Upload 15 files

README.md

@@ -1,15 +1,55 @@
 ---
-title: Limitwise
-emoji: 💻
-colorFrom: indigo
-colorTo: indigo
+title: Corvana Bank LimitWise
+emoji: 💳
+colorFrom: green
+colorTo: gray
 sdk: gradio
 sdk_version: 6.31.0
-python_version: '3.13'
+python_version: "3.12"
 app_file: app.py
 pinned: false
 license: cc-by-4.0
-short_description: Credit default prediction model
+models:
+  - drdavidprivacy/limitwise-model
+datasets:
+  - drdavidprivacy/limitwise-data
+  - drdavidprivacy/practicum-case-packs
+short_description: LimitWise v2, the signed model and its repositories
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# LimitWise v2: Corvana Bank
+
+**Kept by:** Tom Becker, AI Governance Lead. **Signed by:** Anika Rao, Chief AI Officer (Week 6 memo).
+
+**Program:** AI Governance Practicum, Track 1, cohort 1
+
+This page is the public home of a governed system. The model is LimitWise v2, the version the Chief AI Officer signed in Week 6. The app has four tabs: one account (the three Week 6 cases and the record's own reasons), a batch upload (score and rank a file), monitoring (read a month against the reference file with Evidently's data-drift report), and the repositories (the links below).
+
+## The signed version
+
+- File: `limitwise_v2.pkl`, 18,636 bytes, SHA-256 `7842edc29acd316873ffa482db705ab2ea09427f8b4e1186436713d316de8ecb`
+- Recalibration object: `limitwise_v2_recalibrated.pkl`, 89,015 bytes, SHA-256 `b6508dfcc69ca108e3f08fc2be0843dab1d2517355970b8e622e52a34127ca37` (the isotonic recalibration the Week 6 memo ratified; the odds correction to the 3 percent book rate, 0.1089, is applied in the app)
+- Plain-language reasons: `reasons.csv` (one reason for every value of every input)
+- Operating point: the top 10 percent of the ranked file
+- Libraries: scikit-learn 1.9.1, fairlearn 0.15.0
+- Data: the remediated working file in the program's dataset repository, loaded by link, never copied here: https://huggingface.co/datasets/drdavidprivacy/practicum-case-packs/resolve/main/corvana_modeling_extract_remediated.csv (MD5 `f79f13a5119f78572e9f6cc3243ea6ae`)
+
+## Monitoring (Week 8)
+
+- `limitwise_reference.csv`: the Week 6 test set reweighted to the 3 percent book rate (every paying account four times, 578 defaulters once, 19,266 accounts), the file every month is read against. The Week 6 cut on this page is the score of the last account inside the top 10 percent of that file.
+- `limitwise_month_1.csv` to `_month_5.csv`: five monthly files of 20,000 accounts, drawn at the book rate. One has no drift, one has data drift the model shrugs off, one has drift that moves the scores and the outcomes, one has a performance drop, one has a fairness gap. Tab 3 runs Evidently's data-drift report on any of them, or on a file of your own, and reads the caught rate and the groups the Week 6 memo watches.
+
+## The repositories
+
+- Model repository (the model card, the signed file and its hash, the weights): https://huggingface.co/drdavidprivacy/limitwise-model
+- Dataset repository (the datasheet and the remediation log): https://huggingface.co/datasets/drdavidprivacy/limitwise-data
+- Portfolio page (the six deliverables as PDFs, the page each learner builds under their own name): https://huggingface.co/spaces/drdavidprivacy/corvana-portfolio
+- Collection: https://huggingface.co/collections/drdavidprivacy/limitwise-governance-repository-6ae5ff9b374cab20b18250eb
+
+## What this page is not
+
+A score says where a cardholder's record ranks on one list. It is not a decision about the cardholder. The decisions, and the conditions on them, are on the portfolio page.
+
+## Source
+
+Built for the AI Governance Practicum (AI Governance Collective). The organization, the people, and the program are fictional. The data is a public dataset, remediated and documented in the dataset repository's datasheet.

app.py

@@ -0,0 +1,383 @@
+"""The public page of a governed system: the signed model, its monitoring page, and the links to its repositories.
+Built for the AI Governance Practicum (AI Governance Collective). One app.py serves all three organizations; space_config.json says which."""
+import json, os, pickle, io, warnings
+import numpy as np, pandas as pd
+import gradio as gr
+import matplotlib
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt
+warnings.filterwarnings("ignore")
+
+HERE = os.path.dirname(os.path.abspath(__file__))
+
+# Hugging Face's free tier runs Gradio Spaces on ZeroGPU, which refuses to start an app that declares no GPU function.
+# This app needs no GPU: the declaration below satisfies the check and is never called. On a plain CPU Space it is skipped.
+try:
+    import spaces
+    @spaces.GPU
+    def _gpu_not_used():
+        return "This app runs on the CPU."
+except Exception:
+    pass
+CFG = json.load(open(os.path.join(HERE, "space_config.json")))
+REF = json.load(open(os.path.join(HERE, "reference.json")))
+MODEL = pickle.load(open(os.path.join(HERE, CFG["model_file"]), "rb"))
+RECAL = pickle.load(open(os.path.join(HERE, CFG["recal_file"]), "rb")) if CFG.get("recal_file") else None
+REASONS = pd.read_csv(os.path.join(HERE, CFG["reasons_file"])) if CFG.get("reasons_file") else None
+EXAMPLES = pd.read_csv(os.path.join(HERE, "examples.csv"), keep_default_na=False)
+INPUTS, KEPT, SHARE = CFG["inputs"], CFG["kept"], CFG["share"]
+SPACE_ID = os.environ.get("SPACE_ID", "")
+WEEK6 = CFG["week6"]
+
+# ----------------------------------------------------------------------------- scoring
+def score_frame(X):
+    """The production score. Where the Week 6 memo ratified fitted steps (the recalibration and the odds correction), they run here."""
+    X = X[INPUTS]
+    if RECAL is not None:
+        pf = RECAL.predict_proba(X)[:, 1]; f = CFG["odds_factor"]
+        return pf * f / (pf * f + 1 - pf)
+    return MODEL.predict_proba(X)[:, 1]
+
+def contributions(X):
+    """Each input's fixed weight times the record's value on the scaled, remover-adjusted inputs. The record's own reasons."""
+    Z = X[INPUTS]
+    for name, step in MODEL.steps[:-1]:
+        Z = step.transform(Z)
+    coef = MODEL.named_steps["model"].coef_[0]
+    return pd.DataFrame(np.asarray(Z) * coef, columns=KEPT, index=X.index)
+
+def allowed_reasons(con):
+    """Drop the columns a reason may never cite: a medication the record marks No (Alderwyn, Week 6 memo, ruling 5)."""
+    drugs = CFG.get("drug_fields", [])
+    if not drugs: return con
+    keep = [c for c in con.index if not any(c == d + "_No" for d in drugs)]
+    return con[keep]
+
+def allowed_reasons_frame(frame):
+    drugs = CFG.get("drug_fields", [])
+    return frame[[c for c in frame.columns if not any(c == d + "_No" for d in drugs)]] if drugs else frame
+
+def readable(col):
+    if REASONS is not None:
+        r = REASONS.loc[(REASONS.input == col) | ((REASONS.input + "_" + REASONS.value.astype(str)) == col), "reason"]
+        if len(r): return r.iloc[0]
+    for c in CFG["categorical"]:
+        if col.startswith(c + "_"): return f"{c} is {col[len(c) + 1:]}"
+    return col
+
+_ENC = MODEL.named_steps[MODEL.steps[0][0]]
+_CAT_KIND = {}
+for _name, _trans, _cols in _ENC.transformers_:
+    if _name == "cat":
+        for _c, _cats in zip(_cols, _trans.categories_): _CAT_KIND[_c] = "num" if np.issubdtype(np.asarray(_cats).dtype, np.number) else "str"
+
+def coerce(df):
+    """Read a file the way the working file was read when the model was fitted: numeric inputs as numbers, categorical inputs as text,
+    except the coded categoricals the encoder learned as numbers (a code column read from a CSV as text would otherwise be unknown to it)."""
+    df = df.copy()
+    for c in CFG["numeric"]:
+        if c in df: df[c] = pd.to_numeric(df[c], errors="coerce")
+    for c in CFG["categorical"]:
+        if c in df: df[c] = pd.to_numeric(df[c], errors="coerce") if _CAT_KIND.get(c) == "num" else df[c].astype(str)
+    return df
+
+def groups_of(frame):
+    g = {}
+    for gname, d in CFG["groups"].items():
+        col = d["col"]
+        if col not in frame: continue
+        if "bands" in d:
+            v = pd.to_numeric(frame[col], errors="coerce"); lab = pd.Series(["unrecorded"] * len(frame), index=frame.index, dtype=object)
+            for lo, hi, name in d["bands"]: lab[(v >= lo) & (v <= hi)] = name
+            g[gname] = lab
+        else:
+            s = frame[col].astype(str); g[gname] = s.map(d["map"]).fillna(s) if d.get("map") else s
+    return pd.DataFrame(g, index=frame.index)
+
+def pct(x, d=1): return f"{x * 100:.{d}f} percent"
+
+# ----------------------------------------------------------------------------- the signed version (the collapsed panel at the top)
+def model_header():
+    a = CFG["artifacts"][CFG["model_file"]]
+    L = [f"**Organization:** {CFG['org']}. **Signed by:** {CFG['caio']} ({CFG['memo_ref']}). **Kept by:** {CFG['lead']}.",
+         f"**File:** `{CFG['model_file']}`, {a['bytes']:,} bytes, SHA-256 `{a['sha256']}`."]
+    if RECAL is not None:
+        r = CFG["artifacts"][CFG["recal_file"]]
+        L.append(f"**Fitted steps the memo ratified:** the isotonic recalibration (`{CFG['recal_file']}`, SHA-256 `{r['sha256']}`), then the odds correction to the {int(CFG['calibration']['book_rate']*100)} percent book rate (factor {CFG['odds_factor']}). Every score on this page has both applied.")
+    L.append(f"**What it is:** a logistic regression with {len(KEPT)} fitted columns from {len(INPUTS)} input fields. The weights are fixed. The same record gets the same score every time.")
+    L.append(f"**Operating point:** the top {int(SHARE*100)} percent of the ranked file gets {CFG['action']}. On the Week 6 test set ({WEEK6['test_rows']:,} {CFG['units']}) that was {WEEK6['flagged']:,} {CFG['units']}, of which {WEEK6['caught']:,} were {CFG['positive_label']}. Test-set ROC-AUC {WEEK6['roc_auc']:.3f}.")
+    L.append(f"**Libraries:** scikit-learn {CFG['versions']['sklearn']}, fairlearn {CFG['versions']['fairlearn']}. The model does not load on another version of scikit-learn, by design.")
+    return "\n\n".join(L)
+
+def weights_table(n=25):
+    return pd.DataFrame([dict(column=c, weight=round(w, 4), meaning=readable(c)) for c, w in CFG["weights"][:n]])
+
+# ----------------------------------------------------------------------------- tab 1: one record
+CAT = [c for c in INPUTS if c in CFG["categorical"]]
+NUM = [c for c in INPUTS if c in CFG["numeric"]]
+FORM = CAT + NUM   # the order of the form's controls
+
+def _typical():
+    d = {c: REF["numeric"][c]["median"] for c in NUM}
+    for c in CAT:
+        r = REF["categorical"][c]; d[c] = max(r, key=r.get)
+    return d
+
+def _case_row(i):
+    row = EXAMPLES.iloc[i]
+    return {c: (float(row[c]) if c in NUM else str(row[c])) for c in FORM}
+
+PRESETS = {f"Typical {CFG['unit']} (training-file medians and most common values)": _typical()}
+for i in range(len(EXAMPLES)):
+    PRESETS[f"{EXAMPLES['case'].iloc[i].capitalize()} (Week 6 case, rank {EXAMPLES['week6_rank'].iloc[i]} of {WEEK6['test_rows']:,})"] = _case_row(i)
+FIRST = list(PRESETS)[0]
+
+def apply_preset(name):
+    d = PRESETS[name]
+    return [d[c] for c in FORM]
+
+def field_label(c):
+    """The input's name, with the meaning of its codes where the program's group definitions give one (SEX: 1 male, 2 female)."""
+    for d in CFG["groups"].values():
+        if d["col"] == c and d.get("map"):
+            return c + " (" + ", ".join(f"{k} {v}" for k, v in d["map"].items()) + ")"
+    return c
+
+def choices_for(c):
+    ch = sorted(REF["categorical"][c], key=str)
+    for v in EXAMPLES[c].astype(str).unique():
+        if v not in ch: ch.append(v)
+    return ch
+
+def score_one(*vals):
+    row = {c: v for c, v in zip(FORM, vals)}
+    X = coerce(pd.DataFrame([row]))
+    p = float(score_frame(X)[0]); cut = REF["cut_score"]; inside = p >= cut
+    verdict = (f"## Score: **{p:.1%}**\n\n"
+               f"### {'🔴 Above the cut' if inside else '🟢 Below the cut'}  (Week 6 cut = {cut:.1%}, the score of the last {CFG['unit']} inside the top {int(SHARE*100)} percent)\n\n"
+               f"At the operating point the {CFG['memo_ref']} set, this {CFG['unit']} {'would' if inside else 'would not'} be {CFG['flagged_word']}: {CFG['action']}. "
+               f"The score says where the record ranks. It is not a decision about the {CFG['person']}.")
+    con = allowed_reasons(contributions(X).iloc[0])
+    up = con[con > 0].sort_values(ascending=False).head(6); down = con[con < 0].sort_values().head(4)
+    items = list(up.items()) + list(down.items())
+    labs = [readable(c) for c, _ in items][::-1]; v = [float(x) for _, x in items][::-1]
+    fig, ax = plt.subplots(figsize=(11, 0.45 * len(v) + 1.8))
+    ax.barh(range(len(v)), v, color=["#d62728" if x > 0 else "#2ca02c" for x in v])
+    ax.set_yticks(range(len(v))); ax.set_yticklabels(labs, fontsize=9); ax.axvline(0, color="black", lw=0.8)
+    ax.set_xlabel(f"Toward {CFG['positive_label']} (red) or away from it (green): the fixed weight times this record's value", fontsize=9)
+    ax.set_title(f"Why {CFG['system']} scored this {CFG['unit']} {p:.1%}: the record's own reasons", fontsize=11, loc="left")
+    fig.tight_layout()
+    return verdict, fig
+
+# ----------------------------------------------------------------------------- tab 2: batch upload
+def score_batch(file, top_n):
+    if file is None: return None, f"Upload a CSV of {CFG['units']} to score (the model's input columns, one row per {CFG['unit']}).", None
+    try: df = coerce(pd.read_csv(file.name if hasattr(file, "name") else file, low_memory=False, keep_default_na=False, na_values=[]))
+    except Exception as e: return None, f"Could not read the file: {e}", None
+    missing = [c for c in INPUTS if c not in df.columns]
+    if missing: return None, "The file is missing input columns: " + ", ".join(missing) + ". Nothing was scored.", None
+    s = score_frame(df); order = np.argsort(-s, kind="stable"); k = int(round(SHARE * len(df)))
+    idc = CFG.get("id_col"); ids = df[idc].values if idc and idc in df.columns else np.arange(1, len(df) + 1)
+    con = allowed_reasons_frame(contributions(df)); top = [readable(con.columns[i]) for i in np.argmax(con.values, axis=1)]
+    out = pd.DataFrame({(idc if idc and idc in df.columns else "row"): ids, "score": np.round(s, 4), "ranked_in": np.where(np.arange(len(df)) >= 0, "", ""), "top_reason": top})
+    out = out.iloc[order].reset_index(drop=True); out.insert(0, "rank", np.arange(1, len(df) + 1)); out["ranked_in"] = np.where(out["rank"] <= k, "yes", "no")
+    path = os.path.join("/tmp", "ranked_list.csv"); out.to_csv(path, index=False)
+    msg = (f"Scored **{len(df):,}** {CFG['units']}. The top {int(SHARE*100)} percent, the operating point, is **{k:,}** {CFG['units']}. "
+           f"Showing the top **{int(top_n)}** by score (the team's capacity). The download carries every record with its score, rank, whether it is ranked in, and its own top reason.")
+    return out.head(int(top_n)), msg, path
+
+# ----------------------------------------------------------------------------- tab 3: monitoring (Evidently)
+import html as _html
+REFERENCE = pd.read_csv(os.path.join(HERE, f"{CFG['short']}_reference.csv"), low_memory=False, keep_default_na=False, na_values=[])
+MONTHS = {f"Month {i}": f"{CFG['short']}_month_{i}.csv" for i in (1, 2, 3, 4, 5) if os.path.exists(os.path.join(HERE, f"{CFG['short']}_month_{i}.csv"))}
+MIN_GROUP = 30   # a group's rate is shown but not judged below this many of the people it is a rate of
+SCORE_COL, FLAG_COL = "predicted_score", "flagged"
+
+def monitoring_frame(df):
+    """The columns Evidently reads: every field in the file except the identifier, plus the model's score and the flag. Numbers as numbers, text as text."""
+    df = coerce(df); out = pd.DataFrame(index=df.index)
+    skip = {CFG.get("id_col")} | {SCORE_COL, FLAG_COL}
+    num, cat = [], []
+    for c in df.columns:
+        if c in skip or c is None: continue
+        if c in CFG["numeric"] or (c not in CFG["categorical"] and pd.api.types.is_numeric_dtype(df[c]) and c != CFG["outcome"]):
+            out[c] = pd.to_numeric(df[c], errors="coerce").astype("float64").to_numpy(); num.append(c)
+        else:
+            out[c] = df[c].astype(str).to_numpy().astype(object); cat.append(c)
+    s = score_frame(df); out[SCORE_COL] = s.astype("float64"); num.append(SCORE_COL)
+    out[FLAG_COL] = np.where(s >= REF["cut_score"], "flagged", "not flagged").astype(object); cat.append(FLAG_COL)
+    return out, num, cat
+
+def rates(frame, s):
+    """Caught rate (the share of {positives} the model flagged) and false alarm rate (the share of the others it flagged), for the file and by group."""
+    y = (frame[CFG["outcome"]].astype(str) == str(CFG["positive"])).to_numpy(); flag = s >= REF["cut_score"]
+    whole = dict(caught=float(flag[y].mean()) if y.any() else None, false_alarm=float(flag[~y].mean()) if (~y).any() else None, positives=int(y.sum()))
+    G = groups_of(frame); by = {}
+    for dim in G.columns:
+        rows = []
+        for name in sorted(G[dim].astype(str).unique(), key=str):
+            m = (G[dim].astype(str) == name).to_numpy(); yp, yn = y & m, (~y) & m
+            rows.append(dict(group=name, n=int(m.sum()), positives=int(yp.sum()), negatives=int(yn.sum()),
+                             caught=float(flag[yp].mean()) if yp.sum() else None, false_alarm=float(flag[yn].mean()) if yn.sum() else None))
+        by[dim] = rows
+    return whole, by
+
+def gap(rows, key, denom):
+    use = [r[key] for r in rows if r[key] is not None and r[denom] >= MIN_GROUP]
+    return (max(use) - min(use)) if len(use) >= 2 else None
+
+def fmt(x, d=1): return "" if x is None else f"{x * 100:.{d}f}"
+
+def group_table(by_ref, by_cur):
+    out = []
+    for dim, rows in by_cur.items():
+        ref = {r["group"]: r for r in by_ref.get(dim, [])}
+        for r in rows:
+            rr = ref.get(r["group"], {})
+            out.append({"group": f"{dim}: {r['group']}", f"{CFG['units']} this month": r["n"], f"of which {CFG['positive_label']}": r["positives"],
+                        "caught rate, reference": fmt(rr.get("caught")), "caught rate, this month": fmt(r["caught"]) + ("" if r["positives"] >= MIN_GROUP else " (too few to judge)"),
+                        "false alarm rate, reference": fmt(rr.get("false_alarm")), "false alarm rate, this month": fmt(r["false_alarm"]) + ("" if r["negatives"] >= MIN_GROUP else " (too few to judge)")})
+    return pd.DataFrame(out)
+
+def monitor(choice, file):
+    """Run Evidently's data-drift report for one month against the reference file, and summarize the signals above it:
+    the inputs, the scores, the outcomes, the caught rate, and the groups the Week 6 memo watches."""
+    if file is not None:
+        try: cur = pd.read_csv(file.name if hasattr(file, "name") else file, low_memory=False, keep_default_na=False, na_values=[]); label = "your file"
+        except Exception as e: return f"Could not read the file: {e}", None, "", None
+    elif choice in MONTHS:
+        cur = pd.read_csv(os.path.join(HERE, MONTHS[choice]), low_memory=False, keep_default_na=False, na_values=[]); label = choice
+    else:
+        return "Pick a month, or upload a file.", None, "", None
+    missing = [c for c in INPUTS if c not in cur.columns]
+    if missing: return "The file is missing input columns: " + ", ".join(missing) + ". Nothing was read.", None, "", None
+    from evidently import Dataset, DataDefinition, Report
+    from evidently.presets import DataDriftPreset
+    ref_f, num, cat = monitoring_frame(REFERENCE)
+    cur_f, num_c, cat_c = monitoring_frame(cur)
+    cols = [c for c in ref_f.columns if c in cur_f.columns]
+    num = [c for c in num if c in cols]; cat = [c for c in cat if c in cols]
+    dd = DataDefinition(numerical_columns=num, categorical_columns=cat)
+    # two-sided proportion test for the yes/no columns (the flag and the outcome), where a distance score reads a real shift as small
+    binary = [c for c in (FLAG_COL, CFG["outcome"]) if c in cat]
+    preset = DataDriftPreset(per_column_method={c: "z" for c in binary}, per_column_threshold={c: 0.05 for c in binary})
+    snap = Report([preset]).run(Dataset.from_pandas(cur_f[cols], data_definition=dd), Dataset.from_pandas(ref_f[cols], data_definition=dd))
+    ms = snap.dict()["metrics"]
+    def is_drift(m):
+        c = m.get("config", {}); col = c.get("column")
+        if not col: return False
+        if col in binary: return float(m["value"]) < c.get("threshold", 0.05)   # a p-value
+        return float(m["value"]) > c.get("threshold", 0.1)                        # a distance
+    drifted = [m["config"]["column"] for m in ms[1:] if is_drift(m)]
+    inputs_drifted = [c for c in drifted if c in INPUTS]; other_drifted = [c for c in drifted if c not in INPUTS and c not in (SCORE_COL, FLAG_COL, CFG["outcome"])]
+    s_ref, s_cur = ref_f[SCORE_COL].values, cur_f[SCORE_COL].values
+    f_ref, f_cur = float((s_ref >= REF["cut_score"]).mean()), float((s_cur >= REF["cut_score"]).mean())
+    L = [f"**{label}:** {len(cur):,} {CFG['units']} read against the reference file ({len(REFERENCE):,} {CFG['units']}, {CFG.get('reference_label', 'the Week 6 test set')}). Every column in the file is compared, plus the model's score and whether the {CFG['unit']} is flagged at the Week 6 cut.", ""]
+    L.append(f"**The inputs.** {len(inputs_drifted)} of {len(INPUTS)} input columns drifted" + (": " + ", ".join(inputs_drifted) + "." if inputs_drifted else ".") + (f" Other fields in the file that drifted (not model inputs): {', '.join(other_drifted)}." if other_drifted else ""))
+    L.append(f"**The scores.** Mean score {s_ref.mean():.4f} in the reference, {s_cur.mean():.4f} this month. Share flagged at the Week 6 cut: {pct(f_ref)} in the reference, {pct(f_cur)} this month." + (" The score column drifted." if SCORE_COL in drifted else " The score column did not drift."))
+    table = None
+    if CFG["outcome"] in cur.columns:
+        o_ref = float((REFERENCE[CFG["outcome"]].astype(str) == str(CFG["positive"])).mean()); o_cur = float((cur[CFG["outcome"]].astype(str) == str(CFG["positive"])).mean())
+        L.append(f"**The outcomes.** {CFG['positive_label'].capitalize()}: {pct(o_ref)} of the reference, {pct(o_cur)} this month." + (" The outcome column drifted." if CFG["outcome"] in drifted else " The outcome column did not drift."))
+        w_ref, by_ref = rates(REFERENCE, s_ref); w_cur, by_cur = rates(cur, s_cur)
+        L.append(f"**The caught rate.** Of the {CFG['units']} whose outcome is {CFG['positive_label']}, the share the model flagged at the Week 6 cut: {pct(w_ref['caught'])} in the reference, {pct(w_cur['caught'])} this month. "
+                 f"False alarm rate (the share of the others it flagged): {pct(w_ref['false_alarm'])} in the reference, {pct(w_cur['false_alarm'])} this month.")
+        parts = []
+        for dim in by_cur:
+            g_ref, g_cur = gap(by_ref[dim], "caught", "positives"), gap(by_cur[dim], "caught", "positives")
+            f_ref, f_cur = gap(by_ref[dim], "false_alarm", "negatives"), gap(by_cur[dim], "false_alarm", "negatives")
+            bits = []
+            if g_ref is not None and g_cur is not None: bits.append(f"caught-rate gap {fmt(g_ref)} to {fmt(g_cur)} points")
+            if f_ref is not None and f_cur is not None: bits.append(f"false alarm gap {fmt(f_ref)} to {fmt(f_cur)} points")
+            if bits: parts.append(f"{dim}: " + ", ".join(bits))
+        L.append("**The groups.** The widest gap between any two groups (groups with at least 30 of the people the rate is about), reference to this month. " + "; ".join(parts) + ". The table below has every group.")
+        table = group_table(by_ref, by_cur)
+    else:
+        L.append("**The outcomes.** This file carries no outcome column, so the outcome rate, the caught rate, and the group gaps cannot be read yet.")
+    L.append("\nThe full report is below (Evidently, data drift preset: Wasserstein distance for numbers, Jensen-Shannon distance for categories, drift called at 0.1, and a proportion test for the flag and the outcome). Download it as a page to keep with your monitoring read.")
+    report_html = snap.get_html_str(as_iframe=False)
+    path = os.path.join("/tmp", f"{CFG['short']}_{label.replace(' ', '_')}_evidently_report.html"); open(path, "w", encoding="utf-8").write(report_html)
+    frame = f'<iframe srcdoc="{_html.escape(report_html, quote=True)}" style="width:100%;height:1400px;border:1px solid #ddd;border-radius:8px;background:white"></iframe>'
+    return "\n".join(L), table, frame, path
+
+# ----------------------------------------------------------------------------- tab 4: the repositories
+def repositories_markdown():
+    short = CFG["short"]
+    L = [f"## The repositories: {CFG['system']} at {CFG['org']}", "",
+         "A governed system leaves its record in several places, and this page is only one of them. Each link opens on Hugging Face.", "",
+         f"- **Model repository**, [drdavidprivacy/{short}-model](https://huggingface.co/drdavidprivacy/{short}-model): the model card, the signed file and its SHA-256, and the fixed weights as a table. The card on the page is the AI Governance Lead's exemplar.",
+         f"- **Dataset repository**, [drdavidprivacy/{short}-data](https://huggingface.co/datasets/drdavidprivacy/{short}-data): the datasheet and the remediation log for the working file. The data itself is linked, not copied.",
+         f"- **Portfolio page**, [drdavidprivacy/{CFG['portfolio']}](https://huggingface.co/spaces/drdavidprivacy/{CFG['portfolio']}): the six deliverables of the practicum as PDFs, in course order, on a page with no code behind it. Each learner builds one of these under their own name.",
+         f"- **Collection**, [{CFG['collection_name']}]({CFG['collection_url']}): all of the above in one place.", ""]
+    L += [f"**Monitoring targets from the {CFG['memo_ref']}:**"] + [f"- {m}" for m in CFG["monitoring_lines"]]
+    L += ["", f"**Data:** [{CFG['data_file']}]({CFG['data_url']}) in the program's dataset repository (MD5 `{CFG['md5']}`). The data is not copied into this page."]
+    return "\n".join(L)
+
+# ----------------------------------------------------------------------------- the page
+INTRO = f"""
+# {CFG['system']} {CFG['version']}: {CFG['org']}
+**Kept by: {CFG['lead']}** · a teaching deployment for the *AI Governance Practicum*, Track 1
+
+---
+
+{CFG['system']} estimates, for each {CFG['unit']}, the likelihood of {CFG['positive_label']}, ranks the file, and sends the top {int(SHARE*100)} percent to a person for {CFG['action']}. This page runs the signed version ({CFG['memo_ref']}). It is for **education**, not real decisions. A score says where a {CFG['person']}'s record ranks on one list. It is not a decision about the {CFG['person']}.
+"""
+
+with gr.Blocks(title=f"{CFG['org']} {CFG['system']}") as demo:
+    gr.Markdown(INTRO)
+    with gr.Accordion("About the signed version: the file, its hash, the fitted steps, the weights", open=False):
+        gr.Markdown(model_header())
+        gr.Markdown("**The fixed weights.** The largest by size, with what each column means. A positive weight pushes the score up.")
+        gr.Dataframe(weights_table(), interactive=False, wrap=True)
+
+    with gr.Tab(f"1 · Single {CFG['unit']}"):
+        gr.Markdown(f"Score one {CFG['unit']}, then read the **reasons**. Load a Week 6 case to fill the form, or edit any field and score again.")
+        with gr.Row():
+            preset = gr.Dropdown(list(PRESETS), label="Load an example", value=FIRST)
+        d0 = PRESETS[FIRST]
+        widgets = {}
+        with gr.Accordion("Categorical inputs", open=True):
+            for i in range(0, len(CAT), 4):
+                with gr.Row():
+                    for c in CAT[i:i + 4]:
+                        widgets[c] = gr.Dropdown(choices_for(c), value=d0[c], label=field_label(c), allow_custom_value=True)
+        with gr.Accordion("Numeric inputs", open=(len(CAT) == 0)):
+            for i in range(0, len(NUM), 4):
+                with gr.Row():
+                    for c in NUM[i:i + 4]:
+                        widgets[c] = gr.Number(value=d0[c], label=c)
+        ordered = [widgets[c] for c in FORM]
+        score_btn = gr.Button(f"Score this {CFG['unit']}", variant="primary")
+        out_md = gr.Markdown(); out_plot = gr.Plot()
+        preset.change(apply_preset, preset, ordered)
+        score_btn.click(score_one, ordered, [out_md, out_plot])
+
+    with gr.Tab("2 · Batch upload"):
+        gr.Markdown(f"Upload a CSV of {CFG['units']} (the same columns as the working file). {CFG['system']} scores every row, ranks the file, and marks the top {int(SHARE*100)} percent, the operating point the {CFG['memo_ref']} set. Top-N is how many of the ranked list the team can work. `{CFG['short']}_month_1.csv` in this repository is a file to try.")
+        with gr.Row():
+            up = gr.File(label=f"{CFG['units'].capitalize()} CSV", file_types=[".csv"])
+            with gr.Column():
+                topn = gr.Slider(5, 100, value=20, step=5, label="Top-N by capacity (how many the team can work)")
+        batch_btn = gr.Button("Score & rank", variant="primary")
+        batch_msg = gr.Markdown(); batch_tbl = gr.Dataframe(interactive=False, wrap=True); batch_dl = gr.File(label="Download the full ranked list")
+        batch_btn.click(score_batch, [up, topn], [batch_tbl, batch_msg, batch_dl])
+
+    with gr.Tab("3 · Monitoring"):
+        gr.Markdown(f"**One month at a time.** Pick one of the five monthly files, or upload a file of your own, and read it against the reference file ({CFG.get('reference_label', 'the Week 6 test set')}). The page scores every row with the signed version, runs Evidently's data-drift report (every column in the file, the model's score, and the flag at the Week 6 cut), and, where the file carries the outcome, reads the caught rate and the groups the Week 6 memo watches. The five files and the reference are in this repository's Files tab (`{CFG['short']}_reference.csv`, `{CFG['short']}_month_1.csv` to `_5.csv`).")
+        if CFG.get("reference_note"): gr.Markdown(CFG["reference_note"])
+        with gr.Row():
+            m_choice = gr.Dropdown(choices=list(MONTHS), value=(list(MONTHS)[0] if MONTHS else None), label="Monthly file")
+            mf = gr.File(label="Or upload a file of your own (CSV)", file_types=[".csv"])
+        mon_btn = gr.Button("Run the monitoring report", variant="primary")
+        m_msg = gr.Markdown()
+        m_tbl = gr.Dataframe(label="The groups: caught rate and false alarm rate, reference and this month (percent)", interactive=False, wrap=True)
+        m_dl = gr.File(label="Download the report (HTML)")
+        m_html = gr.HTML()
+        mon_btn.click(monitor, [m_choice, mf], [m_msg, m_tbl, m_html, m_dl])
+    with gr.Tab("4 · The repositories"):
+        gr.Markdown(repositories_markdown())
+
+    gr.Markdown(f"---\n*{CFG['system']} is a fictional model built on a public dataset for the AI Governance Practicum (AI Governance Collective). The organization, the people, and the program are fictional. Data: [{CFG['data_file']}]({CFG['data_url']}).*")
+
+if __name__ == "__main__":
+    demo.launch()

examples.csv

@@ -0,0 +1,4 @@
+case,case_note,ID,LIMIT_BAL,BILL_AMT1,BILL_AMT2,BILL_AMT3,BILL_AMT4,BILL_AMT5,BILL_AMT6,PAY_AMT1,PAY_AMT2,PAY_AMT3,PAY_AMT4,PAY_AMT5,PAY_AMT6,STATUS_0,STATUS_2,STATUS_3,STATUS_4,STATUS_5,STATUS_6,PAID_1,PAID_2,PAID_3,PAID_4,PAID_5,UTIL_RECENT,PAY_RATIO_RECENT,SEX,AGE,MARRIAGE_MODEL,EDUCATION_MODEL,default_payment_next_month,week6_rank
+caught,"caught (flagged, defaulted)",15501,50000.0,47237.0,50632.0,49546.0,49448.0,19887.0,38212.0,4500.0,0.0,1713.0,1003.0,699.0,800.0,delay of 2 months,delay of 2 months,delay of 2 months,"no delinquency recorded, code undefined","no delinquency recorded, code undefined","no delinquency recorded, code undefined",paid part,paid nothing,paid part,paid part,paid part,0.9447,0.0889,1,28,married,university,1,279
+missed,"missed (not flagged, defaulted; the first account below the cut)",9857,20000.0,18540.0,19726.0,8564.0,19775.0,19878.0,19898.0,2000.0,8568.0,19000.0,800.0,1000.0,20000.0,delay of 2 months,"no delinquency recorded, code undefined",paid duly,"no delinquency recorded, code undefined","no delinquency recorded, code undefined","no delinquency recorded, code undefined",paid part,paid in full,paid part,paid part,paid part,0.927,0.1014,2,28,single,graduate school,1,601
+false alarm,"false alarm (flagged, paid; the highest-scored paying account)",13697,30000.0,28859.0,28116.0,30226.0,29454.0,31775.0,30743.0,0.0,2900.0,0.0,2787.0,0.0,1445.0,delay of 2 months,delay of 2 months,delay of 2 months,delay of 2 months,delay of 2 months,delay of 2 months,paid nothing,paid part,paid nothing,paid part,paid nothing,0.962,0.0,2,49,single,university,0,1

limitwise_v2.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7842edc29acd316873ffa482db705ab2ea09427f8b4e1186436713d316de8ecb
+size 18636

limitwise_v2_recalibrated.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6508dfcc69ca108e3f08fc2be0843dab1d2517355970b8e622e52a34127ca37
+size 89015

reasons.csv

@@ -0,0 +1,94 @@
+input,value,reason
+LIMIT_BAL,(amount),the account's credit limit
+BILL_AMT1,(amount),the statement balance in the most recent month
+BILL_AMT2,(amount),the statement balance 1 month before the most recent
+BILL_AMT3,(amount),the statement balance 2 months before the most recent
+BILL_AMT4,(amount),the statement balance 3 months before the most recent
+BILL_AMT5,(amount),the statement balance 4 months before the most recent
+BILL_AMT6,(amount),the statement balance 5 months before the most recent
+PAY_AMT1,(amount),the payment made in the most recent month
+PAY_AMT2,(amount),the payment made 1 month before the most recent
+PAY_AMT3,(amount),the payment made 2 months before the most recent
+PAY_AMT4,(amount),the payment made 3 months before the most recent
+PAY_AMT5,(amount),the payment made 4 months before the most recent
+PAY_AMT6,(amount),the payment made 5 months before the most recent
+UTIL_RECENT,(amount),how much of the limit was in use on the most recent statement
+PAY_RATIO_RECENT,(amount),how much of the most recent statement balance was paid
+STATUS_0,delay of 1 month,payment was 1 month late in the most recent month
+STATUS_0,delay of 2 months,payment was 2 months late in the most recent month
+STATUS_0,delay of 3 months,payment was 3 months late in the most recent month
+STATUS_0,delay of 4 months,payment was 4 months late in the most recent month
+STATUS_0,delay of 5 months,payment was 5 months late in the most recent month
+STATUS_0,delay of 6 months,payment was 6 months late in the most recent month
+STATUS_0,delay of 7 months,payment was 7 months late in the most recent month
+STATUS_0,delay of 8 months,payment was 8 months late in the most recent month
+STATUS_0,"no delinquency recorded, code undefined",no delinquency recorded in the most recent month
+STATUS_0,paid duly,paid on time in the most recent month
+STATUS_2,delay of 1 month,payment was 1 month late in the month before the most recent
+STATUS_2,delay of 2 months,payment was 2 months late in the month before the most recent
+STATUS_2,delay of 3 months,payment was 3 months late in the month before the most recent
+STATUS_2,delay of 4 months,payment was 4 months late in the month before the most recent
+STATUS_2,delay of 5 months,payment was 5 months late in the month before the most recent
+STATUS_2,delay of 6 months,payment was 6 months late in the month before the most recent
+STATUS_2,delay of 7 months,payment was 7 months late in the month before the most recent
+STATUS_2,delay of 8 months,payment was 8 months late in the month before the most recent
+STATUS_2,"no delinquency recorded, code undefined",no delinquency recorded in the month before the most recent
+STATUS_2,paid duly,paid on time in the month before the most recent
+STATUS_3,delay of 1 month,payment was 1 month late in two months before the most recent
+STATUS_3,delay of 2 months,payment was 2 months late in two months before the most recent
+STATUS_3,delay of 3 months,payment was 3 months late in two months before the most recent
+STATUS_3,delay of 4 months,payment was 4 months late in two months before the most recent
+STATUS_3,delay of 5 months,payment was 5 months late in two months before the most recent
+STATUS_3,delay of 6 months,payment was 6 months late in two months before the most recent
+STATUS_3,delay of 7 months,payment was 7 months late in two months before the most recent
+STATUS_3,delay of 8 months,payment was 8 months late in two months before the most recent
+STATUS_3,"no delinquency recorded, code undefined",no delinquency recorded in two months before the most recent
+STATUS_3,paid duly,paid on time in two months before the most recent
+STATUS_4,delay of 1 month,payment was 1 month late in three months before the most recent
+STATUS_4,delay of 2 months,payment was 2 months late in three months before the most recent
+STATUS_4,delay of 3 months,payment was 3 months late in three months before the most recent
+STATUS_4,delay of 4 months,payment was 4 months late in three months before the most recent
+STATUS_4,delay of 5 months,payment was 5 months late in three months before the most recent
+STATUS_4,delay of 6 months,payment was 6 months late in three months before the most recent
+STATUS_4,delay of 7 months,payment was 7 months late in three months before the most recent
+STATUS_4,delay of 8 months,payment was 8 months late in three months before the most recent
+STATUS_4,"no delinquency recorded, code undefined",no delinquency recorded in three months before the most recent
+STATUS_4,paid duly,paid on time in three months before the most recent
+STATUS_5,delay of 2 months,payment was 2 months late in four months before the most recent
+STATUS_5,delay of 3 months,payment was 3 months late in four months before the most recent
+STATUS_5,delay of 4 months,payment was 4 months late in four months before the most recent
+STATUS_5,delay of 5 months,payment was 5 months late in four months before the most recent
+STATUS_5,delay of 6 months,payment was 6 months late in four months before the most recent
+STATUS_5,delay of 7 months,payment was 7 months late in four months before the most recent
+STATUS_5,delay of 8 months,payment was 8 months late in four months before the most recent
+STATUS_5,"no delinquency recorded, code undefined",no delinquency recorded in four months before the most recent
+STATUS_5,paid duly,paid on time in four months before the most recent
+STATUS_6,delay of 2 months,payment was 2 months late in five months before the most recent
+STATUS_6,delay of 3 months,payment was 3 months late in five months before the most recent
+STATUS_6,delay of 4 months,payment was 4 months late in five months before the most recent
+STATUS_6,delay of 5 months,payment was 5 months late in five months before the most recent
+STATUS_6,delay of 6 months,payment was 6 months late in five months before the most recent
+STATUS_6,delay of 7 months,payment was 7 months late in five months before the most recent
+STATUS_6,delay of 8 months,payment was 8 months late in five months before the most recent
+STATUS_6,"no delinquency recorded, code undefined",no delinquency recorded in five months before the most recent
+STATUS_6,paid duly,paid on time in five months before the most recent
+PAID_1,no balance,no balance was due in the most recent month
+PAID_1,paid in full,the balance was paid in full in the most recent month
+PAID_1,paid nothing,nothing was paid on the balance in the most recent month
+PAID_1,paid part,part of the balance was paid in the most recent month
+PAID_2,no balance,no balance was due in the month before the most recent
+PAID_2,paid in full,the balance was paid in full in the month before the most recent
+PAID_2,paid nothing,nothing was paid on the balance in the month before the most recent
+PAID_2,paid part,part of the balance was paid in the month before the most recent
+PAID_3,no balance,no balance was due in two months before the most recent
+PAID_3,paid in full,the balance was paid in full in two months before the most recent
+PAID_3,paid nothing,nothing was paid on the balance in two months before the most recent
+PAID_3,paid part,part of the balance was paid in two months before the most recent
+PAID_4,no balance,no balance was due in three months before the most recent
+PAID_4,paid in full,the balance was paid in full in three months before the most recent
+PAID_4,paid nothing,nothing was paid on the balance in three months before the most recent
+PAID_4,paid part,part of the balance was paid in three months before the most recent
+PAID_5,no balance,no balance was due in four months before the most recent
+PAID_5,paid in full,the balance was paid in full in four months before the most recent
+PAID_5,paid nothing,nothing was paid on the balance in four months before the most recent
+PAID_5,paid part,part of the balance was paid in four months before the most recent

reference.json

@@ -0,0 +1,739 @@
+{
+ "rows": 23996,
+ "test_rows": 5999,
+ "cut_score": 0.05812364061688247,
+ "share": 0.1,
+ "positive_rate": 0.2212035339223204,
+ "numeric": {
+  "LIMIT_BAL": {
+   "mean": 167270.87847974664,
+   "std": 129532.29976426934,
+   "median": 140000.0,
+   "edges": [
+    10000.0,
+    30000.0,
+    50000.0,
+    70000.0,
+    100000.0,
+    140000.0,
+    180000.0,
+    210000.0,
+    270000.0,
+    360000.0,
+    1000000.0
+   ],
+   "shares": {
+    "0": 0.08284714119019837,
+    "1": 0.06105184197366228,
+    "2": 0.13927321220203367,
+    "3": 0.09814135689281547,
+    "4": 0.102975495915986,
+    "5": 0.10380896816136023,
+    "6": 0.09159859976662778,
+    "7": 0.11485247541256877,
+    "8": 0.094599099849975,
+    "9": 0.11085180863477247
+   }
+  },
+  "BILL_AMT1": {
+   "mean": 51202.179904984165,
+   "std": 73978.79134449174,
+   "median": 22223.0,
+   "edges": [
+    -165580.0,
+    264.0,
+    1876.0,
+    5992.0,
+    13451.0,
+    22223.0,
+    36908.000000000015,
+    52061.5,
+    83381.0,
+    141741.0,
+    964511.0
+   ],
+   "shares": {
+    "0": 0.09943323887314552,
+    "1": 0.10055842640440073,
+    "2": 0.09997499583263877,
+    "3": 0.10001666944490749,
+    "4": 0.10001666944490749,
+    "5": 0.10001666944490749,
+    "6": 0.09997499583263877,
+    "7": 0.09997499583263877,
+    "8": 0.10001666944490749,
+    "9": 0.10001666944490749
+   }
+  },
+  "BILL_AMT2": {
+   "mean": 49259.96032672112,
+   "std": 71485.52373749421,
+   "median": 21080.5,
+   "edges": [
+    -69777.0,
+    0.0,
+    1479.0,
+    5518.500000000003,
+    12883.0,
+    21080.5,
+    34702.000000000015,
+    50711.0,
+    80199.0,
+    136949.5,
+    983931.0
+   ],
+   "shares": {
+    "0": 0.021711951991998667,
+    "1": 0.1782797132855476,
+    "2": 0.10001666944490749,
+    "3": 0.09997499583263877,
+    "4": 0.10001666944490749,
+    "5": 0.10001666944490749,
+    "6": 0.09997499583263877,
+    "7": 0.09997499583263877,
+    "8": 0.10001666944490749,
+    "9": 0.10001666944490749
+   }
+  },
+  "BILL_AMT3": {
+   "mean": 46981.74654109018,
+   "std": 68871.26278346537,
+   "median": 19982.5,
+   "edges": [
+    -61506.0,
+    0.0,
+    1189.0,
+    5176.000000000002,
+    12218.0,
+    19982.5,
+    31192.00000000002,
+    49175.5,
+    76593.0,
+    132127.5,
+    693131.0
+   ],
+   "shares": {
+    "0": 0.021378563093848973,
+    "1": 0.17861310218369728,
+    "2": 0.10001666944490749,
+    "3": 0.09997499583263877,
+    "4": 0.10001666944490749,
+    "5": 0.10001666944490749,
+    "6": 0.09997499583263877,
+    "7": 0.09997499583263877,
+    "8": 0.10001666944490749,
+    "9": 0.10001666944490749
+   }
+  },
+  "BILL_AMT4": {
+   "mean": 43342.34055675946,
+   "std": 64707.60457814837,
+   "median": 19007.5,
+   "edges": [
+    -81334.0,
+    0.0,
+    992.0,
+    4651.000000000002,
+    11179.0,
+    19007.5,
+    28481.000000000007,
+    45288.0,
+    70415.0,
+    123268.5,
+    891586.0
+   ],
+   "shares": {
+    "0": 0.021920320053342223,
+    "1": 0.17807134522420404,
+    "2": 0.10001666944490749,
+    "3": 0.09997499583263877,
+    "4": 0.10001666944490749,
+    "5": 0.10001666944490749,
+    "6": 0.09997499583263877,
+    "7": 0.09993332222037006,
+    "8": 0.1000583430571762,
+    "9": 0.10001666944490749
+   }
+  },
+  "BILL_AMT5": {
+   "mean": 40329.93294715786,
+   "std": 61075.619260422885,
+   "median": 18104.5,
+   "edges": [
+    -61372.0,
+    0.0,
+    763.0,
+    3589.500000000001,
+    9793.0,
+    18104.5,
+    26516.00000000001,
+    40631.5,
+    65494.0,
+    116593.0,
+    927171.0
+   ],
+   "shares": {
+    "0": 0.021628604767461243,
+    "1": 0.17836306051008502,
+    "2": 0.10001666944490749,
+    "3": 0.09997499583263877,
+    "4": 0.10001666944490749,
+    "5": 0.10001666944490749,
+    "6": 0.09997499583263877,
+    "7": 0.09997499583263877,
+    "8": 0.10001666944490749,
+    "9": 0.10001666944490749
+   }
+  },
+  "BILL_AMT6": {
+   "mean": 38869.986789464914,
+   "std": 59786.0983742078,
+   "median": 16979.0,
+   "edges": [
+    -339603.0,
+    0.0,
+    468.0,
+    2650.0,
+    8686.0,
+    16979.0,
+    25321.000000000015,
+    38824.5,
+    62658.0,
+    113650.5,
+    961664.0
+   ],
+   "shares": {
+    "0": 0.022628771461910318,
+    "1": 0.17732122020336724,
+    "2": 0.10001666944490749,
+    "3": 0.10001666944490749,
+    "4": 0.10001666944490749,
+    "5": 0.10001666944490749,
+    "6": 0.09997499583263877,
+    "7": 0.09997499583263877,
+    "8": 0.10001666944490749,
+    "9": 0.10001666944490749
+   }
+  },
+  "PAY_AMT1": {
+   "mean": 5679.0855975996,
+   "std": 16692.74539074982,
+   "median": 2120.0,
+   "edges": [
+    0.0,
+    326.0,
+    1270.500000000001,
+    1733.0,
+    2120.0,
+    3000.0,
+    4326.0,
+    6229.0,
+    10400.0,
+    873552.0
+   ],
+   "shares": {
+    "0": 0.19999166527754625,
+    "1": 0.10001666944490749,
+    "2": 0.09997499583263877,
+    "3": 0.09984997499583263,
+    "4": 0.07213702283713952,
+    "5": 0.12797966327721286,
+    "6": 0.10001666944490749,
+    "7": 0.09989164860810135,
+    "8": 0.10014169028171362
+   }
+  },
+  "PAY_AMT2": {
+   "mean": 5834.127937989665,
+   "std": 20435.617347531064,
+   "median": 2013.0,
+   "edges": [
+    0.0,
+    264.0,
+    1165.0,
+    1600.0,
+    2013.0,
+    3000.0,
+    4068.0,
+    6000.0,
+    10405.5,
+    1227082.0
+   ],
+   "shares": {
+    "0": 0.19969994999166527,
+    "1": 0.10026671111851976,
+    "2": 0.09664110685114186,
+    "3": 0.10330888481413569,
+    "4": 0.08884814135689281,
+    "5": 0.11118519753292215,
+    "6": 0.09201533588931489,
+    "7": 0.10801800300050009,
+    "8": 0.10001666944490749
+   }
+  },
+  "PAY_AMT3": {
+   "mean": 5218.560510085014,
+   "std": 17543.482408307405,
+   "median": 1804.0,
+   "edges": [
+    0.0,
+    2.0,
+    780.0,
+    1206.0,
+    1804.0,
+    2500.0,
+    3569.5,
+    5362.0,
+    10000.0,
+    896040.0
+   ],
+   "shares": {
+    "0": 0.1994082347057843,
+    "1": 0.09909984997499584,
+    "2": 0.10135022503750625,
+    "3": 0.10001666944490749,
+    "4": 0.09909984997499584,
+    "5": 0.10101683613935655,
+    "6": 0.09997499583263877,
+    "7": 0.0928071345224204,
+    "8": 0.10722620436739456
+   }
+  },
+  "PAY_AMT4": {
+   "mean": 4796.66569428238,
+   "std": 15067.01440928473,
+   "median": 1500.0,
+   "edges": [
+    0.0,
+    500.0,
+    1000.0,
+    1500.0,
+    2133.0,
+    3200.0,
+    5000.0,
+    9698.0,
+    621000.0
+   ],
+   "shares": {
+    "0": 0.29242373728954824,
+    "1": 0.0877229538256376,
+    "2": 0.10943490581763628,
+    "3": 0.11039339889981664,
+    "4": 0.09968328054675779,
+    "5": 0.07976329388231372,
+    "6": 0.12056176029338223,
+    "7": 0.10001666944490749
+   }
+  },
+  "PAY_AMT5": {
+   "mean": 4801.144690781797,
+   "std": 15152.516693963798,
+   "median": 1500.0,
+   "edges": [
+    0.0,
+    500.0,
+    1000.0,
+    1500.0,
+    2149.0,
+    3200.0,
+    5000.0,
+    9500.0,
+    426529.0
+   ],
+   "shares": {
+    "0": 0.29500750125020836,
+    "1": 0.08334722453742291,
+    "2": 0.1087681280213369,
+    "3": 0.1128104684114019,
+    "4": 0.09914152358726454,
+    "5": 0.07888814802467078,
+    "6": 0.12160360060010002,
+    "7": 0.1004334055675946
+   }
+  },
+  "PAY_AMT6": {
+   "mean": 5242.996249374896,
+   "std": 17640.841186151476,
+   "median": 1500.0,
+   "edges": [
+    0.0,
+    416.0,
+    1000.0,
+    1500.0,
+    2100.0,
+    3199.5,
+    5000.0,
+    9619.0,
+    528666.0
+   ],
+   "shares": {
+    "0": 0.29996666111018505,
+    "1": 0.09268211368561427,
+    "2": 0.10160026671111852,
+    "3": 0.10485080846807801,
+    "4": 0.10089181530255042,
+    "5": 0.08197199533255542,
+    "6": 0.11801966994499083,
+    "7": 0.10001666944490749
+   }
+  },
+  "UTIL_RECENT": {
+   "mean": 0.42326252708784795,
+   "std": 0.4107782042096859,
+   "median": 0.31445,
+   "edges": [
+    -0.6199,
+    0.0012,
+    0.0111,
+    0.0403,
+    0.1286,
+    0.31445,
+    0.5236,
+    0.7359,
+    0.9028,
+    0.98125,
+    6.4553
+   ],
+   "shares": {
+    "0": 0.09893315552592098,
+    "1": 0.10085014169028171,
+    "2": 0.09997499583263877,
+    "3": 0.10022503750625104,
+    "4": 0.10001666944490749,
+    "5": 0.09997499583263877,
+    "6": 0.09989164860810135,
+    "7": 0.1001000166694449,
+    "8": 0.10001666944490749,
+    "9": 0.10001666944490749
+   }
+  },
+  "PAY_RATIO_RECENT": {
+   "mean": 0.5656217642401378,
+   "std": 30.33630146650582,
+   "median": 0.0783,
+   "edges": [
+    0.0,
+    0.0281,
+    0.0392,
+    0.0458,
+    0.0544,
+    0.0783,
+    0.1288,
+    0.35830000000000056,
+    1.0,
+    1.0003,
+    4444.3333
+   ],
+   "shares": {
+    "0": 0.0997624703087886,
+    "1": 0.0991570024684458,
+    "2": 0.09943644916398864,
+    "3": 0.10101998043873131,
+    "4": 0.1005076614969028,
+    "5": 0.10008849145358856,
+    "6": 0.10004191700433142,
+    "7": 0.04494434353313772,
+    "8": 0.15453402263518234,
+    "9": 0.1005076614969028
+   }
+  },
+  "AGE": {
+   "mean": 35.437114519086514,
+   "std": 9.188789081976623,
+   "median": 34.0,
+   "edges": [
+    21.0,
+    25.0,
+    27.0,
+    29.0,
+    31.0,
+    34.0,
+    36.0,
+    40.0,
+    43.0,
+    49.0,
+    75.0
+   ],
+   "shares": {
+    "0": 0.0891398566427738,
+    "1": 0.08163860643440574,
+    "2": 0.09705784297382897,
+    "3": 0.10026671111851976,
+    "4": 0.11872812135355892,
+    "5": 0.0764294049008168,
+    "6": 0.1350641773628938,
+    "7": 0.08105517586264377,
+    "8": 0.11664444074012335,
+    "9": 0.10397566261043507
+   }
+  }
+ },
+ "categorical": {
+  "STATUS_0": {
+   "no delinquency recorded, code undefined": 0.5817219536589432,
+   "paid duly": 0.18944824137356225,
+   "delay of 1 month": 0.12347891315219203,
+   "delay of 2 months": 0.08997332888814802,
+   "delay of 3 months": 0.010585097516252709,
+   "delay of 4 months": 0.0025420903483913985,
+   "delay of 5 months": 0.0007501250208368062,
+   "delay of 8 months": 0.0007084514085680946,
+   "delay of 6 months": 0.0004167361226871145,
+   "delay of 7 months": 0.0003750625104184031
+  },
+  "STATUS_2": {
+   "no delinquency recorded, code undefined": 0.6493165527587932,
+   "paid duly": 0.20282547091181863,
+   "delay of 2 months": 0.1310218369728288,
+   "delay of 3 months": 0.010501750291715286,
+   "delay of 4 months": 0.0032505417569594934,
+   "delay of 1 month": 0.0009584930821803634,
+   "delay of 5 months": 0.0008751458576429405,
+   "delay of 7 months": 0.0007501250208368062,
+   "delay of 6 months": 0.00045840973495582597,
+   "delay of 8 months": 4.167361226871145e-05
+  },
+  "STATUS_3": {
+   "no delinquency recorded, code undefined": 0.6624437406234372,
+   "paid duly": 0.19769961660276714,
+   "delay of 2 months": 0.12693782297049508,
+   "delay of 3 months": 0.007709618269711618,
+   "delay of 4 months": 0.00258376396066011,
+   "delay of 7 months": 0.000833472245374229,
+   "delay of 6 months": 0.000833472245374229,
+   "delay of 5 months": 0.0007501250208368062,
+   "delay of 1 month": 0.00012502083680613436,
+   "delay of 8 months": 8.33472245374229e-05
+  },
+  "STATUS_4": {
+   "no delinquency recorded, code undefined": 0.6939073178863144,
+   "paid duly": 0.18898983163860644,
+   "delay of 2 months": 0.10547591265210869,
+   "delay of 3 months": 0.0058759793298883145,
+   "delay of 4 months": 0.002333722287047841,
+   "delay of 7 months": 0.0019169861643607267,
+   "delay of 5 months": 0.0012502083680613436,
+   "delay of 6 months": 0.00012502083680613436,
+   "delay of 1 month": 8.33472245374229e-05,
+   "delay of 8 months": 4.167361226871145e-05
+  },
+  "STATUS_5": {
+   "no delinquency recorded, code undefined": 0.7158276379396566,
+   "paid duly": 0.18494749124854143,
+   "delay of 2 months": 0.08780630105017503,
+   "delay of 3 months": 0.006001000166694449,
+   "delay of 4 months": 0.0027921320220036672,
+   "delay of 7 months": 0.0018753125520920154,
+   "delay of 5 months": 0.0005834305717619603,
+   "delay of 6 months": 0.00012502083680613436,
+   "delay of 8 months": 4.167361226871145e-05
+  },
+  "STATUS_6": {
+   "no delinquency recorded, code undefined": 0.7049508251375229,
+   "paid duly": 0.19153192198699784,
+   "delay of 2 months": 0.09309884980830138,
+   "delay of 3 months": 0.0061676946157692945,
+   "delay of 4 months": 0.001666944490748458,
+   "delay of 7 months": 0.0015835972662110352,
+   "delay of 6 months": 0.0005417569594932489,
+   "delay of 5 months": 0.0004167361226871145,
+   "delay of 8 months": 4.167361226871145e-05
+  },
+  "PAID_1": {
+   "paid part": 0.5919736622770462,
+   "paid in full": 0.2281630271711952,
+   "no balance": 0.10522587097849642,
+   "paid nothing": 0.07463743957326222
+  },
+  "PAID_2": {
+   "paid part": 0.5868894815802633,
+   "paid in full": 0.22703783963993998,
+   "no balance": 0.11781130188364727,
+   "paid nothing": 0.06826137689614936
+  },
+  "PAID_3": {
+   "paid part": 0.5809718286381064,
+   "paid in full": 0.21453575595932656,
+   "no balance": 0.1284380730121687,
+   "paid nothing": 0.0760543423903984
+  },
+  "PAID_4": {
+   "paid part": 0.570345057509585,
+   "paid in full": 0.20899316552758793,
+   "no balance": 0.13952325387564593,
+   "paid nothing": 0.0811385230871812
+  },
+  "PAID_5": {
+   "paid part": 0.5539673278879813,
+   "paid in full": 0.2150775129188198,
+   "no balance": 0.15781796966161027,
+   "paid nothing": 0.0731371895315886
+  },
+  "SEX": {
+   "2": 0.6050591765294215,
+   "1": 0.3949408234705784
+  },
+  "MARRIAGE_MODEL": {
+   "single": 0.5344224037339557,
+   "married": 0.45282547091181863,
+   "other": 0.010668444740790132,
+   "unknown": 0.0020836806134355724
+  },
+  "EDUCATION_MODEL": {
+   "university": 0.46970328388064675,
+   "graduate school": 0.3530171695282547,
+   "high school": 0.16136022670445074,
+   "unknown": 0.012002000333388899,
+   "other": 0.003917319553258876
+  }
+ },
+ "groups": {
+  "sex": {
+   "female": 14519,
+   "male": 9477
+  },
+  "age band": {
+   "25 to 34": 10466,
+   "35 to 44": 7186,
+   "45 to 54": 3382,
+   "under 25": 2139,
+   "55 to 64": 736,
+   "65 and over": 87
+  },
+  "marital status": {
+   "single": 12824,
+   "married": 10866,
+   "other": 256,
+   "unknown": 50
+  },
+  "education": {
+   "university": 11271,
+   "graduate school": 8471,
+   "high school": 3872,
+   "unknown": 288,
+   "other": 94
+  }
+ },
+ "portfolio": {
+  "false_alarm": 0.04195205479452055,
+  "miss": 0.6955538809344386,
+  "flagged": 600,
+  "caught": 404
+ },
+ "group_rates": {
+  "sex": {
+   "male": {
+    "n": 2407,
+    "false_alarm": 0.045921123716909776,
+    "miss": 0.7140287769784173,
+    "flagged_share": 0.10137100124636476
+   },
+   "female": {
+    "n": 3592,
+    "false_alarm": 0.03934774902516838,
+    "miss": 0.682230869001297,
+    "flagged_share": 0.09910913140311804
+   }
+  },
+  "age band": {
+   "35 to 44": {
+    "n": 1830,
+    "false_alarm": 0.04073033707865169,
+    "miss": 0.6945812807881774,
+    "flagged_share": 0.0994535519125683
+   },
+   "25 to 34": {
+    "n": 2543,
+    "false_alarm": 0.03844258255298177,
+    "miss": 0.7315175097276264,
+    "flagged_share": 0.0849390483680692
+   },
+   "under 25": {
+    "n": 545,
+    "false_alarm": 0.059113300492610835,
+    "miss": 0.5683453237410072,
+    "flagged_share": 0.15412844036697249
+   },
+   "45 to 54": {
+    "n": 851,
+    "false_alarm": 0.04186046511627907,
+    "miss": 0.7184466019417476,
+    "flagged_share": 0.099882491186839
+   },
+   "55 to 64": {
+    "n": 207,
+    "false_alarm": 0.059602649006622516,
+    "miss": 0.5892857142857143,
+    "flagged_share": 0.15458937198067632
+   },
+   "65 and over": {
+    "n": 23,
+    "false_alarm": 0.0,
+    "miss": 0.8333333333333334,
+    "flagged_share": 0.043478260869565216
+   }
+  },
+  "marital status": {
+   "single": {
+    "n": 3138,
+    "false_alarm": 0.044,
+    "miss": 0.6943573667711599,
+    "flagged_share": 0.09719566602931803
+   },
+   "married": {
+    "n": 2790,
+    "false_alarm": 0.03777148253068933,
+    "miss": 0.6964285714285714,
+    "flagged_share": 0.1017921146953405
+   },
+   "other": {
+    "n": 67,
+    "false_alarm": 0.08,
+    "miss": 0.7058823529411765,
+    "flagged_share": 0.13432835820895522
+   },
+   "unknown": {
+    "n": 4,
+    "false_alarm": 0.5,
+    "miss": 0.0,
+    "flagged_share": 0.5
+   }
+  },
+  "education": {
+   "university": {
+    "n": 2756,
+    "false_alarm": 0.045933014354066985,
+    "miss": 0.6801801801801802,
+    "flagged_share": 0.11211901306240929
+   },
+   "graduate school": {
+    "n": 2113,
+    "false_alarm": 0.030232558139534883,
+    "miss": 0.712468193384224,
+    "flagged_share": 0.07808802650260294
+   },
+   "high school": {
+    "n": 1044,
+    "false_alarm": 0.056338028169014086,
+    "miss": 0.7110266159695817,
+    "flagged_share": 0.11494252873563218
+   },
+   "unknown": {
+    "n": 57,
+    "false_alarm": 0.07407407407407407,
+    "miss": 0.6666666666666666,
+    "flagged_share": 0.08771929824561403
+   },
+   "other": {
+    "n": 29,
+    "false_alarm": 0.0,
+    "miss": 0.5,
+    "flagged_share": 0.034482758620689655
+   }
+  }
+ },
+ "cut_score_test_set": 0.13423639065119367,
+ "reference_rows": 19266,
+ "reference_positive_rate": 0.03000103809820409,
+ "reference_note": "the Week 6 test set reweighted to the 3 percent book rate (each paying account four times, 578 defaulters once); the cut is the top 10 percent of this file"
+}

requirements.txt

@@ -0,0 +1,8 @@
+gradio==6.31.0
+scikit-learn==1.9.1
+fairlearn==0.15.0
+pandas>=2.2,<3.1
+numpy>=1.26
+scipy>=1.11
+matplotlib>=3.8
+evidently==0.7.23

space_config.json

@@ -0,0 +1,733 @@
+{
+ "org": "Corvana Bank",
+ "system": "LimitWise",
+ "version": "v2",
+ "short": "limitwise",
+ "emoji": "\ud83d\udcb3",
+ "portfolio": "corvana-portfolio",
+ "collection_name": "LimitWise Governance Repository",
+ "collection_url": "https://huggingface.co/collections/drdavidprivacy/limitwise-governance-repository-6ae5ff9b374cab20b18250eb",
+ "model_file": "limitwise_v2.pkl",
+ "recal_file": "limitwise_v2_recalibrated.pkl",
+ "odds_factor": 0.1089,
+ "reasons_file": "reasons.csv",
+ "data_file": "corvana_modeling_extract_remediated.csv",
+ "md5": "f79f13a5119f78572e9f6cc3243ea6ae",
+ "id_col": "ID",
+ "outcome": "default_payment_next_month",
+ "positive": 1,
+ "positive_label": "default",
+ "negative_label": "paid",
+ "unit": "account",
+ "units": "accounts",
+ "action": "an outreach call or a hardship offer, reviewed by a person",
+ "flagged_word": "on the outreach queue",
+ "person": "cardholder",
+ "share": 0.1,
+ "book_size": 4600000,
+ "capacity": 460000,
+ "sample_rows": 4500,
+ "month_rows": 4000,
+ "month_seed": 11,
+ "groups": {
+  "sex": {
+   "col": "SEX",
+   "map": {
+    "1": "male",
+    "2": "female"
+   }
+  },
+  "age band": {
+   "col": "AGE",
+   "bands": [
+    [
+     0,
+     24,
+     "under 25"
+    ],
+    [
+     25,
+     34,
+     "25 to 34"
+    ],
+    [
+     35,
+     44,
+     "35 to 44"
+    ],
+    [
+     45,
+     54,
+     "45 to 54"
+    ],
+    [
+     55,
+     64,
+     "55 to 64"
+    ],
+    [
+     65,
+     200,
+     "65 and over"
+    ]
+   ]
+  },
+  "marital status": {
+   "col": "MARRIAGE_MODEL"
+  },
+  "education": {
+   "col": "EDUCATION_MODEL"
+  }
+ },
+ "min_group": 1000,
+ "group_check": "false_alarm",
+ "group_tripwire": 0.2,
+ "watch_groups": [
+  "under 25",
+  "high school"
+ ],
+ "calibration": {
+  "bands": [
+   [
+    0,
+    0.01,
+    "under 1 percent"
+   ],
+   [
+    0.01,
+    0.03,
+    "1 to 3 percent"
+   ],
+   [
+    0.03,
+    0.08,
+    "3 to 8 percent"
+   ],
+   [
+    0.08,
+    0.2,
+    "8 to 20 percent"
+   ],
+   [
+    0.2,
+    1.0001,
+    "over 20 percent"
+   ]
+  ],
+  "tripwire": 0.25,
+  "book_rate": 0.03
+ },
+ "watch_fields": [
+  "STATUS_0",
+  "STATUS_2",
+  "STATUS_3",
+  "STATUS_4",
+  "STATUS_5",
+  "STATUS_6"
+ ],
+ "watch_fields_label": "the six status fields (Week 6 memo, ruling 6)",
+ "week6": {
+  "roc_auc": 0.758,
+  "flagged": 600,
+  "caught": 402,
+  "false_alarms": 198,
+  "precision": 0.67,
+  "recall": 0.303,
+  "test_rows": 5999,
+  "test_positives": 1327
+ },
+ "lead": "Tom Becker, AI Governance Lead",
+ "caio": "Anika Rao, Chief AI Officer",
+ "memo_ref": "Week 6 memo",
+ "monitoring_lines": [
+  "The calibration table by score band, predicted against observed default, every band within 25 percent (ruling 1).",
+  "The six status fields' distributions against the training file (ruling 6).",
+  "False alarm and contact rates by group, with under 25 and high school reviewed monthly by the AI Governance Office and Customer Advocacy (rulings 3 and 4)."
+ ],
+ "inputs": [
+  "LIMIT_BAL",
+  "BILL_AMT1",
+  "BILL_AMT2",
+  "BILL_AMT3",
+  "BILL_AMT4",
+  "BILL_AMT5",
+  "BILL_AMT6",
+  "PAY_AMT1",
+  "PAY_AMT2",
+  "PAY_AMT3",
+  "PAY_AMT4",
+  "PAY_AMT5",
+  "PAY_AMT6",
+  "STATUS_0",
+  "STATUS_2",
+  "STATUS_3",
+  "STATUS_4",
+  "STATUS_5",
+  "STATUS_6",
+  "PAID_1",
+  "PAID_2",
+  "PAID_3",
+  "PAID_4",
+  "PAID_5",
+  "UTIL_RECENT",
+  "PAY_RATIO_RECENT",
+  "SEX",
+  "AGE",
+  "MARRIAGE_MODEL",
+  "EDUCATION_MODEL"
+ ],
+ "categorical": [
+  "STATUS_0",
+  "STATUS_2",
+  "STATUS_3",
+  "STATUS_4",
+  "STATUS_5",
+  "STATUS_6",
+  "PAID_1",
+  "PAID_2",
+  "PAID_3",
+  "PAID_4",
+  "PAID_5",
+  "SEX",
+  "MARRIAGE_MODEL",
+  "EDUCATION_MODEL"
+ ],
+ "numeric": [
+  "LIMIT_BAL",
+  "BILL_AMT1",
+  "BILL_AMT2",
+  "BILL_AMT3",
+  "BILL_AMT4",
+  "BILL_AMT5",
+  "BILL_AMT6",
+  "PAY_AMT1",
+  "PAY_AMT2",
+  "PAY_AMT3",
+  "PAY_AMT4",
+  "PAY_AMT5",
+  "PAY_AMT6",
+  "UTIL_RECENT",
+  "PAY_RATIO_RECENT",
+  "AGE"
+ ],
+ "kept": [
+  "LIMIT_BAL",
+  "BILL_AMT1",
+  "BILL_AMT2",
+  "BILL_AMT3",
+  "BILL_AMT4",
+  "BILL_AMT5",
+  "BILL_AMT6",
+  "PAY_AMT1",
+  "PAY_AMT2",
+  "PAY_AMT3",
+  "PAY_AMT4",
+  "PAY_AMT5",
+  "PAY_AMT6",
+  "UTIL_RECENT",
+  "PAY_RATIO_RECENT",
+  "STATUS_0_delay of 1 month",
+  "STATUS_0_delay of 2 months",
+  "STATUS_0_delay of 3 months",
+  "STATUS_0_delay of 4 months",
+  "STATUS_0_delay of 5 months",
+  "STATUS_0_delay of 6 months",
+  "STATUS_0_delay of 7 months",
+  "STATUS_0_delay of 8 months",
+  "STATUS_0_no delinquency recorded, code undefined",
+  "STATUS_0_paid duly",
+  "STATUS_2_delay of 1 month",
+  "STATUS_2_delay of 2 months",
+  "STATUS_2_delay of 3 months",
+  "STATUS_2_delay of 4 months",
+  "STATUS_2_delay of 5 months",
+  "STATUS_2_delay of 6 months",
+  "STATUS_2_delay of 7 months",
+  "STATUS_2_delay of 8 months",
+  "STATUS_2_no delinquency recorded, code undefined",
+  "STATUS_2_paid duly",
+  "STATUS_3_delay of 1 month",
+  "STATUS_3_delay of 2 months",
+  "STATUS_3_delay of 3 months",
+  "STATUS_3_delay of 4 months",
+  "STATUS_3_delay of 5 months",
+  "STATUS_3_delay of 6 months",
+  "STATUS_3_delay of 7 months",
+  "STATUS_3_delay of 8 months",
+  "STATUS_3_no delinquency recorded, code undefined",
+  "STATUS_3_paid duly",
+  "STATUS_4_delay of 1 month",
+  "STATUS_4_delay of 2 months",
+  "STATUS_4_delay of 3 months",
+  "STATUS_4_delay of 4 months",
+  "STATUS_4_delay of 5 months",
+  "STATUS_4_delay of 6 months",
+  "STATUS_4_delay of 7 months",
+  "STATUS_4_delay of 8 months",
+  "STATUS_4_no delinquency recorded, code undefined",
+  "STATUS_4_paid duly",
+  "STATUS_5_delay of 2 months",
+  "STATUS_5_delay of 3 months",
+  "STATUS_5_delay of 4 months",
+  "STATUS_5_delay of 5 months",
+  "STATUS_5_delay of 6 months",
+  "STATUS_5_delay of 7 months",
+  "STATUS_5_delay of 8 months",
+  "STATUS_5_no delinquency recorded, code undefined",
+  "STATUS_5_paid duly",
+  "STATUS_6_delay of 2 months",
+  "STATUS_6_delay of 3 months",
+  "STATUS_6_delay of 4 months",
+  "STATUS_6_delay of 5 months",
+  "STATUS_6_delay of 6 months",
+  "STATUS_6_delay of 7 months",
+  "STATUS_6_delay of 8 months",
+  "STATUS_6_no delinquency recorded, code undefined",
+  "STATUS_6_paid duly",
+  "PAID_1_no balance",
+  "PAID_1_paid in full",
+  "PAID_1_paid nothing",
+  "PAID_1_paid part",
+  "PAID_2_no balance",
+  "PAID_2_paid in full",
+  "PAID_2_paid nothing",
+  "PAID_2_paid part",
+  "PAID_3_no balance",
+  "PAID_3_paid in full",
+  "PAID_3_paid nothing",
+  "PAID_3_paid part",
+  "PAID_4_no balance",
+  "PAID_4_paid in full",
+  "PAID_4_paid nothing",
+  "PAID_4_paid part",
+  "PAID_5_no balance",
+  "PAID_5_paid in full",
+  "PAID_5_paid nothing",
+  "PAID_5_paid part"
+ ],
+ "artifacts": {
+  "limitwise_v2.pkl": {
+   "bytes": 18636,
+   "sha256": "7842edc29acd316873ffa482db705ab2ea09427f8b4e1186436713d316de8ecb"
+  },
+  "limitwise_v2_recalibrated.pkl": {
+   "bytes": 89015,
+   "sha256": "b6508dfcc69ca108e3f08fc2be0843dab1d2517355970b8e622e52a34127ca37"
+  },
+  "reasons.csv": {
+   "bytes": 7799,
+   "sha256": "ab349230ee6c1a086e4a6f6b29d93e8dec6c58f084ccc27746f28def712998d2"
+  }
+ },
+ "weights": [
+  [
+   "STATUS_2_delay of 8 months",
+   -0.5039640072953564
+  ],
+  [
+   "STATUS_0_delay of 2 months",
+   0.4551672900610481
+  ],
+  [
+   "STATUS_0_no delinquency recorded, code undefined",
+   -0.29541722019909206
+  ],
+  [
+   "BILL_AMT1",
+   -0.22289125235966772
+  ],
+  [
+   "BILL_AMT2",
+   0.22269939355880997
+  ],
+  [
+   "LIMIT_BAL",
+   -0.18063967864614505
+  ],
+  [
+   "STATUS_3_delay of 6 months",
+   0.17892881317393983
+  ],
+  [
+   "STATUS_3_delay of 1 month",
+   -0.1666802133117096
+  ],
+  [
+   "PAY_AMT1",
+   -0.16324871972903526
+  ],
+  [
+   "STATUS_5_delay of 7 months",
+   0.15235838602958815
+  ],
+  [
+   "STATUS_4_delay of 1 month",
+   0.14913252735793928
+  ],
+  [
+   "STATUS_4_delay of 7 months",
+   -0.14666812791171707
+  ],
+  [
+   "STATUS_0_delay of 3 months",
+   0.1356422562245205
+  ],
+  [
+   "STATUS_4_delay of 5 months",
+   -0.13392612402039197
+  ],
+  [
+   "UTIL_RECENT",
+   0.13305523853156798
+  ],
+  [
+   "PAY_AMT2",
+   -0.13110437667603564
+  ],
+  [
+   "STATUS_4_delay of 6 months",
+   -0.12287955895658455
+  ],
+  [
+   "PAID_3_paid nothing",
+   0.11291845600022514
+  ],
+  [
+   "PAID_4_paid part",
+   -0.10724778099203863
+  ],
+  [
+   "BILL_AMT3",
+   0.10585801079767654
+  ],
+  [
+   "PAID_4_no balance",
+   0.10148095550974805
+  ],
+  [
+   "BILL_AMT6",
+   -0.09763006096339935
+  ],
+  [
+   "PAID_1_paid part",
+   -0.09056534286569744
+  ],
+  [
+   "PAID_5_paid in full",
+   -0.08923148289651153
+  ],
+  [
+   "STATUS_3_delay of 2 months",
+   0.08860161441479777
+  ],
+  [
+   "STATUS_6_no delinquency recorded, code undefined",
+   -0.080771470924185
+  ],
+  [
+   "PAID_3_paid in full",
+   -0.07626009426053693
+  ],
+  [
+   "PAID_1_paid nothing",
+   0.07073519837099726
+  ],
+  [
+   "BILL_AMT4",
+   0.0677806743545396
+  ],
+  [
+   "STATUS_5_paid duly",
+   -0.06666764266580824
+  ],
+  [
+   "STATUS_5_delay of 2 months",
+   0.06417630050812508
+  ],
+  [
+   "STATUS_3_paid duly",
+   -0.05855938417367776
+  ],
+  [
+   "PAID_5_paid nothing",
+   0.05660922499989594
+  ],
+  [
+   "STATUS_2_paid duly",
+   -0.05575862688354833
+  ],
+  [
+   "STATUS_6_paid duly",
+   0.051982154815802435
+  ],
+  [
+   "PAID_1_no balance",
+   0.0514916828905529
+  ],
+  [
+   "STATUS_5_delay of 6 months",
+   0.051270206053641826
+  ],
+  [
+   "PAID_2_paid nothing",
+   0.04988887717888915
+  ],
+  [
+   "STATUS_4_delay of 2 months",
+   0.04986130866939965
+  ],
+  [
+   "STATUS_6_delay of 3 months",
+   0.048227639483102186
+  ],
+  [
+   "STATUS_2_delay of 7 months",
+   -0.046987798245092256
+  ],
+  [
+   "STATUS_0_delay of 1 month",
+   0.04600743034304016
+  ],
+  [
+   "STATUS_0_delay of 4 months",
+   0.04594619106710718
+  ],
+  [
+   "PAID_2_paid part",
+   -0.04515798904117099
+  ],
+  [
+   "STATUS_6_delay of 2 months",
+   0.04299847088182787
+  ],
+  [
+   "STATUS_2_no delinquency recorded, code undefined",
+   0.04256803115041107
+  ],
+  [
+   "STATUS_2_delay of 4 months",
+   -0.04245810777257909
+  ],
+  [
+   "PAID_5_paid part",
+   0.04132424452731096
+  ],
+  [
+   "STATUS_0_paid duly",
+   -0.04040908259405443
+  ],
+  [
+   "STATUS_5_delay of 5 months",
+   0.04029083926098087
+  ],
+  [
+   "STATUS_0_delay of 8 months",
+   -0.03989328482470391
+  ],
+  [
+   "PAID_4_paid nothing",
+   0.030239839152150578
+  ],
+  [
+   "PAID_2_no balance",
+   0.029314204273452448
+  ],
+  [
+   "STATUS_0_delay of 7 months",
+   0.027682360024328365
+  ],
+  [
+   "BILL_AMT5",
+   0.02610742724500418
+  ],
+  [
+   "STATUS_2_delay of 2 months",
+   0.025343683133782442
+  ],
+  [
+   "STATUS_6_delay of 7 months",
+   -0.025146023500980816
+  ],
+  [
+   "STATUS_4_no delinquency recorded, code undefined",
+   -0.02296436281798626
+  ],
+  [
+   "PAID_1_paid in full",
+   0.022724009839817483
+  ],
+  [
+   "PAID_4_paid in full",
+   0.02245439367494789
+  ],
+  [
+   "STATUS_3_no delinquency recorded, code undefined",
+   -0.021565816950964413
+  ],
+  [
+   "STATUS_4_paid duly",
+   0.020762235068190773
+  ],
+  [
+   "STATUS_6_delay of 4 months",
+   0.019296387292338728
+  ],
+  [
+   "STATUS_4_delay of 3 months",
+   -0.019237043787947525
+  ],
+  [
+   "STATUS_2_delay of 1 month",
+   -0.01721996140750323
+  ],
+  [
+   "STATUS_5_delay of 4 months",
+   -0.016779851753297246
+  ],
+  [
+   "STATUS_3_delay of 3 months",
+   0.016488680440516868
+  ],
+  [
+   "STATUS_2_delay of 5 months",
+   0.014957419929230715
+  ],
+  [
+   "PAY_RATIO_RECENT",
+   -0.014744579451479636
+  ],
+  [
+   "STATUS_6_delay of 6 months",
+   0.013758271323696863
+  ],
+  [
+   "PAID_3_no balance",
+   -0.013308180673404297
+  ],
+  [
+   "PAY_AMT4",
+   -0.011886073604529012
+  ],
+  [
+   "STATUS_3_delay of 4 months",
+   -0.011652950713743126
+  ],
+  [
+   "PAID_3_paid part",
+   0.011198663075826232
+  ],
+  [
+   "STATUS_3_delay of 5 months",
+   -0.010570088837797562
+  ],
+  [
+   "PAY_AMT6",
+   -0.010199223770709179
+  ],
+  [
+   "PAY_AMT5",
+   -0.010192317105689951
+  ],
+  [
+   "STATUS_5_delay of 3 months",
+   -0.008719319679223644
+  ],
+  [
+   "STATUS_2_delay of 6 months",
+   -0.0066645897593760675
+  ],
+  [
+   "PAY_AMT3",
+   -0.005409485111934878
+  ],
+  [
+   "STATUS_2_delay of 3 months",
+   0.0053150067807837854
+  ],
+  [
+   "STATUS_4_delay of 8 months",
+   0.005105130451058765
+  ],
+  [
+   "STATUS_5_delay of 8 months",
+   0.005105130451058765
+  ],
+  [
+   "STATUS_6_delay of 8 months",
+   0.005105130451058765
+  ],
+  [
+   "STATUS_4_delay of 4 months",
+   -0.003551067523748044
+  ],
+  [
+   "PAID_5_no balance",
+   0.003222446885544461
+  ],
+  [
+   "STATUS_0_delay of 5 months",
+   0.003219742087970856
+  ],
+  [
+   "STATUS_6_delay of 5 months",
+   0.002269724948883509
+  ],
+  [
+   "STATUS_5_no delinquency recorded, code undefined",
+   0.002162756013280376
+  ],
+  [
+   "STATUS_3_delay of 7 months",
+   -0.0006486499754519148
+  ],
+  [
+   "PAID_2_paid in full",
+   -0.00031853069544259726
+  ],
+  [
+   "STATUS_0_delay of 6 months",
+   -0.0003018817489643535
+  ],
+  [
+   "STATUS_3_delay of 8 months",
+   -0.0001632838231806132
+  ]
+ ],
+ "intercept": -0.2309387424099975,
+ "data_url": "https://huggingface.co/datasets/drdavidprivacy/practicum-case-packs/resolve/main/corvana_modeling_extract_remediated.csv",
+ "dataset_repo": "https://huggingface.co/datasets/drdavidprivacy/practicum-case-packs",
+ "versions": {
+  "sklearn": "1.9.1",
+  "fairlearn": "0.15.0",
+  "pandas": "3.0.5",
+  "numpy": "2.5.3"
+ },
+ "docs": [
+  [
+   "01_Use-Case-Assessment.pdf",
+   "Use Case Assessment (Week 1)"
+  ],
+  [
+   "02_EDA-Observation-Log.pdf",
+   "EDA Observation Log (Week 2)"
+  ],
+  [
+   "03_Dataset-Datasheet.pdf",
+   "Dataset Datasheet (Week 3)"
+  ],
+  [
+   "04_Model-Selection-Rationale.pdf",
+   "Model Selection Rationale (Week 4)"
+  ],
+  [
+   "05_Model-Card.pdf",
+   "Model Card (Week 6)"
+  ],
+  [
+   "06_Go-No-Go-Decision-Record.pdf",
+   "Go/No-Go Decision Record (Week 7)"
+  ]
+ ],
+ "reference_label": "the Week 6 test set reweighted to the 3 percent book rate",
+ "reference_note": "The reference file is the Week 6 test set reweighted to the bank's book rate, the way the Week 6 calibration check reweighted it: every paying account counted four times and 578 defaulters once, so the file defaults at 3.0 percent, as the book does. The Week 6 cut is the score of the last account inside the top 10 percent of that file, which is the queue the record approved."
+}