,25.0,21.0,24.0,0.5077400207519531,0.4224700927734375,1.547739028930664,tempered indoor space (comfort HVAC / radiators)
-b_ad4a5872f2db,r_99bd12b98a34,d_83ab792e75e4,32.2802848815918,19.827587127685547,35.0,27.0,34.0,1.029733657836914,0.5053138732910156,2.6652698516845703,electrical/server room (constant heat load)
-b_afdd8b1a066d,r_3a26ffb366db,d_07a29dc86c9f,17.175640106201172,-5.655172348022461,34.084747314453125,6.295833468437195,25.427430629730225,4.550929069519043,1.3258132934570312,11.649395942687988,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_3a26ffb366db,d_343dcb2826c7,20.255680084228516,12.0,30.827587127685547,16.0,24.40678024291992,3.1734542846679688,0.5332965850830078,5.223049163818359,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_3a26ffb366db,d_4aa47e2a60fe,18.91823959350586,8.0,34.28813552856445,14.0,23.0,2.778209686279297,0.6459293365478516,6.324509620666504,unheated/semi-outdoor space (vent-only)
-b_afdd8b1a066d,r_51e3a4545cad,d_48afe2506d60,23.52060317993164,14.699999809265137,39.0,19.0,34.016666412353516,5.964414596557617,3.2423019409179688,-1.8272762298583984,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_51e3a4545cad,d_9be15b20b63a,22.38367462158203,6.288135528564453,34.36666488647461,14.0,28.284053230285632,5.9241485595703125,2.8046607971191406,4.975702285766602,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_68e2b99fbf02,d_2fb7a37dadcd,22.424646377563477,15.0,31.28333282470703,18.0,27.0,0.2704010009765625,0.2249946594238281,5.0167083740234375,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_6dbe4292d4d7,d_5b949effa7b2,26.3875789642334,15.0,34.587154388427734,21.0,30.99034061431882,0.7063674926757812,0.4138126373291015,3.2672653198242188,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_6dbe4292d4d7,d_695002e46e72,28.46686553955078,19.0,39.14864730834961,24.0,33.98694152832031,0.47039794921875,0.302377700805664,-0.9927005767822266,electrical/server room (constant heat load)
-b_afdd8b1a066d,r_6dbe4292d4d7,d_abf323d095a5,29.50378036499023,18.0,40.0,25.0,34.48117027282713,0.505889892578125,0.3248481750488281,-1.021444320678711,electrical/server room (constant heat load)
-b_afdd8b1a066d,r_7d9d533d77cb,d_063c79be3016,18.37929344177246,-9.5625,37.900001525878906,5.0,27.16410255432128,5.329996109008789,3.173423767089844,13.324606895446776,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_19e17286a070,20.910457611083984,-4.97777795791626,38.84341812133789,7.430517196655273,29.209650993347168,5.150821685791016,3.5081329345703125,12.732205390930176,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_20b5bd087fd4,20.1632022857666,-7.111111164093018,33.32222366333008,9.0,26.0,7.633728981018066,4.181018829345703,2.812583923339844,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_2e144a557e02,22.128326416015625,12.587718963623049,37.69643020629883,16.126236534118654,27.53168125152588,6.467029571533203,4.070728302001953,3.3912887573242188,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_421ba7bf1017,18.807729721069336,-3.0,31.0,9.0,25.08997106552124,7.710260391235352,4.184925079345703,6.975255966186523,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_4a112772cc86,22.075944900512695,14.017241477966309,38.20000076293945,17.05646667480469,28.105240631103495,5.480070114135742,3.987335205078125,4.732692718505859,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_7d9d533d77cb,d_60979a1a7557,19.54648780822754,-9.420000076293944,35.0,3.7863096237182616,28.89245128631591,7.858083724975586,4.102529525756836,8.83682918548584,AHU on weekday schedule (unheated fan room / intake)
-b_afdd8b1a066d,r_7d9d533d77cb,d_65412692e8d8,21.178909301757812,-4.3684210777282715,38.907562255859375,8.548306846618653,29.678222656249993,5.312686920166016,3.743999481201172,12.256613731384276,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_68ddb9c04383,22.98233413696289,14.351351737976074,39.14414596557617,19.0,30.397431182861315,5.035602569580078,3.826095581054687,5.048423767089844,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_7d9d533d77cb,d_6ac18187fab8,20.98558807373047,4.0,47.07746505737305,11.05646562576294,32.9953779220581,5.734331130981445,2.5890941619873047,2.960655212402344,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_9f0e0e2d9148,18.61534118652344,8.0,34.80970001220703,14.195465278625488,24.0,7.811578750610352,3.863382339477539,2.41339111328125,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_ab77e1f67872,21.06841468811035,11.663793563842772,37.795555114746094,15.0,27.20855445861816,7.206335067749023,4.684823989868164,4.642030715942383,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_b0638b76cb66,20.19568252563477,-0.1111111119389534,36.89119338989258,14.956771850585938,27.0,1.7816658020019531,1.6437664031982422,6.80738639831543,unheated/semi-outdoor space (vent-only)
-b_afdd8b1a066d,r_7d9d533d77cb,d_d622ad296e2c,19.92671585083008,-7.201438903808594,33.41798782348633,10.63059139251709,25.0,5.062166213989258,2.9494571685791016,3.381166458129883,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_7d9d533d77cb,d_dd190a4f9355,20.548566818237305,-2.299999952316284,37.37614822387695,9.0,29.0,3.46114730834961,2.189249038696289,12.46059513092041,unheated/semi-outdoor space (vent-only)
-b_afdd8b1a066d,r_b90eecea9c7e,d_153b5d46b36b,32.8201904296875,20.0,45.0,23.0,42.0,14.458879470825195,6.780588150024414,3.6633472442626953,hot plant (boiler / DH substation / hot pipes)
-b_afdd8b1a066d,r_bcd426cb4dc2,d_4b83491730b1,21.391611099243164,15.383333206176758,33.0,18.0,26.282777404785143,5.908191680908203,2.8028125762939453,1.9076290130615232,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_bcd426cb4dc2,d_7139cdf34e3f,18.70762062072754,12.0,29.0,13.741821670532229,23.208225631713866,6.734949111938477,3.6359329223632817,1.4302253723144531,AHU on weekday schedule (tempered plant room)
-b_afdd8b1a066d,r_bcd426cb4dc2,d_75b7d0c1fee9,19.5544376373291,13.0,34.86440658569336,15.0,24.344263076782227,5.908725738525391,5.650356292724609,-1.053668975830078,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_bcd426cb4dc2,d_aa9ddf9d3867,23.230985641479492,16.0,35.375,20.0,26.63333320617676,1.255718231201172,0.8246574401855469,2.4876708984375,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_bcd426cb4dc2,d_fbc4ecb3674c,23.200796127319336,-1.0,51.0,10.684631109237673,41.90454959869385,10.72762680053711,6.223699569702148,15.698846817016602,hot plant (boiler / DH substation / hot pipes)
-b_afdd8b1a066d,r_d1e39dbbd605,d_33d44c40edcf,29.59678649902344,22.66666603088379,35.0,28.0,33.0,0.6726055145263672,0.5595588684082031,1.178386688232422,electrical/server room (constant heat load)
-b_afdd8b1a066d,r_d1e39dbbd605,d_5b130675e18c,24.461700439453125,19.0,32.30864334106445,23.0,28.0,0.4589900970458984,0.4393558502197265,2.0855979919433594,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_d1e39dbbd605,d_718d9aa4fc07,25.373376846313477,17.979999542236328,30.870370864868164,23.0,29.0,0.4324493408203125,0.3284988403320312,0.1084499359130859,electrical/server room (constant heat load)
-b_afdd8b1a066d,r_d1e39dbbd605,d_8e00a7af973d,26.59076881408691,20.0,34.0,24.122164916992187,31.0,0.6700763702392578,0.5634212493896484,-1.142059326171875,electrical/server room (constant heat load)
-b_afdd8b1a066d,r_d1e39dbbd605,d_bc75d2d68089,28.88253402709961,20.960784912109375,35.0,25.0,32.0,1.0479202270507812,0.9130821228027344,-0.3352489471435547,electrical/server room (constant heat load)
-b_afdd8b1a066d,r_d1e39dbbd605,d_db2988877013,28.384761810302734,21.352941513061523,35.0,26.0,32.0,0.6939125061035156,0.5196151733398438,0.3444652557373047,electrical/server room (constant heat load)
-b_afdd8b1a066d,r_ea358c0b18f2,d_320ecf4e93b7,21.440645217895508,13.733333587646484,27.899999618530273,17.0,26.0,3.6706295013427734,1.5823078155517578,-3.891447067260742,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_ea358c0b18f2,d_7881e067e23d,18.615158081054688,5.0333333015441895,29.5,12.0,24.0,2.7196178436279297,1.2625808715820312,6.322819709777832,unheated/semi-outdoor space (vent-only)
-b_afdd8b1a066d,r_ea358c0b18f2,d_9b4f45db9dbb,19.135238647460938,13.0,27.75,16.0,22.03333282470703,1.01470947265625,0.6020774841308594,3.681661605834961,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_f04bff6ae6b7,d_0fa23d252abf,20.07319450378418,16.0,31.03333282470703,18.0,22.15517234802246,0.5939960479736328,0.7212276458740234,0.3552627563476562,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_f04bff6ae6b7,d_21d63ad64295,20.951635360717773,1.0,52.412696838378906,7.0,44.43765888214111,7.223274230957031,6.238458633422852,20.841522216796875,hot plant (boiler / DH substation / hot pipes)
-b_afdd8b1a066d,r_f04bff6ae6b7,d_6d7dd78e506a,22.579225540161133,14.28125,30.91176414489746,19.04908514022827,25.242872714996334,1.0356674194335938,0.7206439971923828,2.7627601623535156,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_f04bff6ae6b7,d_9c81be30f5ac,24.27267265319824,18.0,34.88172149658203,19.59361915588379,29.0,1.138833999633789,1.1958332061767578,-1.0938034057617188,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_f04bff6ae6b7,d_fa3b18f14362,20.383930206298828,-1.0,51.344825744628906,8.0,42.0,6.715166091918945,6.565418243408203,19.58801651000977,hot plant (boiler / DH substation / hot pipes)
-b_afdd8b1a066d,r_f77e432fe269,d_22d7dbfeed76,24.826642990112305,10.083333015441896,32.6065559387207,17.91666603088379,29.0,3.0217342376708984,3.219118118286133,4.084932327270508,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_f77e432fe269,d_253d0f123532,18.20348358154297,9.366666793823242,27.350000381469727,13.0,23.0,8.477237701416016,8.413885116577148,-0.2676048278808594,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_f77e432fe269,d_5bcd3e4ebaef,23.93353271484375,22.0,41.71666717529297,23.0,26.840554332733152,1.272878646850586,1.0534000396728516,-1.2392082214355469,tempered indoor space (comfort HVAC / radiators)
-b_afdd8b1a066d,r_f77e432fe269,d_fff7c4acbe60,24.192604064941406,7.766666889190674,35.0,15.0,29.816667556762695,4.157417297363281,3.451143264770508,4.80061149597168,tempered indoor space (comfort HVAC / radiators)
-b_b57e86051e7e,r_103dff28bb7b,d_719d519dcc91,10.709359169006348,-14.0,28.327272415161133,-0.9787878870964044,21.33333396911621,5.492204666137695,1.5001559257507324,15.32200813293457,AHU on weekday schedule (unheated fan room / intake)
-b_b57e86051e7e,r_103dff28bb7b,d_871279fba467,11.45345401763916,-13.0,27.0,0.0,21.078672790527357,6.455916404724121,1.4730138778686523,14.425049781799316,AHU on weekday schedule (unheated fan room / intake)
-b_b57e86051e7e,r_103dff28bb7b,d_fa0a75ea9a67,11.6719331741333,-14.0,27.0,0.0,21.0,7.808426856994629,1.872406005859375,13.389366149902344,AHU on weekday schedule (unheated fan room / intake)
-b_b7e9d5634ad3,r_484904c730d7,d_3703f0819960,10.064297676086426,-14.0,26.700000762939453,-1.0,20.0,3.5134458541870117,3.3551454544067383,15.242825508117676,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_b7e9d5634ad3,r_484904c730d7,d_57e99ff00802,33.426979064941406,21.0,71.0,22.0,48.51666641235352,5.9800262451171875,6.521940231323242,0.4002037048339844,hot plant (boiler / DH substation / hot pipes)
-b_b7e9d5634ad3,r_484904c730d7,d_e1772c236402,32.737823486328125,21.0,62.894737243652344,22.0,47.26586189270018,5.976654052734375,5.391002655029297,0.368438720703125,hot plant (boiler / DH substation / hot pipes)
-b_b7e9d5634ad3,r_484904c730d7,d_f44ef99c6628,32.07578659057617,19.0,66.31034851074219,20.0,46.507204818725576,5.241447448730469,5.656639099121094,1.1315898895263672,hot plant (boiler / DH substation / hot pipes)
-b_b9acdf6071f3,r_8b5670cba0a2,d_1a1ecc949d00,13.674932479858398,-5.0,34.09961700439453,1.6796099066734318,23.0,4.347890853881836,4.928074836730957,14.274335861206056,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_b9acdf6071f3,r_8b5670cba0a2,d_321f6f3714a5,18.348468780517575,15.0,27.005952835083008,16.80114679336548,21.13997201919554,0.7471694946289062,0.932220458984375,-0.2281837463378906,tempered indoor space (comfort HVAC / radiators)
-b_b9acdf6071f3,r_8b5670cba0a2,d_bedac424bb48,13.008367538452148,-8.0,36.69643020629883,0.7562070786952978,20.850000381469727,3.16042423248291,3.547490119934082,12.74433135986328,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_b9acdf6071f3,r_e45cc68e00a2,d_4493250abcd3,12.80185604095459,-6.267516136169434,33.016666412353516,2.0,22.817457008361806,4.306715965270996,4.503781318664551,13.094619750976562,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_b9acdf6071f3,r_e45cc68e00a2,d_88ce8bc64560,15.23102855682373,-6.0,36.74418640136719,1.979952800273896,23.033361053466795,2.908860206604004,4.005146026611328,12.905418395996094,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_b9acdf6071f3,r_e45cc68e00a2,d_a9e869083651,11.883903503417969,-7.0,33.06666564941406,0.6314689218997959,22.93458271026611,4.306100845336914,4.508148193359375,13.329483032226562,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_b9acdf6071f3,r_e45cc68e00a2,d_c76f521a00b2,18.583328247070312,13.685315132141112,26.126983642578125,17.0,20.582263946533207,0.5234413146972656,0.8279094696044922,0.566436767578125,tempered indoor space (comfort HVAC / radiators)
-b_b9acdf6071f3,r_e45cc68e00a2,d_d64d5d2342bb,18.904878616333008,13.721311569213867,26.816667556762695,17.0,20.491525650024418,0.5938358306884766,0.7298831939697266,1.033975601196289,tempered indoor space (comfort HVAC / radiators)
-b_be0092b8c92e,r_2781546d7942,d_11dc61654452,21.749755859375,0.3853211104869842,40.0,11.101123809814451,31.117021560668945,2.919748306274414,3.3665122985839844,12.18596076965332,unheated/semi-outdoor space (vent-only)
-b_be0092b8c92e,r_2781546d7942,d_3b9d4c40eb81,24.13689422607422,16.0,40.62626266479492,20.786884307861328,29.0,1.756914138793945,1.5889015197753906,1.657663345336914,tempered indoor space (comfort HVAC / radiators)
-b_be0092b8c92e,r_2781546d7942,d_76b6e22bbcb7,21.36078834533691,0.9485294222831726,40.88118743896485,10.0,31.94495391845703,4.580160140991211,4.225452423095703,15.073039054870604,unheated/semi-outdoor space (vent-only)
-b_be0092b8c92e,r_279cb4f4fc29,d_586c0414175c,12.998275756835938,-10.084745407104492,42.0,0.0,25.299999237060547,6.259702682495117,6.072766304016113,15.502180099487305,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_be0092b8c92e,r_279cb4f4fc29,d_a431a5e84fcd,17.15972900390625,-7.966666698455811,44.0,4.725306749343872,26.0,3.48922348022461,3.439056396484375,11.175566673278809,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_be0092b8c92e,r_279cb4f4fc29,d_d20225121466,22.015748977661133,12.980392456054688,34.03333282470703,19.0,25.0,0.7013950347900391,0.6931991577148438,-0.0232524871826171,tempered indoor space (comfort HVAC / radiators)
-b_be0092b8c92e,r_38045b3776de,d_32458a23f61c,23.175893783569336,15.0,37.0,19.21449022293091,27.491713905334475,1.131326675415039,0.933490753173828,0.4438095092773437,tempered indoor space (comfort HVAC / radiators)
-b_be0092b8c92e,r_38045b3776de,d_469285226e03,21.551294326782227,6.480769157409668,46.0,14.035004997253418,26.657876110076906,1.100383758544922,1.1244525909423828,5.357625961303711,tempered indoor space (comfort HVAC / radiators)
-b_be0092b8c92e,r_38045b3776de,d_f57c1450927f,23.39523124694824,1.0,44.13043594360352,10.649999618530272,32.1592903137207,2.787343978881836,2.6310958862304688,11.596471786499023,unheated/semi-outdoor space (vent-only)
-b_be0092b8c92e,r_829b4d35c026,d_3a7d369c9aee,23.13812255859375,5.162499904632568,37.0,17.695108699798585,28.0,1.3806114196777344,1.554828643798828,-0.1603469848632812,tempered indoor space (comfort HVAC / radiators)
-b_be0092b8c92e,r_829b4d35c026,d_8ba95cfeaaa6,18.972122192382812,-8.181818008422852,50.0,4.367522430419923,31.0,5.854375839233398,5.115383148193359,12.183226585388184,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_be0092b8c92e,r_829b4d35c026,d_e330916ab84b,18.824626922607425,-9.94444465637207,50.0,3.621505260467529,30.0,3.892635345458984,3.733549118041992,8.351956367492676,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_be0092b8c92e,r_a6ff229f5d88,d_1269eb694e96,23.91531753540039,8.42391300201416,39.0,17.0,32.0,2.7740306854248047,2.781085968017578,6.956708908081055,unheated/semi-outdoor space (vent-only)
-b_be0092b8c92e,r_a6ff229f5d88,d_50d206f364b2,23.411149978637695,15.631068229675291,34.2260856628418,19.773672008514403,28.0,1.197235107421875,1.3168964385986328,-0.0404815673828125,tempered indoor space (comfort HVAC / radiators)
-b_be0092b8c92e,r_a6ff229f5d88,d_c9f1e52bf751,21.032804489135746,2.0,38.0,10.0,31.0,4.9825286865234375,4.66026496887207,14.530720710754396,unheated/semi-outdoor space (vent-only)
-b_be0092b8c92e,r_f1736c7a40b7,d_14ff7b047c65,18.42421531677246,-2.0,40.33898162841797,7.285254192352295,33.599998474121094,2.575572967529297,2.63336181640625,13.998181343078612,unheated/semi-outdoor space (vent-only)
-b_be0092b8c92e,r_f1736c7a40b7,d_8f262c1bda75,21.742494583129883,10.0,30.89655113220215,18.041881561279297,24.784762191772455,0.9371490478515624,0.7974319458007812,-0.448699951171875,tempered indoor space (comfort HVAC / radiators)
-b_be0092b8c92e,r_f1736c7a40b7,d_9f3c85e3a7e1,13.911396026611328,-8.203390121459961,48.96154022216797,0.8783333301544197,29.0,6.64543342590332,6.120530128479004,17.77406120300293,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_c2dc4330feb5,r_65e639cdbefd,d_45de2f2188fa,17.550979614257812,1.0,32.61666488647461,9.0,24.96666717529297,3.058347702026367,1.1936779022216797,9.840798377990724,unheated/semi-outdoor space (vent-only)
-b_c2dc4330feb5,r_65e639cdbefd,d_ba27a8d32e0c,21.35660171508789,-4.0,37.11666488647461,7.0,28.48296041488647,3.710271835327149,0.8432712554931641,9.302997589111328,unheated/semi-outdoor space (vent-only)
-b_c2dc4330feb5,r_65e639cdbefd,d_fdcf3fa44b07,10.69318675994873,-10.100000381469728,31.549999237060547,-1.0,24.0,5.144369125366211,3.8162031173706055,16.704729080200195,"outdoor-air coupled (intake / unheated space, no schedule)"
-b_c376fc4d6ccb,r_39dfb37da081,d_0d8e1768cb3a,14.32970142364502,2.0,29.0,6.0,22.63157844543457,6.693129539489746,3.304469108581543,,AHU on weekday schedule (tempered plant room)
-b_c376fc4d6ccb,r_39dfb37da081,d_12deecea69dd,15.808260917663574,3.0,29.227941513061523,7.416087245941163,23.12491188049316,8.132908821105957,2.865018844604492,,AHU on weekday schedule (tempered plant room)
-b_c376fc4d6ccb,r_39dfb37da081,d_be0fde6cb60d,16.11652946472168,3.0,26.0,8.0,22.2799708366394,7.149377822875977,3.40195369720459,,AHU on weekday schedule (tempered plant room)
-b_ce7925d56654,r_0620a16c3ded,d_36ad37928c5a,22.353919982910156,14.810344696044922,34.49152374267578,18.325989532470704,27.653361892700183,5.857509613037109,3.1679458618164062,3.1739368438720703,AHU on weekday schedule (tempered plant room)
-b_ce7925d56654,r_0620a16c3ded,d_db3f4d1a222e,19.9525146484375,2.0,50.29824447631836,10.0,32.24845142364501,2.9156627655029297,1.7325668334960938,13.68507480621338,unheated/semi-outdoor space (vent-only)
-b_ce7925d56654,r_1bb7730fcee1,d_071ecf7d4231,24.288299560546875,17.071428298950195,30.566667556762695,21.60957384109497,27.33333396911621,0.7572898864746094,0.6144275665283203,3.4217605590820312,tempered indoor space (comfort HVAC / radiators)
-b_ce7925d56654,r_1bb7730fcee1,d_0befe288584a,25.77641487121582,19.03333282470703,33.74137878417969,22.950000762939453,29.41304397583008,0.9263648986816406,0.6142024993896484,3.603811264038086,tempered indoor space (comfort HVAC / radiators)
-b_ce7925d56654,r_1bb7730fcee1,d_3db330ba4098,24.009008407592773,17.0,30.350000381469727,21.40678024291992,27.113362312316895,0.78753662109375,0.6399021148681641,3.4528942108154297,tempered indoor space (comfort HVAC / radiators)
-b_ce7925d56654,r_1bb7730fcee1,d_600bd93ceff1,23.543365478515625,11.0,41.52727127075195,19.5,30.279008293151843,9.783353805541992,3.956365585327149,0.3368663787841797,AHU on weekday schedule (tempered plant room)
-b_ce7925d56654,r_1bb7730fcee1,d_8c43f9975289,25.930747985839844,19.06779670715332,33.25,23.0,29.486111640930176,0.9125385284423828,0.5922679901123047,3.54730224609375,tempered indoor space (comfort HVAC / radiators)
-b_ce7925d56654,r_1bb7730fcee1,d_9fd6f7d6469b,18.246841430664062,6.38095235824585,35.03845977783203,13.072861862182616,24.922925758361817,5.44141960144043,2.8701066970825195,4.413665771484375,AHU on weekday schedule (tempered plant room)
-b_ce7925d56654,r_1bb7730fcee1,d_b94797370f49,26.55043601989746,7.599999904632568,55.46428680419922,14.283333778381348,48.0,9.40432357788086,8.276060104370117,17.22224235534668,hot plant (boiler / DH substation / hot pipes)
-b_ce7925d56654,r_1bb7730fcee1,d_bfbe0f417642,20.71566390991211,9.0,39.47457504272461,16.0,25.15165224075317,2.580970764160156,2.4565486907958984,6.21734619140625,unheated/semi-outdoor space (vent-only)
-b_ce7925d56654,r_1bb7730fcee1,d_d1529e657cbe,22.367523193359375,8.533333778381348,50.53061294555664,14.0,36.0,9.886911392211914,3.4344539642333984,11.793983459472656,AHU on weekday schedule (tempered plant room)
-b_ce7925d56654,r_1bb7730fcee1,d_fe6233c68a67,20.708641052246097,13.0,44.0,17.0,34.0,1.450237274169922,1.634050369262695,4.999774932861328,tempered indoor space (comfort HVAC / radiators)
-b_ce7925d56654,r_6beebf5e37bf,d_6cfe45359687,20.68883895874023,8.0,32.0,14.0,25.189223289489743,4.159399032592773,2.581104278564453,4.334148406982422,AHU on weekday schedule (tempered plant room)
-b_ce7925d56654,r_6beebf5e37bf,d_9af963fa9017,21.240203857421875,-2.0,47.230770111083984,9.0,38.0,10.00107765197754,6.953521728515625,17.274890899658203,AHU on weekday schedule (tempered plant room)
-b_ce7925d56654,r_6beebf5e37bf,d_b541fda0e6c0,20.1202335357666,6.766666889190674,32.0,14.0,25.0,5.48713493347168,3.247526168823242,1.0878753662109375,AHU on weekday schedule (tempered plant room)
-b_ce7925d56654,r_6beebf5e37bf,d_c0b59014fcd8,19.132883071899418,0.0,57.0,7.031157779693604,37.502127838134776,4.860345840454102,3.2780075073242188,17.392492294311523,AHU on weekday schedule (tempered plant room)
-b_d45195c31e8f,r_37ca07e35f31,d_497cf69db25f,17.071674346923828,2.0,71.63333129882812,5.0,28.683332443237305,9.143454551696776,1.786005973815918,14.041909217834473,AHU on weekday schedule (tempered plant room)
-b_d45195c31e8f,r_37ca07e35f31,d_f11c15757bf1,17.913393020629883,2.0,95.0,5.0,28.983333587646484,10.466458320617676,1.6841545104980469,11.870861053466797,AHU on weekday schedule (tempered plant room)
-b_dabc992b5f46,r_f13f5c3c9f5f,d_172fcf48ec9e,23.286230087280273,15.453038215637209,38.155174255371094,18.0,29.549999237060547,9.29738426208496,0.6651668548583984,0.4031219482421875,AHU on weekday schedule (tempered plant room)
-b_dabc992b5f46,r_f13f5c3c9f5f,d_7fb0c50f956d,21.42535400390625,-3.9160306453704834,55.0,8.81910552978516,28.25695514678954,10.984912872314451,0.6708450317382812,7.144880294799805,AHU on weekday schedule (tempered plant room)
-b_dabc992b5f46,r_f13f5c3c9f5f,d_abbd3c7517a2,19.10573387145996,-16.25423812866211,55.0,2.367499899864198,27.0,14.45495891571045,0.7660579681396484,5.743873596191406,AHU on weekday schedule (unheated fan room / intake)
-b_dabc992b5f46,r_f13f5c3c9f5f,d_db6b21a2bcd5,32.901939392089844,20.0,49.0,24.36333351135254,40.0,3.492664337158203,0.7008857727050781,-2.15401840209961,hot plant (boiler / DH substation / hot pipes)
-b_fcce8f31d315,r_5f10a21ecd58,d_184be698fc36,11.664824485778809,-12.0,33.45000076293945,-2.0,25.0,6.559528350830078,2.6753358840942383,18.438655853271484,AHU on weekday schedule (unheated fan room / intake)
-b_fcce8f31d315,r_5f10a21ecd58,d_9ab87ed4625d,14.285085678100586,-11.0,31.46666717529297,0.3661206841468831,23.57098913192748,10.680171966552734,2.4756393432617188,12.451053619384766,AHU on weekday schedule (unheated fan room / intake)
-b_fcce8f31d315,r_5f10a21ecd58,d_c90f83dd9ea8,12.937211990356444,-13.0,31.450000762939453,-2.0,25.0,11.354790687561035,2.426527976989746,14.630364418029783,AHU on weekday schedule (unheated fan room / intake)

research/hvac_origin_features.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:440ab983ffae1595cd1eef907b3902ed7fbebc1778685edf4b9c097db97282e7
-size 1056311

research/hvac_origin_features.py

@@ -1,70 +0,0 @@
-"""Leak-free environment (HVAC-class) features per (device_id, origin_idx).
-
-Motivation (2026-08-19 EDA, research/hvac_room_profiles.py): device temperature
-signatures classify the equipment environment, and death rates are U-shaped in
-it - cold classes (AHU intake/fan rooms) lose 25-40% of batteries, hot plants
-20-30%, mild indoor 10%. Monotone temp features (phases 6/8/13) cannot express
-the U; two separate exposure arms + a schedule signature can.
-
-Unlike the full-history EDA class labels, everything here is EXPANDING: at each
-features.parquet origin only temperature data <= origin is used, so the columns
-are protocol-clean (features use only data <= origin) and runtime-computable.
-
-Columns (per device, life-so-far as of origin):
-  cold_frac  - fraction of observed days whose min temperature < 5 C
-  hot_frac   - fraction of observed days whose max temperature > 32 C
-  sched_amp  - mean weekday daily temp range minus mean weekend daily range
-               (weekday-scheduled AHU signature; ~0 for unscheduled spaces)
-
-Output: research/hvac_origin_features.parquet keyed (device_id, origin_idx).
-Run: venv/bin/python research/hvac_origin_features.py
-"""
-from pathlib import Path
-
-import numpy as np
-import pandas as pd
-
-HVAC_FEATURES = ['cold_frac', 'hot_frac', 'sched_amp']
-
-
-def build(dataset_path: Path = Path('dataset')) -> pd.DataFrame:
-    m = pd.read_parquet(dataset_path / 'train' / 'battery_metrics.parquet',
-                        columns=['end_time', 'temperature', 'device_id'])
-    m = m.dropna(subset=['temperature'])
-    m['device_id'] = m['device_id'].astype(str)
-    m['day'] = m['end_time'].dt.normalize()
-
-    g = m.groupby(['device_id', 'day'])['temperature']
-    daily = pd.DataFrame({'t_min': g.min(), 't_max': g.max()}).reset_index()
-    daily['cold'] = daily['t_min'] < 5.0
-    daily['hot'] = daily['t_max'] > 32.0
-    daily['rng'] = daily['t_max'] - daily['t_min']
-    daily['we'] = daily['day'].dt.dayofweek >= 5
-    daily = daily.sort_values('day')
-
-    table = pd.read_parquet(dataset_path / 'train' / 'features.parquet',
-                            columns=['origin_idx', 'origin'])
-    origins = table.drop_duplicates().sort_values('origin_idx')
-
-    rows = []
-    for _, o in origins.iterrows():
-        h = daily[daily['day'] <= o['origin']]
-        gd = h.groupby('device_id')
-        f = pd.DataFrame({
-            'cold_frac': gd['cold'].mean(),
-            'hot_frac': gd['hot'].mean(),
-        })
-        wd = h[~h['we']].groupby('device_id')['rng'].mean()
-        we = h[h['we']].groupby('device_id')['rng'].mean()
-        f['sched_amp'] = wd - we
-        f['origin_idx'] = o['origin_idx']
-        rows.append(f.reset_index())
-    out = pd.concat(rows, ignore_index=True)
-    out.to_parquet(Path('research') / 'hvac_origin_features.parquet')
-    return out
-
-
-if __name__ == '__main__':
-    t = build()
-    print(t.groupby('origin_idx').size().describe().round(1).to_string())
-    print(t[HVAC_FEATURES].describe().round(3).to_string())

research/hvac_room_classes.csv

@@ -1,80 +0,0 @@
-building_id,room_id,n,room_guess,mix
-b_15bc00efc223,r_89c3bc35e31a,3,AHU on weekday schedule (unheated fan room / intake),3x AHU on weekday schedule
-b_15bc00efc223,r_8fa484a8c87f,2,AHU on weekday schedule (unheated fan room / intake),2x AHU on weekday schedule
-b_2b69c4007582,r_268fb9378735,6,unheated/semi-outdoor space (vent-only),4x unheated/semi-outdoor space + 2x tempered indoor space
-b_2b69c4007582,r_7c97db8498cc,28,AHU on weekday schedule (tempered plant room),13x AHU on weekday schedule + 7x unheated/semi-outdoor space + 4x tempered indoor space + 3x AHU on weekday schedule + 1x electrical/server room
-b_2b69c4007582,r_a86a82a16ca0,17,tempered indoor space (comfort HVAC / radiators),5x tempered indoor space + 5x unheated/semi-outdoor space + 3x AHU on weekday schedule + 2x outdoor-air coupled + 2x AHU on weekday schedule
-b_2b69c4007582,r_e1bac19501a4,14,tempered indoor space (comfort HVAC / radiators),7x tempered indoor space + 5x outdoor-air coupled + 1x hot plant + 1x unheated/semi-outdoor space
-b_4217a3b6e58b,r_84ab2c9a82b8,1,unheated/semi-outdoor space (vent-only),1x unheated/semi-outdoor space
-b_4217a3b6e58b,r_a36e0de3c5c2,8,tempered indoor space (comfort HVAC / radiators),5x tempered indoor space + 3x AHU on weekday schedule
-b_4217a3b6e58b,r_bd4c57fb510d,11,tempered indoor space (comfort HVAC / radiators),6x tempered indoor space + 2x AHU on weekday schedule + 2x unheated/semi-outdoor space + 1x electrical/server room
-b_4217a3b6e58b,r_cdb37c5edd19,3,tempered indoor space (comfort HVAC / radiators),2x tempered indoor space + 1x AHU on weekday schedule
-b_53df020ad893,r_4d95939c1d71,3,tempered indoor space (comfort HVAC / radiators),2x tempered indoor space + 1x unheated/semi-outdoor space
-b_53df020ad893,r_9b5c4b767d88,3,"outdoor-air coupled (intake / unheated space, no schedule)",1x outdoor-air coupled + 1x tempered indoor space + 1x electrical/server room
-b_55038c6689a0,r_ddd012cdb553,10,AHU on weekday schedule (tempered plant room),7x AHU on weekday schedule + 1x tempered indoor space + 1x unheated/semi-outdoor space + 1x hot plant
-b_60c6d1796ab4,r_b4c6b2670bd8,4,tempered indoor space (comfort HVAC / radiators),2x tempered indoor space + 1x unheated/semi-outdoor space + 1x electrical/server room
-b_60c6d1796ab4,r_e301f99df217,1,tempered indoor space (comfort HVAC / radiators),1x tempered indoor space
-b_62e09ab003f8,r_5bd522ea865e,6,tempered indoor space (comfort HVAC / radiators),3x tempered indoor space + 2x outdoor-air coupled + 1x AHU on weekday schedule
-b_62e09ab003f8,r_5c8cfab78cad,27,tempered indoor space (comfort HVAC / radiators),14x tempered indoor space + 4x heating equipment + 4x AHU on weekday schedule + 3x electrical/server room + 2x unheated/semi-outdoor space
-b_6869b0f15a78,r_0facb54e227f,2,hot plant (boiler / DH substation / hot pipes),1x hot plant + 1x heating equipment
-b_6869b0f15a78,r_974c30d91109,6,hot plant (boiler / DH substation / hot pipes),6x hot plant
-b_6869b0f15a78,r_ff96a3eef647,3,"outdoor-air coupled (intake / unheated space, no schedule)",2x outdoor-air coupled + 1x unheated/semi-outdoor space
-b_7aac45259c8a,r_ab8e8ce3ee18,4,tempered indoor space (comfort HVAC / radiators),3x tempered indoor space + 1x unheated/semi-outdoor space
-b_955300d5f090,r_02e6b266abf6,5,"outdoor-air coupled (intake / unheated space, no schedule)",3x outdoor-air coupled + 2x tempered indoor space
-b_955300d5f090,r_0d59c2f44f40,33,AHU on weekday schedule (tempered plant room),11x AHU on weekday schedule + 10x unheated/semi-outdoor space + 9x tempered indoor space + 3x outdoor-air coupled
-b_955300d5f090,r_27b0cfa20d53,5,"outdoor-air coupled (intake / unheated space, no schedule)",3x outdoor-air coupled + 2x tempered indoor space
-b_955300d5f090,r_2823849bded1,8,unheated/semi-outdoor space (vent-only),4x unheated/semi-outdoor space + 2x AHU on weekday schedule + 1x tempered indoor space + 1x outdoor-air coupled
-b_955300d5f090,r_5b4caff693e8,3,unheated/semi-outdoor space (vent-only),3x unheated/semi-outdoor space
-b_955300d5f090,r_786737d33818,4,"outdoor-air coupled (intake / unheated space, no schedule)",2x outdoor-air coupled + 2x tempered indoor space
-b_955300d5f090,r_9b5477e4058b,3,AHU on weekday schedule (unheated fan room / intake),2x AHU on weekday schedule + 1x unheated/semi-outdoor space
-b_955300d5f090,r_9cb7e6657012,8,unheated/semi-outdoor space (vent-only),6x unheated/semi-outdoor space + 2x tempered indoor space
-b_955300d5f090,r_9ea33a360e32,19,AHU on weekday schedule (tempered plant room),11x AHU on weekday schedule + 6x unheated/semi-outdoor space + 2x outdoor-air coupled
-b_955300d5f090,r_baa7dd603600,3,"outdoor-air coupled (intake / unheated space, no schedule)",2x outdoor-air coupled + 1x AHU on weekday schedule
-b_955300d5f090,r_ca9045f4fef2,4,"outdoor-air coupled (intake / unheated space, no schedule)",2x outdoor-air coupled + 2x tempered indoor space
-b_955300d5f090,r_cb32bb120909,2,heating equipment (winter duty),2x heating equipment
-b_955300d5f090,r_e156e26e7583,3,unheated/semi-outdoor space (vent-only),2x unheated/semi-outdoor space + 1x tempered indoor space
-b_955300d5f090,r_e292e2e62ec3,5,unheated/semi-outdoor space (vent-only),4x unheated/semi-outdoor space + 1x tempered indoor space
-b_955300d5f090,r_e2a69ae237a4,5,"outdoor-air coupled (intake / unheated space, no schedule)",2x outdoor-air coupled + 2x tempered indoor space + 1x unheated/semi-outdoor space
-b_955300d5f090,r_f2d7d812a5f0,2,tempered indoor space (comfort HVAC / radiators),1x tempered indoor space + 1x outdoor-air coupled
-b_955300d5f090,r_f7088836b3c8,5,"outdoor-air coupled (intake / unheated space, no schedule)",3x outdoor-air coupled + 1x electrical/server room + 1x tempered indoor space
-b_9a5a45d6e498,r_101aa0c8e52c,3,AHU on weekday schedule (tempered plant room),2x AHU on weekday schedule + 1x AHU on weekday schedule
-b_9a5a45d6e498,r_22232f8e4b91,3,unheated/semi-outdoor space (vent-only),1x unheated/semi-outdoor space + 1x AHU on weekday schedule + 1x AHU on weekday schedule
-b_9a5a45d6e498,r_6359aa838297,15,"outdoor-air coupled (intake / unheated space, no schedule)",8x outdoor-air coupled + 4x AHU on weekday schedule + 3x AHU on weekday schedule
-b_9a5a45d6e498,r_d1124136aa68,3,AHU on weekday schedule (unheated fan room / intake),2x AHU on weekday schedule + 1x outdoor-air coupled
-b_a0e6376410cb,r_0524a19c4bfd,4,AHU on weekday schedule (tempered plant room),1x AHU on weekday schedule + 1x tempered indoor space + 1x outdoor-air coupled + 1x unheated/semi-outdoor space
-b_a0e6376410cb,r_66f8934ecd8c,5,tempered indoor space (comfort HVAC / radiators),2x tempered indoor space + 2x AHU on weekday schedule + 1x AHU on weekday schedule
-b_a0e6376410cb,r_770fc166ffa0,6,AHU on weekday schedule (tempered plant room),4x AHU on weekday schedule + 2x electrical/server room
-b_a0e6376410cb,r_77a0d3ba0f9c,8,"outdoor-air coupled (intake / unheated space, no schedule)",4x outdoor-air coupled + 2x tempered indoor space + 1x electrical/server room + 1x AHU on weekday schedule
-b_a0e6376410cb,r_e60f0e5e874f,3,AHU on weekday schedule (tempered plant room),2x AHU on weekday schedule + 1x electrical/server room
-b_a0e6376410cb,r_f32d30daa895,3,AHU on weekday schedule (tempered plant room),1x AHU on weekday schedule + 1x outdoor-air coupled + 1x unheated/semi-outdoor space
-b_ad4a5872f2db,r_38ffec571713,13,tempered indoor space (comfort HVAC / radiators),6x tempered indoor space + 4x AHU on weekday schedule + 2x unheated/semi-outdoor space + 1x AHU on weekday schedule
-b_ad4a5872f2db,r_99bd12b98a34,4,tempered indoor space (comfort HVAC / radiators),2x tempered indoor space + 1x heating equipment + 1x electrical/server room
-b_afdd8b1a066d,r_3a26ffb366db,3,AHU on weekday schedule (tempered plant room),1x AHU on weekday schedule + 1x tempered indoor space + 1x unheated/semi-outdoor space
-b_afdd8b1a066d,r_51e3a4545cad,2,AHU on weekday schedule (tempered plant room),2x AHU on weekday schedule
-b_afdd8b1a066d,r_68e2b99fbf02,1,tempered indoor space (comfort HVAC / radiators),1x tempered indoor space
-b_afdd8b1a066d,r_6dbe4292d4d7,3,electrical/server room (constant heat load),2x electrical/server room + 1x tempered indoor space
-b_afdd8b1a066d,r_7d9d533d77cb,15,AHU on weekday schedule (tempered plant room),10x AHU on weekday schedule + 2x tempered indoor space + 2x unheated/semi-outdoor space + 1x AHU on weekday schedule
-b_afdd8b1a066d,r_b90eecea9c7e,1,hot plant (boiler / DH substation / hot pipes),1x hot plant
-b_afdd8b1a066d,r_bcd426cb4dc2,5,AHU on weekday schedule (tempered plant room),2x AHU on weekday schedule + 2x tempered indoor space + 1x hot plant
-b_afdd8b1a066d,r_d1e39dbbd605,6,electrical/server room (constant heat load),5x electrical/server room + 1x tempered indoor space
-b_afdd8b1a066d,r_ea358c0b18f2,3,tempered indoor space (comfort HVAC / radiators),2x tempered indoor space + 1x unheated/semi-outdoor space
-b_afdd8b1a066d,r_f04bff6ae6b7,5,tempered indoor space (comfort HVAC / radiators),3x tempered indoor space + 2x hot plant
-b_afdd8b1a066d,r_f77e432fe269,4,tempered indoor space (comfort HVAC / radiators),4x tempered indoor space
-b_b57e86051e7e,r_103dff28bb7b,3,AHU on weekday schedule (unheated fan room / intake),3x AHU on weekday schedule
-b_b7e9d5634ad3,r_484904c730d7,4,hot plant (boiler / DH substation / hot pipes),3x hot plant + 1x outdoor-air coupled
-b_b9acdf6071f3,r_8b5670cba0a2,3,"outdoor-air coupled (intake / unheated space, no schedule)",2x outdoor-air coupled + 1x tempered indoor space
-b_b9acdf6071f3,r_e45cc68e00a2,5,"outdoor-air coupled (intake / unheated space, no schedule)",3x outdoor-air coupled + 2x tempered indoor space
-b_be0092b8c92e,r_2781546d7942,3,unheated/semi-outdoor space (vent-only),2x unheated/semi-outdoor space + 1x tempered indoor space
-b_be0092b8c92e,r_279cb4f4fc29,3,"outdoor-air coupled (intake / unheated space, no schedule)",2x outdoor-air coupled + 1x tempered indoor space
-b_be0092b8c92e,r_38045b3776de,3,tempered indoor space (comfort HVAC / radiators),2x tempered indoor space + 1x unheated/semi-outdoor space
-b_be0092b8c92e,r_829b4d35c026,3,"outdoor-air coupled (intake / unheated space, no schedule)",2x outdoor-air coupled + 1x tempered indoor space
-b_be0092b8c92e,r_a6ff229f5d88,3,unheated/semi-outdoor space (vent-only),2x unheated/semi-outdoor space + 1x tempered indoor space
-b_be0092b8c92e,r_f1736c7a40b7,3,unheated/semi-outdoor space (vent-only),1x unheated/semi-outdoor space + 1x tempered indoor space + 1x outdoor-air coupled
-b_c2dc4330feb5,r_65e639cdbefd,3,unheated/semi-outdoor space (vent-only),2x unheated/semi-outdoor space + 1x outdoor-air coupled
-b_c376fc4d6ccb,r_39dfb37da081,3,AHU on weekday schedule (tempered plant room),3x AHU on weekday schedule
-b_ce7925d56654,r_0620a16c3ded,2,AHU on weekday schedule (tempered plant room),1x AHU on weekday schedule + 1x unheated/semi-outdoor space
-b_ce7925d56654,r_1bb7730fcee1,10,tempered indoor space (comfort HVAC / radiators),5x tempered indoor space + 3x AHU on weekday schedule + 1x hot plant + 1x unheated/semi-outdoor space
-b_ce7925d56654,r_6beebf5e37bf,4,AHU on weekday schedule (tempered plant room),4x AHU on weekday schedule
-b_d45195c31e8f,r_37ca07e35f31,2,AHU on weekday schedule (tempered plant room),2x AHU on weekday schedule
-b_dabc992b5f46,r_f13f5c3c9f5f,4,AHU on weekday schedule (tempered plant room),2x AHU on weekday schedule + 1x AHU on weekday schedule + 1x hot plant
-b_fcce8f31d315,r_5f10a21ecd58,3,AHU on weekday schedule (unheated fan room / intake),3x AHU on weekday schedule

research/hvac_room_profiles.csv

@@ -1,80 +0,0 @@
-building_id,room_id,n_devices,n_obs,t_mean,t_p05,t_p95,t_std,diurnal_wd,diurnal_we,seasonal,weekend_dip,hvac_guess
-b_b57e86051e7e,r_103dff28bb7b,3.0,66319.0,11.276205062866211,0.0,21.101694107055664,6.9822916984558105,5.450651168823242,1.6132917404174805,14.386013984680176,3.130769729614258,"scheduled ventilation (AHU, weekday operating hours)"
-b_15bc00efc223,r_89c3bc35e31a,3.0,53715.0,11.653122901916504,-2.683333396911621,23.5,8.279617309570312,9.701985359191895,2.2121543884277344,15.632421493530273,4.054306983947754,"scheduled ventilation (AHU, weekday operating hours)"
-b_15bc00efc223,r_8fa484a8c87f,2.0,43575.0,12.506261825561523,-2.0,25.071927261352535,8.60757827758789,12.036898612976074,3.9027843475341797,16.2408504486084,3.804507255554199,"scheduled ventilation (AHU, weekday operating hours)"
-b_fcce8f31d315,r_5f10a21ecd58,3.0,52129.0,12.829794883728027,-1.0,24.730128479003906,8.371939659118652,9.278406143188477,2.530604362487793,15.503957748413086,3.9082260131835938,"scheduled ventilation (AHU, weekday operating hours)"
-b_955300d5f090,r_baa7dd603600,3.0,22813.0,13.030999183654785,2.0,24.0,6.816330909729004,6.985386848449707,5.2711029052734375,14.05884075164795,1.609175682067871,"scheduled ventilation (AHU, weekday operating hours)"
-b_6869b0f15a78,r_ff96a3eef647,3.0,42732.0,14.26303482055664,3.0,24.0,6.283768177032471,2.9016199111938477,2.5665111541748047,12.585808753967285,0.18584918975830078,unconditioned/vent-only space (tracks outdoor seasons)
-b_9a5a45d6e498,r_d1124136aa68,3.0,60131.0,14.392990112304688,0.0,25.40186882019043,8.06118392944336,8.496270179748535,3.880277633666992,13.456748962402344,2.502230644226074,"scheduled ventilation (AHU, weekday operating hours)"
-b_9a5a45d6e498,r_6359aa838297,15.0,310701.0,14.63676929473877,0.0,25.5,8.185598373413086,6.5585784912109375,4.417290687561035,14.158870697021484,1.376143455505371,"scheduled ventilation (AHU, weekday operating hours)"
-b_b9acdf6071f3,r_8b5670cba0a2,3.0,54708.0,15.024746894836426,3.0,22.5975248336792,6.055808067321777,2.7288074493408203,3.1013975143432617,8.861148834228516,-0.09499073028564453,unconditioned/vent-only space (tracks outdoor seasons)
-b_c376fc4d6ccb,r_39dfb37da081,3.0,22828.0,15.418498992919922,7.0,22.91666603088379,4.769450664520264,7.3276472091674805,2.849679946899414,,2.858325958251953,"scheduled ventilation (AHU, weekday operating hours)"
-b_955300d5f090,r_5b4caff693e8,3.0,38336.0,15.44373607635498,6.0,23.0,5.293636322021484,1.9263849258422852,1.6529464721679688,12.07291030883789,0.07053470611572266,unconditioned/vent-only space (tracks outdoor seasons)
-b_9a5a45d6e498,r_22232f8e4b91,3.0,68457.0,15.45107650756836,3.8534693717956543,24.0,6.309248447418213,4.553862571716309,2.1406631469726562,10.45508861541748,1.1119260787963867,"scheduled ventilation (AHU, weekday operating hours)"
-b_62e09ab003f8,r_5bd522ea865e,6.0,143893.0,15.460672378540039,5.559321880340576,21.58333396911621,4.837680339813232,2.095858573913574,1.8306760787963867,7.4474077224731445,-0.011209487915039062,unconditioned/vent-only space (tracks outdoor seasons)
-b_b9acdf6071f3,r_e45cc68e00a2,5.0,91811.0,15.472857475280762,3.490897059440613,22.684846878051758,5.966550827026367,2.507327079772949,2.866765022277832,8.20910358428955,-0.07052135467529297,unconditioned/vent-only space (tracks outdoor seasons)
-b_955300d5f090,r_9b5477e4058b,3.0,48564.0,15.747925758361816,3.501027393341065,24.847457885742188,6.226083278656006,3.9120473861694336,2.0980405807495117,11.824091911315918,-0.11090660095214844,"scheduled ventilation (AHU, weekday operating hours)"
-b_955300d5f090,r_27b0cfa20d53,5.0,99104.0,16.236663818359375,3.0,26.0,6.749387741088867,5.302665710449219,4.985854148864746,10.204568862915039,0.505366325378418,unconditioned/vent-only space (tracks outdoor seasons)
-b_955300d5f090,r_e2a69ae237a4,5.0,51742.0,16.24174690246582,4.566666603088379,26.299999237060547,6.256832122802734,4.4181718826293945,5.030272483825684,9.971707344055176,0.3629283905029297,unconditioned/vent-only space (tracks outdoor seasons)
-b_955300d5f090,r_02e6b266abf6,5.0,87995.0,16.66031837463379,1.853181838989257,27.220338821411133,7.780560493469238,4.981172561645508,3.9396772384643555,11.668581008911133,-0.06834030151367188,"scheduled ventilation (AHU, weekday operating hours)"
-b_c2dc4330feb5,r_65e639cdbefd,3.0,49390.0,16.803600311279297,2.0,27.0,7.540674686431885,3.572509765625,1.7866649627685547,11.711344718933105,0.578643798828125,"scheduled ventilation (AHU, weekday operating hours)"
-b_be0092b8c92e,r_279cb4f4fc29,3.0,64879.0,17.39132308959961,3.0,25.399999618530273,7.598820686340332,3.3556251525878906,3.26279354095459,8.884858131408691,-0.045940399169921875,unconditioned/vent-only space (tracks outdoor seasons)
-b_955300d5f090,r_ca9045f4fef2,4.0,67290.0,17.47679901123047,4.71051471233368,26.433332443237305,6.832335472106934,9.236570358276367,8.881999015808105,9.245710372924805,0.14682960510253906,unconditioned/vent-only space (tracks outdoor seasons)
-b_d45195c31e8f,r_37ca07e35f31,2.0,28716.0,17.49253273010254,5.0,28.799999237060547,7.440978527069092,9.542098045349121,1.7154769897460938,12.956385612487793,4.319018363952637,"scheduled ventilation (AHU, weekday operating hours)"
-b_9a5a45d6e498,r_101aa0c8e52c,3.0,61210.0,17.96147346496582,5.796743941307068,27.0,6.666900157928467,5.028185844421387,1.0335731506347656,10.169468879699707,-0.21351242065429688,"scheduled ventilation (AHU, weekday operating hours)"
-b_955300d5f090,r_2823849bded1,8.0,127383.0,17.97539520263672,7.0,25.950000762939453,5.787521839141846,3.756753921508789,2.731151580810547,8.840665817260742,0.15660667419433594,"scheduled ventilation (AHU, weekday operating hours)"
-b_be0092b8c92e,r_f1736c7a40b7,3.0,64707.0,18.026037216186523,4.300000190734863,30.0,7.596157073974609,2.6192264556884766,2.4933204650878906,10.441181182861328,0.13754844665527344,unconditioned/vent-only space (tracks outdoor seasons)
-b_a0e6376410cb,r_0524a19c4bfd,4.0,93377.0,18.350006103515625,5.209935474395752,26.677151107788085,7.000560283660889,4.46868896484375,3.9370861053466797,9.383346557617188,0.000843048095703125,unconditioned/vent-only space (tracks outdoor seasons)
-b_955300d5f090,r_786737d33818,4.0,65678.0,18.406618118286133,6.132564020156862,26.649999618530273,6.368472099304199,8.79362678527832,8.45177173614502,7.106228828430176,-0.08893013000488281,unconditioned/vent-only space (tracks outdoor seasons)
-b_2b69c4007582,r_e1bac19501a4,14.0,184820.0,18.4110050201416,2.75,33.0,8.467782974243164,2.111501693725586,1.9783782958984375,9.746050834655762,-0.15406227111816406,unconditioned/vent-only space (tracks outdoor seasons)
-b_955300d5f090,r_f7088836b3c8,5.0,98662.0,18.486133575439453,3.0833332538604736,29.482695579528805,8.098608016967773,8.001522064208984,7.632041931152344,10.0173978805542,0.2469921112060547,unconditioned/vent-only space (tracks outdoor seasons)
-b_a0e6376410cb,r_f32d30daa895,3.0,25204.0,18.642969131469727,6.0,31.421379756927475,7.864974021911621,4.040626525878906,3.1517505645751953,4.268009185791016,0.6578044891357422,ordinary conditioned space (radiators/mixed HVAC)
-b_afdd8b1a066d,r_3a26ffb366db,3.0,45468.0,18.783185958862305,11.366666793823242,25.0,4.216314792633057,1.7028255462646484,0.8190441131591797,7.732317924499512,-0.15099525451660156,unconditioned/vent-only space (tracks outdoor seasons)
-b_2b69c4007582,r_268fb9378735,6.0,73558.0,18.915037155151367,10.350648880004883,25.598247814178453,4.69200325012207,3.022510528564453,1.7282371520996094,8.123129844665527,0.31634521484375,"scheduled ventilation (AHU, weekday operating hours)"
-b_4217a3b6e58b,r_84ab2c9a82b8,1.0,21896.0,18.937320709228516,12.0,25.0,3.8751296997070312,2.034788131713867,1.2020111083984375,7.84197998046875,-0.2617454528808594,unconditioned/vent-only space (tracks outdoor seasons)
-b_955300d5f090,r_f2d7d812a5f0,2.0,29989.0,19.101951599121094,5.0,28.28333282470703,7.324823379516602,8.971978187561035,8.84556770324707,8.929415702819824,0.13271141052246094,unconditioned/vent-only space (tracks outdoor seasons)
-b_a0e6376410cb,r_66f8934ecd8c,5.0,105456.0,19.104171752929688,8.216666221618652,28.565196990966797,6.08662748336792,5.998128890991211,4.344608306884766,4.040771484375,1.3002948760986328,"scheduled ventilation (AHU, weekday operating hours)"
-b_2b69c4007582,r_a86a82a16ca0,17.0,225362.0,19.12948989868164,8.5,28.070114707946757,5.782099723815918,5.040187835693359,3.356283187866211,8.349100112915039,-0.4173088073730469,"scheduled ventilation (AHU, weekday operating hours)"
-b_2b69c4007582,r_7c97db8498cc,28.0,363341.0,19.539684295654297,9.050000190734863,27.0,5.565213203430176,4.017871856689453,1.643117904663086,8.002903938293457,0.1783599853515625,"scheduled ventilation (AHU, weekday operating hours)"
-b_a0e6376410cb,r_77a0d3ba0f9c,8.0,196823.0,19.580547332763672,5.616666793823242,29.483333587646484,7.332419395446777,1.5293254852294922,1.1736431121826172,8.5603666305542,0.14144134521484375,unconditioned/vent-only space (tracks outdoor seasons)
-b_4217a3b6e58b,r_a36e0de3c5c2,8.0,133786.0,19.65196990966797,12.0,25.81355857849121,4.152437210083008,2.7676048278808594,1.905374526977539,5.410179138183594,0.7837810516357422,unconditioned/vent-only space (tracks outdoor seasons)
-b_afdd8b1a066d,r_ea358c0b18f2,3.0,45289.0,19.730335235595703,15.0,25.0,3.148531198501587,1.801382064819336,0.7093486785888672,2.037677764892578,-0.06750869750976562,ordinary conditioned space (radiators/mixed HVAC)
-b_955300d5f090,r_0d59c2f44f40,33.0,650170.0,19.75297737121582,8.83644232749939,27.327196025848377,5.528252124786377,3.0226173400878906,2.4163055419921875,8.04210376739502,-0.5087146759033203,unconditioned/vent-only space (tracks outdoor seasons)
-b_955300d5f090,r_e156e26e7583,3.0,32706.0,19.970073699951172,11.0,31.0,6.1677045822143555,2.4080944061279297,2.311372756958008,7.634027481079102,-0.025304794311523438,unconditioned/vent-only space (tracks outdoor seasons)
-b_955300d5f090,r_9ea33a360e32,19.0,340159.0,20.102516174316406,9.0,29.53333282470703,6.197927474975586,5.007749557495117,2.8827743530273438,10.937528610229492,0.7304630279541016,"scheduled ventilation (AHU, weekday operating hours)"
-b_60c6d1796ab4,r_e301f99df217,1.0,21566.0,20.13431167602539,17.0,24.450000762939453,2.7492358684539795,1.2161598205566406,1.1746940612792969,5.6898956298828125,0.015888214111328125,unconditioned/vent-only space (tracks outdoor seasons)
-b_7aac45259c8a,r_ab8e8ce3ee18,4.0,63416.0,20.248868942260742,13.0,32.0,5.881930828094482,0.6659278869628906,0.5371837615966797,4.83381462097168,0.06625556945800781,ordinary conditioned space (radiators/mixed HVAC)
-b_a0e6376410cb,r_e60f0e5e874f,3.0,63950.0,20.260671615600586,9.0,31.621057605743406,6.817746639251709,6.955070495605469,4.526788711547852,6.666095733642578,0.946929931640625,"scheduled ventilation (AHU, weekday operating hours)"
-b_be0092b8c92e,r_829b4d35c026,3.0,64334.0,20.287050247192383,6.102907752990722,30.0,7.18626070022583,3.4239864349365234,3.4649810791015625,6.811149597167969,-2.5053977966308594,unconditioned/vent-only space (tracks outdoor seasons)
-b_ce7925d56654,r_6beebf5e37bf,4.0,71741.0,20.301761627197266,10.25,34.846153259277344,6.681120872497559,4.424009323120117,3.2739734649658203,10.035661697387695,0.3622322082519531,"scheduled ventilation (AHU, weekday operating hours)"
-b_53df020ad893,r_9b5c4b767d88,3.0,74412.0,20.308366775512695,6.066666603088379,33.29999923706055,8.158671379089355,2.092344284057617,1.8900814056396484,7.10706901550293,0.06268501281738281,unconditioned/vent-only space (tracks outdoor seasons)
-b_ad4a5872f2db,r_38ffec571713,13.0,320376.0,20.37318992614746,10.0,25.85454559326172,4.5634636878967285,4.120357513427734,1.1937217712402344,3.7553653717041016,-1.5807743072509766,"scheduled ventilation (AHU, weekday operating hours)"
-b_afdd8b1a066d,r_7d9d533d77cb,15.0,359726.0,20.465145111083984,10.0,28.0,5.335058212280273,4.197776794433594,2.84246826171875,7.235029220581055,-1.1072845458984375,"scheduled ventilation (AHU, weekday operating hours)"
-b_955300d5f090,r_e292e2e62ec3,5.0,78903.0,20.466873168945312,9.0,31.0,6.6164679527282715,1.2049236297607422,1.1698131561279297,10.119791984558105,0.040897369384765625,unconditioned/vent-only space (tracks outdoor seasons)
-b_4217a3b6e58b,r_bd4c57fb510d,11.0,179314.0,20.551555633544922,13.466666221618652,27.28024635314942,4.411491394042969,4.941860198974609,3.037557601928711,3.551229476928711,0.8855266571044922,"scheduled ventilation (AHU, weekday operating hours)"
-b_4217a3b6e58b,r_cdb37c5edd19,3.0,22761.0,20.570907592773438,14.0,26.0,3.541339159011841,5.537435531616211,4.149482727050781,3.7983245849609375,-0.23137283325195312,"scheduled ventilation (AHU, weekday operating hours)"
-b_55038c6689a0,r_ddd012cdb553,10.0,159132.0,20.908761978149414,12.095538806915284,32.549329948425274,5.817671775817871,5.147184371948242,1.5370197296142578,4.2563934326171875,1.8159656524658203,"scheduled ventilation (AHU, weekday operating hours)"
-b_53df020ad893,r_4d95939c1d71,3.0,95172.0,21.086084365844727,13.600000381469727,26.233333587646484,3.9949839115142822,0.589630126953125,0.5252399444580078,5.315019607543945,-0.09200859069824219,unconditioned/vent-only space (tracks outdoor seasons)
-b_ce7925d56654,r_0620a16c3ded,2.0,36338.0,21.153152465820312,12.0,29.15466070175173,5.450156211853027,4.241254806518555,2.383960723876953,8.429811477661133,0.011751174926757812,"scheduled ventilation (AHU, weekday operating hours)"
-b_afdd8b1a066d,r_bcd426cb4dc2,5.0,137826.0,21.322505950927734,14.216666221618652,29.0,5.043494701385498,3.3268604278564453,1.839324951171875,4.2544097900390625,0.5657920837402344,"scheduled ventilation (AHU, weekday operating hours)"
-b_afdd8b1a066d,r_f04bff6ae6b7,5.0,101885.0,21.345273971557617,11.0,29.25,6.11444091796875,2.476766586303711,2.366060256958008,7.765541076660156,0.042285919189453125,unconditioned/vent-only space (tracks outdoor seasons)
-b_60c6d1796ab4,r_b4c6b2670bd8,4.0,89435.0,21.937124252319336,10.033728981018067,32.08333206176758,6.96323299407959,1.4300365447998047,1.5240421295166016,6.669073104858398,0.012844085693359375,unconditioned/vent-only space (tracks outdoor seasons)
-b_be0092b8c92e,r_2781546d7942,3.0,64923.0,22.41581153869629,12.0,31.0,5.758373260498047,2.8338470458984375,3.010883331298828,9.638887405395508,-0.09977531433105469,unconditioned/vent-only space (tracks outdoor seasons)
-b_afdd8b1a066d,r_68e2b99fbf02,1.0,30994.0,22.424646377563477,18.0,27.0,2.9214816093444824,0.2704010009765625,0.22499465942382812,5.0167083740234375,0.018611907958984375,unconditioned/vent-only space (tracks outdoor seasons)
-b_62e09ab003f8,r_5c8cfab78cad,27.0,514552.0,22.47916030883789,14.0,29.0,4.8827290534973145,1.7740039825439453,0.8263950347900391,0.9953231811523438,-0.35761260986328125,ordinary conditioned space (radiators/mixed HVAC)
-b_be0092b8c92e,r_38045b3776de,3.0,64925.0,22.707462310791016,14.0,30.89439735412596,5.076829433441162,1.5666751861572266,1.5210018157958984,5.799139022827148,-0.14640045166015625,unconditioned/vent-only space (tracks outdoor seasons)
-b_be0092b8c92e,r_a6ff229f5d88,3.0,64920.0,22.786420822143555,12.947340631484986,30.726507854461662,5.220310688018799,2.9747142791748047,2.9027957916259766,7.148979187011719,0.08974075317382812,unconditioned/vent-only space (tracks outdoor seasons)
-b_afdd8b1a066d,r_f77e432fe269,4.0,57947.0,22.794153213500977,14.0,28.116666793823242,4.418858528137207,4.115118026733398,3.5757102966308594,3.1022262573242188,0.1353893280029297,ordinary conditioned space (radiators/mixed HVAC)
-b_afdd8b1a066d,r_51e3a4545cad,2.0,45675.0,22.952125549316406,16.0,29.0,4.163959503173828,4.368167877197266,2.3036632537841797,1.57427978515625,-0.6721153259277344,"scheduled ventilation (AHU, weekday operating hours)"
-b_955300d5f090,r_9cb7e6657012,8.0,161090.0,22.975360870361328,18.0,32.0,4.734466075897217,0.5722236633300781,0.5345687866210938,8.524869918823242,-0.061370849609375,unconditioned/vent-only space (tracks outdoor seasons)
-b_ce7925d56654,r_1bb7730fcee1,10.0,182392.0,23.209121704101562,15.0,31.236779785156234,5.217009544372559,3.532787322998047,1.877218246459961,5.894094467163086,0.5879249572753906,"scheduled ventilation (AHU, weekday operating hours)"
-b_a0e6376410cb,r_770fc166ffa0,6.0,128007.0,23.35256004333496,16.0,29.0,4.42451286315918,3.544961929321289,0.5764102935791016,3.0026283264160156,-1.1315994262695312,"scheduled ventilation (AHU, weekday operating hours)"
-b_dabc992b5f46,r_f13f5c3c9f5f,4.0,78003.0,24.692785263061523,11.0,36.168824005126936,7.712890148162842,7.881195068359375,0.6740684509277344,2.4397735595703125,-2.2702484130859375,"scheduled ventilation (AHU, weekday operating hours)"
-b_ad4a5872f2db,r_99bd12b98a34,4.0,122602.0,27.08595085144043,21.0,34.233333587646484,5.035804271697998,0.9842624664306641,0.8599147796630859,-1.020639419555664,-0.0600128173828125,ordinary conditioned space (radiators/mixed HVAC)
-b_afdd8b1a066d,r_d1e39dbbd605,6.0,178833.0,27.205974578857422,23.0,31.0,2.564842700958252,0.6400318145751953,0.52691650390625,0.3753814697265625,-0.07421302795410156,ordinary conditioned space (radiators/mixed HVAC)
-b_b7e9d5634ad3,r_484904c730d7,4.0,92169.0,27.652881622314453,4.199999809265137,47.41666793823242,13.675037384033203,4.351097106933594,4.389474868774414,2.913858413696289,0.27538108825683594,ordinary conditioned space (radiators/mixed HVAC)
-b_afdd8b1a066d,r_6dbe4292d4d7,3.0,78474.0,28.119409561157227,23.0,33.39189147949219,3.1326682567596436,0.5608844757080078,0.20102691650390625,0.4177055358886719,0.2509880065917969,ordinary conditioned space (radiators/mixed HVAC)
-b_955300d5f090,r_cb32bb120909,2.0,35181.0,29.49652862548828,20.0,35.0,4.804248332977295,0.9948463439941406,0.7486839294433594,-8.456584930419922,-0.29415321350097656,winter-warm plant (heating equipment nearby)
-b_afdd8b1a066d,r_b90eecea9c7e,1.0,10249.0,32.8201904296875,23.0,42.0,6.613819122314453,14.458879470825195,6.780588150024414,3.6633472442626953,4.620477676391602,heating plant (boiler room / district-heat substation)
-b_6869b0f15a78,r_0facb54e227f,2.0,28500.0,33.71730422973633,24.0,48.0,7.27948522567749,1.7568893432617188,1.9118881225585938,-12.068391799926758,-0.25122833251953125,heating plant (boiler room / district-heat substation)
-b_6869b0f15a78,r_974c30d91109,6.0,85495.0,35.533721923828125,27.0,43.0,4.956151485443115,0.8019790649414062,0.5732841491699219,9.882659912109375,-0.058349609375,heating plant (boiler room / district-heat substation)

research/hvac_room_profiles.py

@@ -1,88 +0,0 @@
-"""Per-room temperature signatures -> heuristic HVAC-type guess.
-
-Each device reports (end_time, voltage, temperature). Devices sit on/near
-technical equipment; the temperature signature is the only HVAC clue we have:
-  - level (mean, p05, p95): freezer / chilled / ambient / hot plant
-  - diurnal amplitude on weekdays vs weekend: scheduled ventilation (AHU)
-  - seasonal swing (winter vs summer monthly means): heating plant peaks in
-    winter, cooling load peaks in summer, tightly conditioned rooms are flat
-Run: venv/bin/python research/hvac_room_profiles.py
-"""
-import pandas as pd
-import numpy as np
-
-devices = pd.read_csv("dataset/train/devices.csv")
-m = pd.read_parquet(
-    "dataset/train/battery_metrics.parquet",
-    columns=["end_time", "temperature", "device_id"],
-)
-m = m.dropna(subset=["temperature"])
-m["end_time"] = pd.to_datetime(m["end_time"])
-m["device_id"] = m["device_id"].astype(str)
-m = m.merge(devices[["device_id", "building_id", "room_id"]], on="device_id")
-
-m["hour"] = m["end_time"].dt.hour
-m["month"] = m["end_time"].dt.month
-m["weekend"] = m["end_time"].dt.dayofweek >= 5
-
-
-def profile(g: pd.DataFrame) -> pd.Series:
-    t = g["temperature"]
-    wd = g[~g["weekend"]]
-    we = g[g["weekend"]]
-    # diurnal amplitude: spread of hour-of-day means (weekdays)
-    hr = wd.groupby("hour")["temperature"].mean()
-    diurnal = hr.max() - hr.min() if len(hr) > 12 else np.nan
-    hr_we = we.groupby("hour")["temperature"].mean()
-    diurnal_we = hr_we.max() - hr_we.min() if len(hr_we) > 12 else np.nan
-    mo = g.groupby("month")["temperature"].mean()
-    winter = mo.reindex([12, 1, 2]).mean()
-    summer = mo.reindex([6, 7, 8]).mean()
-    return pd.Series(
-        dict(
-            n_devices=g["device_id"].nunique(),
-            n_obs=len(g),
-            t_mean=t.mean(),
-            t_p05=t.quantile(0.05),
-            t_p95=t.quantile(0.95),
-            t_std=t.std(),
-            diurnal_wd=diurnal,
-            diurnal_we=diurnal_we,
-            seasonal=summer - winter,  # + = warmer in summer
-            weekend_dip=(wd["temperature"].mean() - we["temperature"].mean()),
-        )
-    )
-
-
-prof = (
-    m.groupby(["building_id", "room_id"])
-    .apply(profile, include_groups=False)
-    .reset_index()
-)
-
-
-def guess(r: pd.Series) -> str:
-    if r.t_mean < 8:
-        return "refrigeration / freezer or cold room"
-    if r.t_mean > 32:
-        return "heating plant (boiler room / district-heat substation)"
-    sched = (r.diurnal_wd or 0) - (r.diurnal_we or 0)
-    if r.diurnal_wd and r.diurnal_wd > 3 and sched > 1:
-        return "scheduled ventilation (AHU, weekday operating hours)"
-    if r.t_std < 1.5 and abs(r.seasonal) < 1.5:
-        return "tightly conditioned (server/tech room, always-on cooling)"
-    if r.seasonal > 5:
-        return "unconditioned/vent-only space (tracks outdoor seasons)"
-    if r.seasonal < -2:
-        return "winter-warm plant (heating equipment nearby)"
-    return "ordinary conditioned space (radiators/mixed HVAC)"
-
-
-prof["hvac_guess"] = prof.apply(guess, axis=1)
-prof = prof.sort_values("t_mean")
-pd.set_option("display.width", 250)
-pd.set_option("display.max_rows", 100)
-print(prof.round(1).to_string(index=False))
-prof.to_csv("research/hvac_room_profiles.csv", index=False)
-print("\nGuess distribution:")
-print(prof.hvac_guess.value_counts().to_string())

research/lab.py

@@ -1,81 +0,0 @@
-"""
-MUTABLE candidate file - the ONLY file experiments may modify.
-
-Contract (enforced by research/run_experiment.py):
-  candidate() -> object with
-    .fit(train_df)   - train_df: (device, origin) rows incl. FEATURES + labels
-                       (dies42, days_to_eol, eol) - labels are legitimate here
-                       (closed windows only, walk-forward supplies them)
-    .predict(rows)   - rows: same columns MINUS all label columns (redacted by
-                       the harness). Must return a DataFrame indexed like rows
-                       with columns: p42, day_q0.05, day_q0.2, day_q0.5,
-                       day_q0.8. The planner runs on p42 MASS (sum sets the
-                       inclusion count) - calibration matters as much as
-                       ranking.
-  CANDIDATE_NAME - short slug for the log.
-  NOTES          - one-line description of the change being tested.
-
-Read-only imports from batteryswap_example are allowed (SklearnRUL,
-DayQuantiles, FEATURES, WINDOW_DAYS...). Do not monkeypatch them.
-"""
-
-from pathlib import Path
-
-import numpy as np
-import pandas as pd
-from sklearn.ensemble import BaggingClassifier
-from sklearn.impute import SimpleImputer
-from sklearn.linear_model import LogisticRegression
-from sklearn.pipeline import make_pipeline
-from sklearn.preprocessing import StandardScaler
-
-from batteryswap_example.models_cv import FEATURES, DayQuantiles
-
-CANDIDATE_NAME = 'hvac_env_logbag'
-NOTES = ('exact logbag replica (15-bag logistic C=1.0, DayQuantiles timing on '
-         'FEATURES only) + 3 static environment features from temperature '
-         'signatures (research/hvac_origin_features.parquet, EXPANDING '
-         'life-so-far at each origin - leak-free): cold_frac (days with '
-         'min<5C), hot_frac (days with max>32C), sched_amp (weekday-weekend '
-         'daily-range gap, AHU schedule). Hypothesis: HVAC-class death rate is '
-         'U-shaped (cold 0.25-0.40 / hot 0.2-0.3 / mild 0.1, 2026-08-19 EDA) - '
-         'two exposure arms express what the rejected monotone temp features '
-         '(phases 6/8/13) could not')
-
-HVAC_FEATURES = ['cold_frac', 'hot_frac', 'sched_amp']
-_HVAC = (pd.read_parquet(Path('research') / 'hvac_origin_features.parquet')
-         .set_index(['device_id', 'origin_idx'])[HVAC_FEATURES])
-
-
-def env_cols(df):
-    dev = df['device_id'] if 'device_id' in df.columns else df.index
-    key = pd.MultiIndex.from_arrays(
-        [np.asarray(dev).astype(str), df['origin_idx'].values])
-    env = _HVAC.reindex(key)
-    env.index = df.index
-    return env
-
-
-class HvacEnvLogbag:
-    """logbag whose WHO-classifier also sees the environment columns."""
-
-    def fit(self, train):
-        X = pd.concat([train[FEATURES], env_cols(train)], axis=1)
-        base = make_pipeline(SimpleImputer(strategy='median'),
-                             StandardScaler(),
-                             LogisticRegression(max_iter=2000, C=1.0))
-        self.clf = BaggingClassifier(base, n_estimators=15, random_state=0)
-        self.clf.fit(X, train['dies42'])
-        dying = train[train['dies42'] == 1]
-        self.days = DayQuantiles().fit(dying[FEATURES], dying['days_to_eol'])
-        return self
-
-    def predict(self, rows):
-        out = self.days.predict(rows[FEATURES])
-        X = pd.concat([rows[FEATURES], env_cols(rows)], axis=1)
-        out['p42'] = self.clf.predict_proba(X)[:, 1]
-        return out
-
-
-def candidate():
-    return HvacEnvLogbag()

research/last_sat_s1.csv

@@ -1,35 +0,0 @@
-room_change,battery_swap,early_swap,daily_limit,overtime,weekly_limit,late_swap,travel,building_change,total_cost,origin,model,planner,p_star,day_q,ec_margin,ec_pmax,split_days,restarts,iters
-7.5,4.0,247.0,100.0,92.806,200.0,1370.0,32.93066666666667,12.0,2066.2366666666667,8,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-7.0,4.5,961.0,100.0,56.8,100.0,390.0,22.20183333333334,11.0,1652.5018333333333,9,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.5,4.0,670.5,100.0,75.47366666666667,100.0,40.0,24.641,9.0,1029.1146666666666,10,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.0,3.75,483.5,100.0,83.41166666666665,100.0,440.0,29.157166666666665,10.0,1255.8188333333333,11,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-8.0,4.5,637.0,100.0,67.59666666666666,200.0,1590.0,27.124999999999996,12.0,2646.221666666667,12,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-7.5,5.0,757.0,300.0,106.648,200.0,1670.0,32.159333333333336,14.0,3092.307333333333,13,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.0,4.25,621.5,200.0,86.50599999999999,100.0,2010.0,28.371666666666666,12.0,3068.6276666666668,14,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-8.0,4.25,408.0,200.0,62.32166666666667,200.0,3170.0,23.285,15.0,4090.856666666667,15,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-7.0,4.25,285.5,300.0,93.36500000000001,300.0,2550.0,36.07033333333334,10.0,3586.185333333334,16,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.5,4.5,290.5,300.0,98.618,300.0,1910.0,38.10116666666667,11.0,2959.2191666666668,17,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.5,4.0,520.5,100.0,58.25,200.0,1820.0,22.438000000000006,10.0,2741.688,18,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.5,3.25,478.5,100.0,53.99099999999999,200.0,1140.0,22.006833333333336,9.0,2012.2478333333333,19,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.0,2.5,228.5,100.0,55.512,200.0,1050.0,22.90316666666667,9.0,1673.4151666666667,20,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.5,3.25,353.5,800.0,345.92866666666663,300.0,1380.0,118.16833333333332,11.0,3318.347,21,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.0,3.5,531.5,400.0,108.80499999999999,200.0,710.0,36.55716666666667,10.0,2006.3621666666666,22,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.0,3.75,511.5,600.0,347.416,300.0,860.0,124.92466666666667,10.0,2763.5906666666665,23,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.5,3.5,501.5,500.0,114.99533333333332,300.0,750.0,37.41066666666667,9.0,2221.906,24,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.0,4.0,583.5,500.0,120.476,300.0,1720.0,38.240833333333335,9.0,3281.2168333333334,25,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.5,3.5,436.5,200.0,83.13933333333333,200.0,2020.0,29.703666666666667,8.0,2986.343,26,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.0,3.25,448.5,100.0,18.547666666666668,100.0,1400.0,6.45,7.0,2088.7476666666666,27,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-4.5,3.0,298.5,100.0,53.15233333333333,100.0,810.0,24.799,7.0,1400.9513333333332,28,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-4.0,2.25,165.0,100.0,52.12833333333333,100.0,440.0,24.62433333333333,6.0,894.0026666666668,29,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.0,2.5,309.5,200.0,85.22933333333333,200.0,310.0,25.614666666666665,6.0,1143.844,30,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.5,3.0,258.0,200.0,129.325,200.0,500.0,44.781666666666666,9.0,1349.6066666666666,31,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-7.0,3.5,265.5,500.0,145.59633333333335,300.0,620.0,55.230999999999995,11.0,1907.8273333333334,32,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-4.5,2.5,114.0,300.0,142.31,300.0,280.0,61.32233333333334,9.0,1213.6323333333335,33,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.0,3.25,208.5,500.0,192.601,300.0,800.0,71.39816666666667,10.0,2090.7491666666665,34,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-5.5,3.0,124.5,500.0,186.30666666666662,300.0,840.0,72.26283333333333,10.0,2041.5695,35,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.5,3.5,258.0,500.0,183.71433333333331,300.0,1200.0,71.00866666666667,9.0,2531.723,36,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-6.0,3.25,345.0,300.0,127.45500000000001,300.0,780.0,50.5915,10.0,1922.2965,37,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-4.5,2.5,211.0,300.0,136.04466666666667,200.0,790.0,53.50816666666667,7.0,1704.5528333333332,38,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-9.0,4.75,525.0,100.0,78.35866666666666,100.0,380.0,31.798166666666667,11.0,1239.9068333333332,39,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-9.5,5.25,550.0,100.0,87.05600000000001,200.0,200.0,35.483000000000004,12.0,1199.289,40,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60
-8.0,6.0,646.0,200.0,106.38633333333333,100.0,300.0,37.063,10.0,1413.4493333333335,41,logbag,tsroute,0.75,0.05,0.8,25.0,12,2,60

research/last_tier1.csv

@@ -1,35 +0,0 @@
-weekly_limit,overtime,building_change,travel,room_change,early_swap,battery_swap,daily_limit,late_swap,total_cost,origin,planner,p_star,day_q,ec_margin,ec_pmax,split_days,model,restarts,iters
-100.0,24.84533333333333,12.0,32.25083333333333,7.5,237.5,4.0,0.0,1060.0,1478.0961666666667,8,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,23.703000000000003,11.0,22.964000000000002,7.0,978.0,4.5,0.0,300.0,1347.1670000000001,9,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,17.628999999999998,11.0,31.258833333333335,6.5,796.5,4.5,0.0,30.0,897.3878333333333,10,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,28.291999999999998,12.0,31.271,6.5,557.5,4.0,0.0,390.0,1029.563,11,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,27.13466666666667,14.0,28.298833333333327,8.5,676.5,4.75,0.0,1590.0,2349.1835,12,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,24.216666666666665,15.0,36.37866666666667,7.5,882.5,5.25,0.0,1680.0,2750.845333333333,13,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,25.173999999999996,12.0,28.472999999999995,6.0,608.0,4.25,0.0,2010.0,2793.897,14,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,24.104000000000003,15.0,22.700833333333332,8.0,404.5,4.25,0.0,3170.0,3748.5548333333336,15,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,58.925333333333334,11.0,36.326,7.5,276.0,4.25,100.0,2570.0,3164.001333333333,16,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,28.730333333333327,10.0,22.291666666666668,7.0,298.5,4.5,0.0,1640.0,2011.0220000000002,17,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,26.260666666666665,12.0,27.17516666666667,7.5,529.5,4.25,0.0,1850.0,2556.685833333333,18,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,26.426666666666666,11.0,24.01366666666667,6.5,558.5,3.5,0.0,1140.0,1769.9403333333335,19,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,27.958000000000002,9.0,25.213,5.0,228.5,2.5,0.0,1020.0,1318.1709999999998,20,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,250.0213333333333,11.0,118.16833333333332,6.5,335.5,3.25,500.0,1370.0,2694.4396666666667,21,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,39.92866666666667,11.0,36.05100000000001,6.0,525.0,3.5,0.0,720.0,1341.4796666666666,22,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-200.0,231.81866666666667,9.0,104.126,6.0,532.0,3.75,400.0,550.0,2036.6946666666668,23,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,47.44799999999999,10.0,39.35933333333334,6.0,518.5,3.75,0.0,750.0,1375.0573333333332,24,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,42.67266666666667,10.0,43.27016666666667,6.5,582.5,4.0,0.0,1680.0,2368.9428333333335,25,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,36.77466666666666,9.0,29.706000000000003,6.0,431.0,3.5,0.0,2020.0,2535.980666666667,26,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,0.0,9.0,6.9111666666666665,5.5,493.5,3.5,0.0,1400.0,1918.4111666666668,27,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,28.88133333333333,8.0,24.702666666666666,4.5,274.0,3.0,0.0,790.0,1233.0839999999998,28,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,29.748999999999995,8.0,24.656166666666664,4.5,210.0,2.5,0.0,440.0,719.4051666666667,29,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,40.089999999999996,6.0,26.473833333333335,5.0,318.0,2.5,0.0,310.0,708.0638333333334,30,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,69.965,9.0,46.48566666666666,5.5,257.5,3.0,0.0,110.0,601.4506666666666,31,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,79.22399999999999,12.0,57.020833333333336,7.5,305.5,3.75,0.0,620.0,1084.9948333333334,32,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-200.0,125.87233333333333,10.0,69.63366666666667,5.0,168.5,2.75,200.0,270.0,1051.7559999999999,33,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-200.0,125.11499999999998,11.0,71.59066666666666,5.5,201.5,3.25,200.0,800.0,1617.9556666666667,34,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,121.88699999999999,12.0,72.7285,6.0,143.5,3.25,100.0,840.0,1399.3655,35,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,118.35299999999998,9.0,70.9795,6.5,257.0,3.5,100.0,1220.0,1885.3325,36,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,100.79233333333335,10.0,58.18600000000001,5.5,339.5,3.25,100.0,730.0,1447.2283333333332,37,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,101.69466666666666,7.0,59.76166666666666,4.5,220.0,2.5,100.0,790.0,1385.4563333333335,38,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-100.0,60.986333333333334,11.0,40.96116666666667,8.0,461.5,4.5,0.0,130.0,816.9475000000001,39,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,63.32833333333333,10.0,41.691,9.0,545.0,5.25,0.0,230.0,904.2693333333334,40,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20
-0.0,62.25433333333333,12.0,40.179500000000004,8.5,614.0,6.0,100.0,270.0,1112.9338333333333,41,tsroute,0.8,0.05,0.8,25.0,7,candidate,0,20

research/last_tier1_baseline.csv

@@ -1,35 +0,0 @@
-building_change,weekly_limit,early_swap,room_change,late_swap,overtime,daily_limit,travel,battery_swap,total_cost,origin,planner,p_star,day_q,ec_margin,ec_pmax,split_days,model,restarts,iters
-12.0,100.0,237.5,7.5,1060.0,24.84533333333333,0.0,32.25083333333333,4.0,1478.0961666666667,8,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,0.0,978.0,7.0,300.0,23.703000000000003,0.0,22.964000000000002,4.5,1347.1670000000001,9,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,0.0,878.0,7.0,20.0,18.956999999999994,0.0,32.12383333333334,4.5,972.5808333333334,10,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,0.0,557.5,6.5,390.0,28.291999999999998,0.0,31.271,4.0,1029.563,11,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-14.0,0.0,756.0,9.0,1590.0,27.19366666666667,0.0,28.146333333333327,4.75,2429.09,12,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-15.0,100.0,832.5,8.0,1690.0,48.894333333333336,0.0,48.63400000000001,5.25,2748.278333333333,13,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,100.0,608.0,6.0,2010.0,25.173999999999996,0.0,28.472999999999995,4.25,2793.897,14,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-15.0,100.0,404.5,8.0,3170.0,24.104000000000003,0.0,22.700833333333332,4.25,3748.5548333333336,15,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,100.0,276.0,7.5,2570.0,58.925333333333334,100.0,36.326,4.25,3164.001333333333,16,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,298.5,7.0,1640.0,28.730333333333327,0.0,22.291666666666668,4.5,2011.0220000000002,17,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,100.0,529.5,7.5,1850.0,26.260666666666665,0.0,27.17516666666667,4.25,2556.685833333333,18,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,588.0,6.0,1130.0,26.426666666666666,0.0,23.63266666666667,3.5,1787.5593333333331,19,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,0.0,228.5,5.0,1020.0,27.958000000000002,0.0,25.213,2.5,1318.1709999999998,20,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,100.0,335.5,6.5,1370.0,250.0213333333333,500.0,118.16833333333332,3.25,2694.4396666666667,21,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,0.0,532.5,6.0,740.0,40.284333333333336,0.0,37.73900000000001,3.5,1371.0233333333333,22,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,200.0,532.0,6.0,550.0,231.81866666666667,400.0,104.126,3.75,2036.6946666666668,23,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,518.5,6.0,750.0,47.44799999999999,0.0,39.35933333333334,3.75,1375.0573333333332,24,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,582.5,6.5,1680.0,42.67266666666667,0.0,43.27016666666667,4.0,2368.9428333333335,25,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,0.0,435.5,6.0,2020.0,35.633333333333326,0.0,29.739333333333338,3.5,2539.3726666666666,26,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,0.0,493.5,5.5,1400.0,0.0,0.0,6.9111666666666665,3.5,1918.4111666666668,27,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-7.0,100.0,299.5,4.5,800.0,28.88133333333333,0.0,24.505166666666668,3.0,1267.3864999999998,28,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-8.0,0.0,210.0,4.5,440.0,29.748999999999995,0.0,24.656166666666664,2.5,719.4051666666667,29,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-7.0,0.0,313.0,5.0,310.0,40.477666666666664,0.0,26.6025,2.5,704.5801666666666,30,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,100.0,257.5,5.5,110.0,69.965,0.0,46.48566666666666,3.0,601.4506666666666,31,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,0.0,305.5,7.5,620.0,79.22399999999999,0.0,57.020833333333336,3.75,1084.9948333333334,32,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,200.0,168.5,5.0,270.0,125.87233333333333,200.0,69.63366666666667,2.75,1051.7559999999999,33,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,100.0,203.5,5.5,800.0,116.42533333333333,100.0,72.75016666666667,3.25,1412.4255,34,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,100.0,178.0,6.0,840.0,121.959,100.0,72.62383333333334,3.25,1433.8328333333334,35,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-9.0,100.0,257.0,6.5,1220.0,118.35299999999998,100.0,70.9795,3.5,1885.3325,36,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-11.0,100.0,342.0,6.0,730.0,100.79233333333335,100.0,58.06166666666667,3.25,1451.104,37,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-7.0,100.0,212.0,4.5,780.0,101.69466666666666,100.0,59.76166666666666,2.5,1367.4563333333333,38,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,100.0,504.0,8.5,380.0,61.19,0.0,42.419000000000004,4.75,1112.859,39,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-10.0,0.0,545.0,9.0,230.0,63.55433333333333,0.0,41.792500000000004,5.25,904.5968333333333,40,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20
-12.0,0.0,622.5,8.5,270.0,59.859333333333325,0.0,48.086,6.0,1026.9453333333333,41,tsroute,0.8,0.05,0.8,25.0,7,logbag,0,20

research/log.md

@@ -1,169 +0,0 @@
-# Autoresearch log
-
-One line per run, appended by run_experiment.py. Newest last.
-
-- 2026-08-19 10:36 lab@a510518ac3 tier0 BASELINE written: {"auc_all": 0.930567841842267, "auc_crossed": 0.8021138140264079, "count_err": 5.702596246648099, "crossed_count_err": 2.170831477801592, "dying_crossed_mean_p": 0.45313844808967824, "max_fit_s": 2.1}
-- 2026-08-19 10:37 lab@a510518ac3 tier1 BASELINE written: mean 1697.4
-- 2026-08-19 10:38 lab@a510518ac3 tier0 [lah_km14] {"auc_all": 0.9392896322708449, "auc_crossed": 0.6738938288632838, "count_err": 6.421327992384114, "crossed_count_err": 3.811722142568241, "dying_crossed_mean_p": 0.3302330619100046, "max_fit_s": 2.1} -> REJECT (screen)
-- 2026-08-19 10:40 lab@100054a46b tier0 [lah_cov] {"auc_all": 0.9455405609464158, "auc_crossed": 0.7583108203408955, "count_err": 6.686080189930141, "crossed_count_err": 3.2967281861400948, "dying_crossed_mean_p": 0.4597015461571341, "max_fit_s": 2.2} -> REJECT (screen)
-- 2026-08-19 11:13 lab@e28710fb9b tier0 [batch_demote_reorder] {"auc_all": 0.9555776554839356, "auc_crossed": 0.8332489435168008, "count_err": 5.702596246648099, "crossed_count_err": 2.1708314778015914, "dying_crossed_mean_p": 0.4583350850596277, "max_fit_s": 2.1} -> PROMOTE to tier 1
-- 2026-08-19 11:26 lab@e28710fb9b tier1 [batch_demote_reorder] mean_delta +86.3, wins 6/34 -> REJECT (tier 1)
-- 2026-08-19 11:35 lab@02c8a8d930 tier0 [batch_demote_agegate35] {"auc_all": 0.9493515837463812, "auc_crossed": 0.8191064444118955, "count_err": 5.702596246648098, "crossed_count_err": 2.1708314778015914, "dying_crossed_mean_p": 0.4539273042126147, "max_fit_s": 2.1} -> REJECT (screen)
-- 2026-08-19 11:38 lab@f05f907772 tier0 [agegate35_arrblend875] {"auc_all": 0.9499577772396863, "auc_crossed": 0.8235763244691817, "count_err": 5.702596246648099, "crossed_count_err": 2.1708314778015922, "dying_crossed_mean_p": 0.4566653338247193, "max_fit_s": 2.2} -> PROMOTE to tier 1
-- 2026-08-19 11:40 lab@f05f907772 tier1 [agegate35_arrblend875] mean_delta -7.6, wins 9/34 -> REJECT (tier 1)
-
-## Session summary 2026-08-19 (autonomous, 3 tier-0 + 2 tier-1 runs)
-
-- Family tested: mass-preserving within-cohort reorder (reassign logbag's own
-  multiset of crossed p42 values by an adjusted rank; count_err invariant by
-  construction - all three candidates hit baseline 5.703 exactly).
-- Offline forensics (tier-0 origins): crossed "ghost" negatives are largely
-  FUTURE positives - the same devices swap miss/ghost roles across origins
-  (d_e76cf26c8d1c ghost@16->dying@34, d_cfdd8b69ddd4 ghost@34->dying@40,
-  d_7fb0c50f956d ghost@34->dying@40). Every arrival model ranks the cohort
-  WORSE than logbag: covariate logistic 0.757, OOF-stacked logistic 0.754,
-  fit-time adaptive blend 0.779 vs logbag 0.802. Oracle reorder gives
-  dying_crossed_mean_p only 0.599 (gate arm needs 0.544 = 91% of oracle).
-- batch_demote_reorder (demote crossed rows whose last-report day is shared by
-  >=3 stale devices - phase-15 batch removals): t0 PROMOTE (auc_crossed 0.833)
-  but t1 +86.3 (6/34). Forensics: all six losing origins (+433..+898) demoted
-  a device with dropout age 4-31d that then died - the label chases the
-  dropout inside ~40d.
-- batch_demote_agegate35 (add staleness>35 age gate; full-panel audit: ZERO
-  dying devices demoted at any of 34 origins): t0 REJECT at 0.8191 vs gate
-  0.8221 - the missing 0.003 IS the fresh-batch ambiguity, unresolvable
-  without re-adding the tier-1 harm. kmin=2 audited unsafe (demotes a dying
-  pair d_497cf69db25f/d_f11c15757bf1).
-- agegate35_arrblend875 (safe demotion + 0.875 logbag/0.125 arrival rank
-  blend; blend weight admittedly tuned at the screen boundary): t0 PROMOTE
-  0.8236, t1 mean -7.6 wins 9/34 -> REJECT. First cost-negative candidate in
-  this channel; value concentrated (s_39 -296, s_10 -75, s_12 -80, s_22 -30,
-  s_28 -34, s_35 -35) vs blend noise (s_34 +206, s_41 +86).
-- Implication for the label-lag bucket: the only recoverable, safe signal is
-  aged-batch ghost demotion, worth ~ -8/origin panel mean - real but below
-  the pre-registered bar (-15 mean AND 18/34 wins). The remainder of the miss
-  bucket is the arrival process itself: for a crossed, still-reporting device,
-  WHICH 42d window the label lands in is unpredictable from every leak-free
-  covariate tried, and logbag's ~0.15-0.30 is the honest mixture probability.
-  Do not re-run: KM arrival (any bucket), covariate/stacked/adaptive arrival
-  logistics, ungated or kmin<3 batch demotion. A future session could price
-  the safe demotion as a plan post-filter (not via p42) or attack a different
-  miss bucket entirely.
-- Ops note: batteryswap_example/models_cv.py was edited externally at 13:06
-  (comment-only, EMERGENCY_DAY note) while frozen-pinned; restored to the
-  pinned HEAD state to unblock the runner. The edit is preserved at
-  /private/tmp/models_cv_human_edit_1306.py for the maintainer to re-apply.
-- 2026-08-19 12:05 lab@f05f907772 [sat] BASELINE written: mean 2115.1 (factor 4.0)
-
-## [sat] PAUSED mid-session: shared-file conflict on models_cv.py (2026-08-19, autonomous optimizer-tuning session)
-
-- Experiments completed: NONE. Zero s0 and zero s1 runs produced numbers.
-  Experiment 1 (sat_constructive: constructive=true, all else champion) was
-  configured in opt_lab.json but both launch attempts aborted at check_frozen.
-  opt_lab.json is left holding sat_constructive as the first experiment to run
-  on resume.
-- What happened: at session start batteryswap_example/models_cv.py carried an
-  uncommitted nn_order visit-ordering edit (mtime 14:07:35) belonging to
-  ANOTHER live session. To clear the frozen-hash violation I copied that state
-  byte-exactly to /private/tmp/models_cv_human_edit_nn_order_20260819.py
-  (sha256 67565183fec483edc064a23f2c9a9436218b1b8df00113d01c2584b604ad2726)
-  and then ran git checkout on the file - this wiped the other session's
-  working copy (recoverable from the /tmp copy above; the earlier 13:06
-  comment edit is also still at /private/tmp/models_cv_human_edit_1306.py,
-  sha256 3a17b540...). Working tree now holds the HEAD state
-  (sha256 22d0d92e3821140f036f8312d6c45367af9d9c58a62c204a171c7c45d2dba1b1).
-  Per coordinator directive I have NOT restored anything and will not touch
-  the file again; restoration is for the owning session/maintainer.
-- Forensic finding that blocks any naive resume: the frozen pin for
-  models_cv.py (fda0187468ab723771927949f27b0bebc44925d52925b9b4a9e5d1ca36048140,
-  frozen.json mtime 14:04:20) matches NO recoverable version: not git HEAD
-  (22d0d9...), not the 13:06 comment edit (3a17b5...), not the 14:07:35
-  nn_order state (675651...), not any of 12 committed revisions, not any
-  dangling git blob, and no leave-N-out subset of the 13 line-level diff
-  blocks between HEAD and the nn_order state (8192 combinations hashed). The
-  sat baseline (baselines/sat_champion.csv, mean 2115.1, factor 4.0, written
-  14:05:30) was computed under that now-lost 14:04:20 working-tree state.
-  models_cv.cpython-313.pyc was recompiled at 14:07:45 from the nn_order
-  state, so the pyc no longer witnesses the pinned source either.
-- Consequence: candidate-vs-baseline comparisons on the sat block are
-  currently impossible regardless of the shared-file conflict - the pinned
-  evaluation code exists nowhere. Resume requires a human to settle
-  models_cv.py, run --refreeze, and re-run --baseline for the sat block
-  (and to re-verify tier2_champion.csv comparability for the s1 native leg
-  if the settled code changes planner behavior, e.g. if nn_order defaults on).
-- Campaign state: budget untouched (0/10 s0, 0/3 s1). Hypothesis queue as
-  pre-registered: constructive=true first, then restarts/iters up, split_days
-  3-5 vs 10-14, alpha 0.6-0.75, ec_pmax 10-15, day_q 0.2, then combinations
-  of winners. Baseline decomposition to beat: late_swap 1022.4,
-  early_swap 445.7, overtime 111.5, daily_limit 261.8, weekly_limit 211.8,
-  travel 41.3 (mean 2115.1); ~585/scenario limit-driven attack surface.
-- 2026-08-19 12:31 lab@f05f907772 [sat] s0 [sat_constructive] delta +16.2, wins 0/6 -> REJECT (s0)
-- 2026-08-19 12:32 lab@f05f907772 [sat] s0 [sat_search6x120] delta -13.4, wins 1/6 -> REJECT (s0)
-- 2026-08-19 12:33 lab@f05f907772 [sat] s0 [sat_sd4] delta +19.4, wins 1/6 -> REJECT (s0)
-- 2026-08-19 12:33 lab@f05f907772 [sat] s0 [sat_sd12] delta -4.1, wins 4/6 -> REJECT (s0)
-- 2026-08-19 12:34 lab@f05f907772 [sat] s0 [sat_alpha07] delta -10.5, wins 4/6 -> REJECT (s0)
-- 2026-08-19 12:35 lab@f05f907772 [sat] s0 [sat_pm12] delta +116.3, wins 0/6 -> REJECT (s0)
-- 2026-08-19 12:36 lab@f05f907772 [sat] s0 [sat_dayq02] delta +0.0, wins 0/6 -> REJECT (s0)
-- 2026-08-19 12:36 lab@f05f907772 [sat] s0 [sat_alpha075] delta -53.9, wins 1/6 -> REJECT (s0)
-- 2026-08-19 12:37 lab@f05f907772 [sat] s0 [sat_combo_a075_sd12_r6i120] delta -54.6, wins 3/6 -> REJECT (s0)
-- 2026-08-19 12:38 lab@f05f907772 [sat] s0 [sat_combo_a075_sd12] delta -59.8, wins 4/6 -> PROMOTE to s1
-- 2026-08-19 12:44 lab@f05f907772 [sat] s1 [sat_combo_a075_sd12] sat +20.0 (18/34), native +42.0 -> REJECT (s1)
-
-## Session summary 2026-08-19 [sat] optimizer knobs under saturation (10 s0 + 1 s1; ALL REJECTED)
-
-Baseline: sat mean 2115.1 (s0-subset 1945.3); components late 1022.4 /
-early 445.7 / OT 111.5 / daily 261.8 / weekly 211.8 / travel 41.3.
-s0-subset components: 861.7 / 455.4 / 102.5 / 266.7 / 200.0 / 38.0.
-
-- sat_constructive (constructive=true): s0 +16.2, 0/6. Only weekly +16.7
-  moved - the greedy limit-aware start lands the search in WORSE weekly
-  basins. Initialization is not the lever.
-- sat_search6x120 (restarts 6, iters 120): s0 -13.4, 1/6, five exact ties.
-  One origin found a daily-limit escape (-16.7/origin daily = -80 once).
-  The champion search is effectively converged at 2x60.
-- sat_sd4 (split_days 4): s0 +19.4, 1/6; weekly +16.7. Finer buckets hurt.
-- sat_sd12 (split_days 12): s0 -4.1, 4/6, worst s_28 +20.7; late -8.4.
-  Broad but tiny timing win, no limit movement.
-- sat_alpha07 (alpha 0.7): s0 -10.5, 4/6; late +76.6 / early -64.1 /
-  daily -16.7; five winners avg ~-104 but s_16 +457.7 (excluded band died).
-- sat_pm12 (ec_pmax 12): s0 +116.3, 0/6, worst s_40 +697.9; late +133.3 and
-  daily/weekly BOTH +16.7. The ec top-up inclusions are load-bearing.
-- sat_dayq02 (day_q 0.2): s0 +0.0 exactly - plans identical; dead knob (the
-  downstream day-reassignment search dominates the initial quantile).
-- sat_alpha075 (alpha 0.75): s0 -53.9 but 1/6 - five exact ties (band empty),
-  one origin -324 (early -226, daily -100 there). Value totally concentrated.
-- sat_combo_a075_sd12_r6i120: s0 -54.6, 3/6, worst s_22 +88.7 (sd12 x r6i120
-  interaction creates a new loss). Mean gate passed, wins gate failed.
-- sat_combo_a075_sd12: s0 -59.8, 4/6 -> PROMOTE. s1: sat +20.0 (18/34,
-  worst s_17 +628.3), native +42.0 (13/34, worst s_23 +539.8) -> REJECT.
-  s1 forensics: 18 wins sum -1317 (median -47.9, best s_10 -345); 10 losses
-  sum +1999, of which FOUR blow-ups sum +1835 (s_17 +628, s_8 +459, s_31
-  +391 on a 959 base, s_23 +357). Full-block component deltas: late +43.5,
-  early -27.0, OT +1.3, daily +5.9, weekly -2.9, travel +0.3.
-
-Synthesis - what planning under saturation actually requires (or doesn't):
-1. The ~585/scenario limit spend is the shadow price of scheduling MASS, not
-   of scheduling choices. With the inclusion set fixed, required swap-hours
-   in peak weeks exceed the /4 limits and the day-assignment search has
-   already harvested nearly all rearrangement gains: +search finds one -80
-   escape across 6 origins, day_q is inert, both sd directions wash, the
-   constructive start is worse. Paying the fine is cheaper than the timing
-   cost of moving work out of peak weeks - the champion's non-adaptation is
-   the rational policy, consistent with the public leader's ~108/scenario
-   daily-limit spend.
-2. The only knob family that moves >50/scenario is inclusion (alpha,
-   ec_pmax), and it does NOT buy congestion relief: over 34 origins limits
-   moved daily +5.9 / weekly -2.9 while the cost went late +43.5 vs early
-   -27.0. Exclusion is a mortality bet with negative expectation and a fat
-   left... right tail: 4 of 34 origins contribute +1835. Saturation makes
-   this WORSE than natively (emergencies consume scarce worker-hours), so
-   inclusion-trimming is anti-correlated with the regime it targets.
-3. Screening lesson: 6-origin s0 is high-variance for inclusion knobs -
-   concentrated mortality events flipped -59.8 (4/6) to +20.0 at s1. The s1
-   mean gate (not the wins gate, which passed 18/34) is what caught it.
-4. Remaining untested allowed knob: ec_margin (alone). Judged indefensible
-   after alpha07/alpha075/pm12: it is the same inclusion-margin family whose
-   mechanism the s1 decomposition just showed to be limit-neutral and
-   mortality-negative. No adoption candidate this session; champion stands.
-- 2026-08-19 (manual, outside runner) pretrained-FM screen [chronos-bolt-small features]: tier0 replica (lab.py untouched, baseline reproduced exactly) fm_margin 0.757 / fm_pmax 0.801 / fm_trio 0.752 / fm_pmax_dark 0.781 vs auc_crossed base 0.802 -> ALL REJECT; see EXPERIMENTS.md phase 24. Do not re-test TS foundation models on the voltage series.
-- 2026-08-19 14:19 lab@d19ecc80c9 tier0 [hvac_env_logbag] {"auc_all": 0.9257928574006632, "auc_crossed": 0.7761705738914385, "count_err": 5.228063413500567, "crossed_count_err": 2.077773344847095, "dying_crossed_mean_p": 0.4502460011621175, "max_fit_s": 2.6} -> REJECT (screen)

research/opt_lab.json

@@ -1,4 +0,0 @@
-{
- "name": "champ_sanity",
- "notes": "session 2026-08-19 [sat] CLOSED: 10 s0 + 1 s1, all rejected, champion stands (see log.md summary). Config reset to champion knobs verbatim."
-}

research/program.md

@@ -1,87 +0,0 @@
-# Autoresearch program: the label-lag miss bucket
-
-## Mission
-
-Reduce walk-forward CV cost by improving p42 on **crossed devices** - batteries
-already below the 2.45 V death threshold at planning time whose recorded EOL
-arrives inside the 42-day window. The late-swap forensics (EXPERIMENTS.md):
-these are the largest
-recoverable miss bucket on the panel block (~half of the remaining s_8+ late
-cost; each miss pays 300-500). The incumbent logbag scores them p42 0.001-0.3
-because most crossed devices never get a recorded death (ghosts) - the label
-process, not a model bug. The open question is whether a *time-since-crossing
-arrival model* (or anything better) can separate "label coming" from "ghost"
-without wrecking the probability mass the planner runs on.
-
-## Rules
-
-1. **Mutate only `research/lab.py`.** Every other file is hash-pinned; the
-   runner aborts on any change. Never touch `batteryswap_public/` (the
-   competition evaluator) or `batteryswap_example/` (the champion pipeline).
-   **A frozen-guard failure is a STOP signal, never a fix-it signal**: do NOT
-   checkout, restore, or overwrite a frozen file to clear a hash violation -
-   other live sessions share these files and a restore destroys their work
-   (it happened once; see log.md). Stop runs, report, and wait for a human.
-2. **No labels at predict time.** The harness redacts eol/dies42/days_to_eol
-   from predict rows; do not try to reconstruct them. Labels at fit time are
-   legitimate (closed windows only, supplied by the walk-forward).
-3. **Runtime budget**: fit must stay under ~90 s/origin (deployment refits at
-   every scenario inside a 30-minute Docker limit). Tier 0 reports fit_s.
-4. **The metric is probability MASS, not ranking.** Inclusion count
-   n = 0.8 * sum(p42): a model that ranks better but inflates or compresses
-   the p42 scale LOSES (RF: AUC 0.94 vs 0.85, cost +199). Keep sum(p42) an
-   honest expected-death count. Tier 0 gates on count_err for this reason.
-5. One idea per experiment; write NOTES; the runner appends every result to
-   `research/log.md`. Read the log before each new experiment.
-
-## Protocol (tiers, promotion pre-registered)
-
-- **Tier 0** (~5 min): 6-origin screen, no planner.
-  `python research/run_experiment.py --tier 0`
-  Promote iff (auc_crossed >= baseline+0.02 OR dying_crossed_mean_p >= 1.2x
-  baseline) AND count_err <= 1.1x baseline AND max_fit_s <= 90.
-- **Tier 1** (~30 min): 34-origin panel, reduced day-search (restarts=0,
-  iters=20 - phase 7 showed the search is saturated, so deltas transfer).
-  Promote iff mean delta <= -15 AND wins >= 18/34.
-- **Tier 2** (~90 min): full panel + repeated-split selection stability.
-  ADOPTION RULE (pre-registered, EXPERIMENTS.md): candidate selected in >=60%
-  of 2000 random half-splits AND modal under blocked splits. Anything less is
-  a rejection no matter how good the mean looks - the panel's resolution is
-  ~150 cost units and mean-chasing mines noise. Adoption is a human decision.
-
-## What is already known NOT to work (do not re-run; EXPERIMENTS.md has details)
-
-- Model families on this table: GBM/HGB, RF/ExtraTrees (scale too compressed),
-  MLP, Weibull AFT, discrete hazard (tied, no win), Wiener first-hitting-time
-  (overdispersed mass), conformal cutoffs, all blends, recency weighting,
-  log1p compression, UCB inclusion, C != 1.0, more bags (>15 flat).
-- Feature families: cohort/batch clocks (coverage), spatial peers, telemetry
-  gaps, temperature stress, room cascade, global volatility features
-  (rng_30/vstd_30 as global columns), fleet-rank features, v_proj.
-- Label-process edits: GHOST_ZERO (zeroing crossed-forever devices, +547) and
-  QUARANTINE (dropping looks-dead-labeled-alive rows, +840, AUC collapse).
-  The "looks dead but survives" negatives are LOAD-BEARING; any crossed-device
-  fix must keep them in training.
-- Inclusion mechanics: fixed thresholds, conformal thresholds, EV top-ups at
-  native prices, alpha != 0.8, piggyback/routing-aware inclusion.
-
-## Promising, untested directions (start here)
-
-The seed candidate is the LogbagLAH draft from models_cv.py (KM arrival on
-crossing->EOL lag). Its design space, roughly in order of expected value:
-- KM bucket width (7/14/21), the v_corr gate (2.45-2.55), smooth blending of
-  arrival prob with logbag p42 instead of hard replacement.
-- Covariate arrival model: logistic on (days_below_245, v_corr_last,
-  staleness, reporting rate) trained ONLY on crossed pairs - richer than a
-  1-D KM, same leak-free inputs.
-- domain_split=True (train the WHO model only on not-crossed rows).
-- Censoring correction: ghosts cluster on batch-removal dates (phase 15) -
-  a removal-hazard competing risk may sharpen the arrival curve.
-- Calibration guard: rescale crossed-device p42 so sum over crossed pairs
-  matches the KM-expected arrival count (protects rule 4).
-
-## Baselines
-
-`research/baselines/` holds champion (logbag) results per tier; regenerate
-with `--baseline` only if frozen files legitimately changed (human decision,
-then `--refreeze`).

research/program_opt.md

@@ -1,66 +0,0 @@
-# Autoresearch program: optimizer knobs under worker-limit saturation
-
-## Mission
-
-Find optimizer/planner parameter settings that win in the **limit-saturated
-regime** - the deployment regime local CV never samples. Held-out fleets are
-~4.3x train size (EXPERIMENTS.md phase 16); fixed worker limits bind hard
-there (the public leader rationally pays ~108/scenario in daily-limit
-penalties; we pay ~0 locally because limits never bind). The harness
-(`research/run_sat.py`) reproduces the congestion ratio by dividing each
-scenario's worker limits by 4 on real panel origins - planner and evaluator
-share the transformed settings, so prices and scoring cannot disagree.
-
-## Rules
-
-1. **Mutate only `research/opt_lab.json`.** Everything else is hash-pinned.
-   Set `name` (slug) and `notes` per experiment. Allowed knobs: alpha,
-   ec_margin, ec_pmax, split_days, restarts, iters, constructive, day_q.
-   Unknown keys abort the run.
-2. One hypothesis per experiment. The runner enforces promotion thresholds -
-   trust its verdict, never reinterpret.
-3. Never run `--baseline` or `--refreeze`; never touch `research/baselines/`,
-   `batteryswap_example/`, or `batteryswap_public/`. **A frozen-guard failure
-   is a STOP signal, never a fix-it signal**: do NOT checkout, restore, or
-   overwrite a frozen file to clear a hash violation - other live sessions
-   share these files and a restore destroys their work (it happened once; see
-   log.md). Stop runs, report, and wait for a human.
-4. **No-regression is part of the objective**: a knob set that wins saturated
-   but loses the native panel is a rejection (s1 runs both). Adoption of any
-   change is a human decision under the pre-registered panel rule.
-
-## Protocol
-
-- **s0** (~5 min): 6-origin saturated screen.
-  `venv/bin/python research/run_sat.py --tier s0`
-  Promote iff mean delta <= -25 AND wins >= 4/6.
-- **s1** (~40 min): all 34 saturated origins + native no-regression.
-  `venv/bin/python research/run_sat.py --tier s1`
-  Candidate iff sat delta <= -30 AND sat wins >= 18/34 AND native delta <= +5.
-
-## Context and priors
-
-- The champion's knobs (alpha 0.8, ec 0.8/pm 25, sd 7, restarts 2/iters 60)
-  were ALL tuned where limits never bind. Under saturation the trade-offs
-  invert in known ways: spreading swaps across days/weeks avoids limit
-  penalties but costs timing; batching saves travel but concentrates hours.
-- `constructive=true` (greedy limit-aware placement start, drafted phase 18)
-  exists precisely for this regime and has never been evaluated in it.
-- More search (restarts/iters) may matter more here: the limit-penalty
-  landscape is spikier than the travel landscape the defaults were tuned on.
-- Lower alpha (fewer inclusions) may become optimal if marginal swaps push
-  days over limits; or higher day_q / larger split_days to spread visits.
-- A native-panel joint sweep of these same knobs is running separately - its
-  conclusion (champion expected to survive natively) does NOT bind the
-  saturated block. Do not tune for the native panel; the s1 native leg only
-  guards against regressions.
-- Emergency-day pricing (45 vs exact 48/50/52) was tested and REJECTED
-  natively (phase 21) and is not an available knob here.
-
-## Reporting
-
-Every run is appended to `research/log.md` with a `[sat]` prefix. At session
-end, append a summary block and report: every experiment (name, knobs, s0/s1
-deltas, verdict), the component decomposition (late/early/overtime/limits)
-for the interesting ones, and what the evidence says about planning under
-saturation.

research/run_experiment.py

@@ -1,261 +0,0 @@
-"""
-FROZEN evaluation harness for autoresearch candidates. Do not edit (hash-checked).
-
-Usage (from repo root, venv python):
-    python research/run_experiment.py --tier 0            # screen candidate (~5 min)
-    python research/run_experiment.py --tier 1            # reduced-search panel (~30 min)
-    python research/run_experiment.py --tier 2            # full panel + selection stability (~90 min)
-    python research/run_experiment.py --tier N --baseline # (re)compute the champion baseline
-    python research/run_experiment.py --refreeze          # human-only: re-pin frozen hashes
-
-The candidate lives in research/lab.py (the ONLY mutable file) and must expose
-`candidate()` returning an object with .fit(train_df) and .predict(rows_df)
-following the models_cv contract (predict returns a DataFrame indexed like rows
-with columns p42 and day_q{0.05,0.2,0.5,0.8}).
-
-Anti-leak: at predict time the candidate receives rows with all label columns
-removed. Labels are visible at fit time only (closed windows - that is the
-legitimate walk-forward training signal).
-"""
-
-import argparse
-import hashlib
-import importlib.util
-import json
-import sys
-import time
-from datetime import datetime, timezone
-from pathlib import Path
-
-import numpy as np
-import pandas as pd
-
-ROOT = Path(__file__).resolve().parent.parent
-RESEARCH = ROOT / 'research'
-BASELINES = RESEARCH / 'baselines'
-sys.path.insert(0, str(ROOT))
-
-import os
-os.chdir(ROOT)
-
-from batteryswap_example import models_cv as M
-from batteryswap_example.features import load as load_features
-
-LABEL_COLS = ['eol', 'dies42', 'days_to_eol', 'window_closed']
-TIER0_ORIGINS = [10, 16, 22, 28, 34, 40]
-PANEL = list(range(8, 42))
-CHAMP = dict(planner='tsroute', p_star=0.8, day_q=0.05,
-             ec_margin=0.8, ec_pmax=25.0, split_days=7)
-TIER1_SEARCH = dict(restarts=0, iters=20)   # reduced day-search (phase 7: saturated)
-TIER2_SEARCH = dict(restarts=2, iters=60)   # official settings
-
-FROZEN = [
-    'research/run_experiment.py',
-    'research/program.md',
-    'research/run_sat.py',
-    'research/program_opt.md',
-    'batteryswap_example/models_cv.py',
-    'batteryswap_example/features.py',
-    'batteryswap_example/aux_features.py',
-    'batteryswap_example/fht_features.py',
-]
-
-
-def _sha(path):
-    return hashlib.sha256((ROOT / path).read_bytes()).hexdigest()
-
-
-def check_frozen():
-    pins = json.loads((RESEARCH / 'frozen.json').read_text())
-    import batteryswap_public
-    pub = Path(batteryswap_public.__file__).parent  # evaluator lives in site-packages
-    bad = []
-    for p, h in pins.items():
-        if p.startswith('batteryswap_public/'):
-            actual = hashlib.sha256((pub / p.split('/', 1)[1]).read_bytes()).hexdigest()
-        else:
-            actual = _sha(p)
-        if actual != h:
-            bad.append(p)
-    if bad:
-        sys.exit(f'FROZEN-FILE VIOLATION: {bad} changed. Revert them; only '
-                 f'research/lab.py is mutable. (Humans: --refreeze after '
-                 f'intentional harness changes.)')
-
-
-def refreeze():
-    pins = {p: _sha(p) for p in FROZEN}
-    import batteryswap_public
-    pub = Path(batteryswap_public.__file__).parent
-    for f in ('evaluate.py', 'utils.py'):
-        pins[f'batteryswap_public/{f}'] = hashlib.sha256((pub / f).read_bytes()).hexdigest()
-    (RESEARCH / 'frozen.json').write_text(json.dumps(pins, indent=1))
-    print('frozen.json updated')
-
-
-class Guarded:
-    """Labels visible at fit (closed windows), REDACTED at predict."""
-
-    def __init__(self, inner):
-        self.inner = inner
-
-    def fit(self, train):
-        self.inner.fit(train)
-        return self
-
-    def predict(self, rows):
-        safe = rows.drop(columns=[c for c in LABEL_COLS if c in rows.columns])
-        return self.inner.predict(safe)
-
-
-def load_lab():
-    spec = importlib.util.spec_from_file_location('lab', RESEARCH / 'lab.py')
-    lab = importlib.util.module_from_spec(spec)
-    spec.loader.exec_module(lab)
-    return lab
-
-
-def load_table():
-    table = load_features(Path('dataset'))
-    table['v_proj'] = table['v_last'] + table['slope_lastk'] * table['staleness']
-    table['r_vlast'] = table.groupby('origin_idx')['v_last'].rank(pct=True)
-    table['r_tau'] = table.groupby('origin_idx')['tau_projected'].rank(pct=True)
-    return table
-
-
-def log_line(msg):
-    stamp = datetime.now(timezone.utc).strftime('%Y-%m-%d %H:%M')
-    lab_sha = _sha('research/lab.py')[:10]
-    with open(RESEARCH / 'log.md', 'a') as f:
-        f.write(f'- {stamp} lab@{lab_sha} {msg}\n')
-    print(f'[log] {msg}')
-
-
-# ------------------------------------------------------------------ tier 0
-
-def tier0(table, factory, name):
-    from sklearn.metrics import roc_auc_score
-    out = []
-    for idx in TIER0_ORIGINS:
-        rows = table[table['origin_idx'] == idx].set_index('device_id')
-        T = rows['origin'].iloc[0]
-        train = M.training_pairs(table, T)
-        t0 = time.time()
-        model = Guarded(factory()).fit(train)
-        fit_s = time.time() - t0
-        p = model.predict(rows)['p42']
-        y = rows['dies42']
-        crossed = rows['days_below_245'].notna() & (rows['days_below_245'] >= 0)
-        cy, cp = y[crossed], p[crossed]
-        dying_crossed = crossed & (y == 1)
-        out.append(dict(
-            origin=idx, fit_s=round(fit_s, 1),
-            auc_all=roc_auc_score(y, p) if 0 < y.sum() < len(y) else np.nan,
-            auc_crossed=(roc_auc_score(cy, cp)
-                         if 0 < cy.sum() < len(cy) else np.nan),
-            sum_p42=float(p.sum()), deaths=int(y.sum()),
-            crossed_sum_p42=float(cp.sum()), crossed_deaths=int(cy.sum()),
-            dying_crossed_mean_p=(float(p[dying_crossed].mean())
-                                  if dying_crossed.any() else np.nan)))
-    df = pd.DataFrame(out)
-    print(f'\n=== tier 0 [{name}] ===')
-    print(df.round(3).to_string(index=False))
-    summary = dict(
-        auc_all=float(df['auc_all'].mean()),
-        auc_crossed=float(df['auc_crossed'].mean()),
-        count_err=float((df['sum_p42'] - df['deaths']).abs().mean()),
-        crossed_count_err=float((df['crossed_sum_p42'] - df['crossed_deaths']).abs().mean()),
-        dying_crossed_mean_p=float(df['dying_crossed_mean_p'].mean()),
-        max_fit_s=float(df['fit_s'].max()))
-    print('summary:', {k: round(v, 3) for k, v in summary.items()})
-    return summary
-
-
-# --------------------------------------------------------------- tiers 1/2
-
-def run_panel(table, ctx, model_name, search, tag):
-    cfg = dict(CHAMP, model=model_name, **search)
-    df = M.walk_forward(table, ctx, PANEL, [cfg])
-    df.to_csv(RESEARCH / f'last_{tag}.csv', index=False)
-    return df
-
-
-def compare_panel(cand, base, tier):
-    m = cand.merge(base, on='origin', suffixes=('_cand', '_base'))
-    delta = m['total_cost_cand'] - m['total_cost_base']
-    wins = int((delta < -0.5).sum())
-    ties = int((delta.abs() <= 0.5).sum())
-    print(f'\n=== tier {tier} panel: candidate vs champion ===')
-    print(f'mean {m["total_cost_cand"].mean():.1f} vs {m["total_cost_base"].mean():.1f} '
-          f'(delta {delta.mean():+.1f}); wins {wins}/{len(m)}, ties {ties}')
-    worst = m.loc[delta.idxmax()]
-    print(f'worst origin s_{int(worst["origin"])}: {delta.max():+.1f}')
-    return delta.mean(), wins
-
-
-def main():
-    ap = argparse.ArgumentParser()
-    ap.add_argument('--tier', type=int, choices=(0, 1, 2))
-    ap.add_argument('--baseline', action='store_true')
-    ap.add_argument('--refreeze', action='store_true')
-    args = ap.parse_args()
-
-    if args.refreeze:
-        refreeze()
-        return
-    check_frozen()
-    BASELINES.mkdir(exist_ok=True)
-
-    table = load_table()
-    ctx = M.load_context(Path('dataset')) if args.tier != 0 else None
-
-    if args.tier == 0:
-        if args.baseline:
-            s = tier0(table, lambda: M.SklearnRUL('logbag'), 'logbag baseline')
-            (BASELINES / 'tier0.json').write_text(json.dumps(s, indent=1))
-            log_line(f'tier0 BASELINE written: {json.dumps(s)}')
-            return
-        lab = load_lab()
-        s = tier0(table, lab.candidate, lab.CANDIDATE_NAME)
-        b = json.loads((BASELINES / 'tier0.json').read_text())
-        print('\nbaseline :', {k: round(v, 3) for k, v in b.items()})
-        promote = (
-            (s['auc_crossed'] >= b['auc_crossed'] + 0.02
-             or s['dying_crossed_mean_p'] >= 1.2 * b['dying_crossed_mean_p'])
-            and s['count_err'] <= 1.1 * b['count_err']
-            and s['max_fit_s'] <= 90)
-        verdict = 'PROMOTE to tier 1' if promote else 'REJECT (screen)'
-        print(f'\nverdict: {verdict}')
-        log_line(f'tier0 [{lab.CANDIDATE_NAME}] {json.dumps(s)} -> {verdict}')
-        return
-
-    search = TIER1_SEARCH if args.tier == 1 else TIER2_SEARCH
-    tag = f'tier{args.tier}'
-    if args.baseline:
-        df = run_panel(table, ctx, 'logbag', search, f'{tag}_baseline')
-        df.to_csv(BASELINES / f'{tag}_champion.csv', index=False)
-        log_line(f'{tag} BASELINE written: mean {df["total_cost"].mean():.1f}')
-        return
-
-    lab = load_lab()
-    M.MODELS['candidate'] = lambda: Guarded(lab.candidate())
-    base = pd.read_csv(BASELINES / f'{tag}_champion.csv')
-    cand = run_panel(table, ctx, 'candidate', search, tag)
-    dmean, wins = compare_panel(cand, base, args.tier)
-
-    if args.tier == 1:
-        promote = dmean <= -15 and wins >= 18
-        verdict = 'PROMOTE to tier 2' if promote else 'REJECT (tier 1)'
-    else:
-        both = pd.concat([base, cand], ignore_index=True)
-        M.selection_stability(both)
-        verdict = ('CANDIDATE for adoption - APPLY THE PRE-REGISTERED RULE to '
-                   'the selection-stability output above (>=60% random AND '
-                   'modal blocked); adoption itself is a human decision')
-    print(f'\nverdict: {verdict}')
-    log_line(f'{tag} [{lab.CANDIDATE_NAME}] mean_delta {dmean:+.1f}, '
-             f'wins {wins}/34 -> {verdict}')
-
-
-if __name__ == '__main__':
-    main()

research/run_sat.py

@@ -1,152 +0,0 @@
-"""
-FROZEN saturated-regime harness. Do not edit (hash-checked).
-
-The held-out fleets are ~4.3x train size (EXPERIMENTS.md phase 16): the fixed
-worker limits bind there in a way no local scenario reproduces, and the
-public leader rationally pays ~108/scenario in daily-limit penalties. Same
-congestion ratio without fabricating data: divide the per-scenario worker
-limits (weekly, daily, overtime_start) by SAT_FACTOR on REAL panel origins.
-The planner and the evaluator both read the transformed settings, so the
-planner's prices cannot disagree with the scorer's (the phase-11 property).
-
-Usage (repo root, venv python):
-    python research/run_sat.py --baseline    # champion knobs on the sat block
-    python research/run_sat.py --tier s0     # opt_lab.json screen, 6 origins (~5 min)
-    python research/run_sat.py --tier s1     # 34 sat origins + native no-regression
-
-Mutable surface for optimizer experiments: research/opt_lab.json ONLY.
-Allowed knobs: alpha, ec_margin, ec_pmax, split_days, restarts, iters,
-constructive, day_q. Promotion thresholds are enforced here - do not
-reinterpret them.
-"""
-
-import argparse
-import importlib.util
-import json
-import os
-import sys
-from pathlib import Path
-
-import pandas as pd
-
-ROOT = Path(__file__).resolve().parent.parent
-RESEARCH = ROOT / 'research'
-BASELINES = RESEARCH / 'baselines'
-sys.path.insert(0, str(ROOT))
-os.chdir(ROOT)
-
-_spec = importlib.util.spec_from_file_location('re_mod', RESEARCH / 'run_experiment.py')
-RE = importlib.util.module_from_spec(_spec)
-_spec.loader.exec_module(RE)
-
-from batteryswap_example import models_cv as M
-
-SAT_FACTOR = float(os.environ.get('BATTERYSWAP_SAT_FACTOR', '4'))
-S0_ORIGINS = [10, 16, 22, 28, 34, 40]
-PANEL = list(range(8, 42))
-CHAMP = dict(model='logbag', planner='tsroute', p_star=0.8, day_q=0.05,
-             ec_margin=0.8, ec_pmax=25.0, split_days=7, restarts=2, iters=60)
-KNOB_MAP = {'alpha': 'p_star'}
-ALLOWED = {'alpha', 'ec_margin', 'ec_pmax', 'split_days', 'restarts', 'iters',
-           'constructive', 'day_q'}
-
-
-def sat_ctx(ctx):
-    out = {}
-    for idx, c in ctx.items():
-        s = c['settings']
-        s2 = s.model_copy(update=dict(
-            worker_limit_weekly_hours=s.worker_limit_weekly_hours / SAT_FACTOR,
-            worker_limit_daily_hours=s.worker_limit_daily_hours / SAT_FACTOR,
-            overtime_start=s.overtime_start / SAT_FACTOR))
-        out[idx] = dict(c, settings=s2)
-    return out
-
-
-def opt_cfg():
-    knobs = json.loads((RESEARCH / 'opt_lab.json').read_text())
-    name = knobs.pop('name', 'unnamed')
-    knobs.pop('notes', None)
-    bad = set(knobs) - ALLOWED
-    if bad:
-        sys.exit(f'opt_lab.json: unknown knobs {sorted(bad)}; allowed {sorted(ALLOWED)}')
-    cfg = dict(CHAMP)
-    for k, v in knobs.items():
-        cfg[KNOB_MAP.get(k, k)] = v
-    return cfg, name
-
-
-def compare(cand, base, label):
-    m = cand.merge(base[['origin', 'total_cost']], on='origin',
-                   suffixes=('', '_base'))
-    d = m['total_cost'] - m['total_cost_base']
-    wins = int((d < -0.5).sum())
-    print(f'\n=== {label} ===')
-    print(f'mean {m["total_cost"].mean():.1f} vs {m["total_cost_base"].mean():.1f} '
-          f'(delta {d.mean():+.1f}); wins {wins}/{len(m)}; '
-          f'worst s_{int(m["origin"][d.idxmax()])} {d.max():+.1f}')
-    return float(d.mean()), wins
-
-
-def components(df, label):
-    comps = ['late_swap', 'early_swap', 'overtime', 'daily_limit', 'weekly_limit',
-             'travel']
-    have = [c for c in comps if c in df.columns]
-    print(f'{label} components:',
-          {c: round(float(df[c].mean()), 1) for c in have})
-
-
-def main():
-    ap = argparse.ArgumentParser()
-    ap.add_argument('--tier', choices=('s0', 's1'))
-    ap.add_argument('--baseline', action='store_true')
-    args = ap.parse_args()
-
-    RE.check_frozen()
-    table = RE.load_table()
-    ctx = M.load_context(Path('dataset'))
-    sctx = sat_ctx(ctx)
-
-    if args.baseline:
-        df = M.walk_forward(table, sctx, PANEL, [dict(CHAMP)])
-        df.to_csv(BASELINES / 'sat_champion.csv', index=False)
-        components(df, 'sat champion')
-        RE.log_line(f'[sat] BASELINE written: mean {df["total_cost"].mean():.1f} '
-                    f'(factor {SAT_FACTOR})')
-        return
-
-    cfg, name = opt_cfg()
-    base = pd.read_csv(BASELINES / 'sat_champion.csv')
-
-    if args.tier == 's0':
-        cand = M.walk_forward(table, sctx, S0_ORIGINS, [cfg])
-        dmean, wins = compare(cand, base[base['origin'].isin(S0_ORIGINS)],
-                              f's0 sat screen [{name}]')
-        components(cand, 'candidate')
-        promote = dmean <= -25 and wins >= 4
-        verdict = 'PROMOTE to s1' if promote else 'REJECT (s0)'
-        print(f'\nverdict: {verdict}')
-        RE.log_line(f'[sat] s0 [{name}] delta {dmean:+.1f}, wins {wins}/6 -> {verdict}')
-        return
-
-    # s1: full saturated block + native no-regression
-    cand = M.walk_forward(table, sctx, PANEL, [cfg])
-    cand.to_csv(RESEARCH / 'last_sat_s1.csv', index=False)
-    dsat, wsat = compare(cand, base, f's1 saturated block [{name}]')
-    components(cand, 'candidate')
-
-    native = M.walk_forward(table, ctx, PANEL, [cfg])
-    tier2 = pd.read_csv(BASELINES / 'tier2_champion.csv')
-    dnat, _ = compare(native, tier2, f's1 native no-regression [{name}]')
-
-    ok = dsat <= -30 and wsat >= 18 and dnat <= 5
-    verdict = ('CANDIDATE for adoption - present sat AND native numbers to a '
-               'human; the pre-registered panel rule still applies to any '
-               'native change' if ok else 'REJECT (s1)')
-    print(f'\nverdict: {verdict}')
-    RE.log_line(f'[sat] s1 [{name}] sat {dsat:+.1f} ({wsat}/34), '
-                f'native {dnat:+.1f} -> {verdict}')
-
-
-if __name__ == '__main__':
-    main()