diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 66c0587..7807b05 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -15,7 +15,7 @@ repos:
 
       - id: ruff
         name: ruff (unused imports, undefined names, redefinitions)
-        entry: ruff check faircode scripts tests
+        entry: ruff check faircode scripts tests explainers/code
         language: system
         pass_filenames: false
         always_run: true
diff --git a/Makefile b/Makefile
index e36e5f4..9307591 100644
--- a/Makefile
+++ b/Makefile
@@ -56,6 +56,7 @@ lint:  ## Enforce the em-dash-free rule + check for broken doc links + issue num
 	$(PY) scripts/check_explainer_structure.py
 	$(PY) scripts/check_traction.py
 	$(PY) scripts/check_contributors.py
+  ruff check faircode scripts tests explainers/code
 	ruff check faircode scripts tests
 
 audits:  ## Run every audit's unfair.py/fair.py + the CLI smoke tests (mirrors audits.yml's required run-audits job, plus profiler/benchmark-harness's CLI steps)
diff --git a/explainers/aggregation-bias.md b/explainers/aggregation-bias.md
index 8ab9912..eb4d071 100644
--- a/explainers/aggregation-bias.md
+++ b/explainers/aggregation-bias.md
@@ -44,7 +44,6 @@ Two cautions. This model has only three features, so its absolute accuracy is lo
 Run from the repo root (`PYTHONPATH=. python3 explainers/code/aggregation-bias.py`; that file is generated from the block below).
 
 ```python
-import numpy as np
 from sklearn.linear_model import LogisticRegression
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness
 
diff --git a/explainers/code/aggregation-bias.py b/explainers/code/aggregation-bias.py
index 860ff4f..f4c1797 100644
--- a/explainers/code/aggregation-bias.py
+++ b/explainers/code/aggregation-bias.py
@@ -1,5 +1,4 @@
 # Generated from explainers/aggregation-bias.md by scripts/build_explainers.py. Edit the markdown, not this file.
-import numpy as np
 from sklearn.linear_model import LogisticRegression
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness
 
diff --git a/explainers/code/delayed-feedback.py b/explainers/code/delayed-feedback.py
index 9ffff9a..6299d33 100644
--- a/explainers/code/delayed-feedback.py
+++ b/explainers/code/delayed-feedback.py
@@ -18,7 +18,6 @@ def monitored_gap(mean_delay_group, mean_delay_others, today=90):
     delay = np.minimum(delay, 29.9)
     seen_positive = (y == 1) & (discharge + delay <= today)
     window_closed = discharge + 30 <= today
-    known_negative = (y == 0) & window_closed
     # a naive monitor counts every discharged patient and treats every outcome it has not seen as "no"
     naive = {g: seen_positive[group == g].mean() for g in (True, False)}
     # a careful monitor only counts patients whose 30-day window is closed
diff --git a/explainers/code/fairness-in-ranking.py b/explainers/code/fairness-in-ranking.py
index 92a0b89..50a8dc9 100644
--- a/explainers/code/fairness-in-ranking.py
+++ b/explainers/code/fairness-in-ranking.py
@@ -38,7 +38,7 @@ def report(score, name):
     print(f"{name}\n  female share of the top {K}: {female[top].mean():.3f} (pool {female.mean():.3f})"
           f"\n  exposure ratio female/male, whole list: {whole[female == 1].mean() / whole[female == 0].mean():.3f}"
           f"\n  exposure ratio female/male, top {K} only: {cut[female == 1].mean() / cut[female == 0].mean():.3f}"
-          f"\n  exposure per unit of hire rate, female/male: "
+          "\n  exposure per unit of hire rate, female/male: "
           f"{(cut[female == 1].mean() / hired[female == 1].mean()) / (cut[female == 0].mean() / hired[female == 0].mean()):.3f}")
 
 report(scores(X_biased), "baseline (Gender is a feature)")
diff --git a/explainers/code/feature-importance-by-group.py b/explainers/code/feature-importance-by-group.py
index 3cc3819..73ea037 100644
--- a/explainers/code/feature-importance-by-group.py
+++ b/explainers/code/feature-importance-by-group.py
@@ -1,5 +1,4 @@
 # Generated from explainers/feature-importance-by-group.md by scripts/build_explainers.py. Edit the markdown, not this file.
-import numpy as np
 from sklearn.inspection import permutation_importance
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness
 from faircode.models import build_model
diff --git a/explainers/code/hiring-pipeline-adverse-impact.py b/explainers/code/hiring-pipeline-adverse-impact.py
index c6bdeb1..0fb0f03 100644
--- a/explainers/code/hiring-pipeline-adverse-impact.py
+++ b/explainers/code/hiring-pipeline-adverse-impact.py
@@ -16,7 +16,6 @@
 def ratio_with_interval(passed, reached, n_boot=500):
     """Women's selection rate divided by everyone else's, among the people who reached the stage, with a bootstrap interval."""
     w, m = reached & (women == 1), reached & (women == 0)
-    rate = lambda mask, rows: passed[rows][mask[rows]].mean()
     point = passed[w].mean() / passed[m].mean()
     idx_w, idx_m = np.flatnonzero(w), np.flatnonzero(m)
     boots = [passed[rng.choice(idx_w, len(idx_w))].mean() / passed[rng.choice(idx_m, len(idx_m))].mean() for _ in range(n_boot)]
diff --git a/explainers/code/leave-one-group-out-evaluation.py b/explainers/code/leave-one-group-out-evaluation.py
index 7103f0b..6b58ed8 100644
--- a/explainers/code/leave-one-group-out-evaluation.py
+++ b/explainers/code/leave-one-group-out-evaluation.py
@@ -1,5 +1,4 @@
 # Generated from explainers/leave-one-group-out-evaluation.md by scripts/build_explainers.py. Edit the markdown, not this file.
-import numpy as np
 import pandas as pd
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness
 from faircode.models import build_model
diff --git a/explainers/code/monitoring-fairness-in-production.py b/explainers/code/monitoring-fairness-in-production.py
index 6f28431..8117d42 100644
--- a/explainers/code/monitoring-fairness-in-production.py
+++ b/explainers/code/monitoring-fairness-in-production.py
@@ -37,7 +37,7 @@ def batches_until_alert(batch, window, drift, drift_at=20, n_batches=60, z=3):
             return b - drift_at + 1 if b >= drift_at else -1       # -1 = a false alarm before the drift
     return None
 
-print(f"\nalert rule: the gap over the last W batches is more than 3 standard errors from the launch value")
+print("\nalert rule: the gap over the last W batches is more than 3 standard errors from the launch value")
 print(f"{'batch size':>10} {'window':>7} {'drift':>6} {'median batches to alert':>25} {'never alerted':>14} {'false alarms before the drift':>31}")
 for batch, window in ((200, 3), (500, 3), (1000, 3), (2000, 3), (1000, 1), (1000, 6)):
     for drift in (0.03, 0.06):
diff --git a/explainers/code/small-group-shrinkage.py b/explainers/code/small-group-shrinkage.py
index 2b51af1..11064e4 100644
--- a/explainers/code/small-group-shrinkage.py
+++ b/explainers/code/small-group-shrinkage.py
@@ -49,7 +49,7 @@ def shrink(successes, counts):
     small_raw.append(np.sqrt(np.mean((raw[small] - truth[small]) ** 2))); small_shrunk.append(np.sqrt(np.mean((shrunk[small] - truth[small]) ** 2)))
     estimates.append((raw[worst], shrunk[worst]))
 
-print(f"\nerror (RMSE against the true cell accuracy) from a 600-row audit, averaged over 300 audits")
+print("\nerror (RMSE against the true cell accuracy) from a 600-row audit, averaged over 300 audits")
 print(f"  all {len(cells)} cells:        raw {np.mean(raw_err):.3f} | shrunk {np.mean(shrunk_err):.3f}")
 print(f"  the smallest {small.sum()} cells:   raw {np.mean(small_raw):.3f} | shrunk {np.mean(small_shrunk):.3f}")
 mean_raw, mean_shrunk = np.mean(estimates, axis=0)
diff --git a/explainers/code/survivorship-bias.py b/explainers/code/survivorship-bias.py
index ae27715..1cca3f0 100644
--- a/explainers/code/survivorship-bias.py
+++ b/explainers/code/survivorship-bias.py
@@ -1,5 +1,4 @@
 # Generated from explainers/survivorship-bias.md by scripts/build_explainers.py. Edit the markdown, not this file.
-import numpy as np
 from sklearn.linear_model import LogisticRegression
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness
 
diff --git a/explainers/code/utilization-as-a-label.py b/explainers/code/utilization-as-a-label.py
index 5f9c7ab..9c52476 100644
--- a/explainers/code/utilization-as-a-label.py
+++ b/explainers/code/utilization-as-a-label.py
@@ -1,5 +1,4 @@
 # Generated from explainers/utilization-as-a-label.md by scripts/build_explainers.py. Edit the markdown, not this file.
-import numpy as np
 import pandas as pd
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness
 
diff --git a/explainers/delayed-feedback.md b/explainers/delayed-feedback.md
index 65e732a..b3aabcd 100644
--- a/explainers/delayed-feedback.md
+++ b/explainers/delayed-feedback.md
@@ -59,7 +59,6 @@ def monitored_gap(mean_delay_group, mean_delay_others, today=90):
     delay = np.minimum(delay, 29.9)
     seen_positive = (y == 1) & (discharge + delay <= today)
     window_closed = discharge + 30 <= today
-    known_negative = (y == 0) & window_closed
     # a naive monitor counts every discharged patient and treats every outcome it has not seen as "no"
     naive = {g: seen_positive[group == g].mean() for g in (True, False)}
     # a careful monitor only counts patients whose 30-day window is closed
diff --git a/explainers/fairness-in-ranking.md b/explainers/fairness-in-ranking.md
index 9c4d144..4aa8059 100644
--- a/explainers/fairness-in-ranking.md
+++ b/explainers/fairness-in-ranking.md
@@ -87,7 +87,7 @@ def report(score, name):
     print(f"{name}\n  female share of the top {K}: {female[top].mean():.3f} (pool {female.mean():.3f})"
           f"\n  exposure ratio female/male, whole list: {whole[female == 1].mean() / whole[female == 0].mean():.3f}"
           f"\n  exposure ratio female/male, top {K} only: {cut[female == 1].mean() / cut[female == 0].mean():.3f}"
-          f"\n  exposure per unit of hire rate, female/male: "
+          "\n  exposure per unit of hire rate, female/male: "
           f"{(cut[female == 1].mean() / hired[female == 1].mean()) / (cut[female == 0].mean() / hired[female == 0].mean()):.3f}")
 
 report(scores(X_biased), "baseline (Gender is a feature)")
diff --git a/explainers/feature-importance-by-group.md b/explainers/feature-importance-by-group.md
index b22dca5..b3481bc 100644
--- a/explainers/feature-importance-by-group.md
+++ b/explainers/feature-importance-by-group.md
@@ -46,7 +46,6 @@ Part of the gap is mechanical. Women are a minority of the positive class in thi
 Run from the repo root (`PYTHONPATH=. python3 explainers/code/feature-importance-by-group.py`; that file is generated from the block below).
 
 ```python
-import numpy as np
 from sklearn.inspection import permutation_importance
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness
 from faircode.models import build_model
diff --git a/explainers/hiring-pipeline-adverse-impact.md b/explainers/hiring-pipeline-adverse-impact.md
index 13c6649..b206a5b 100644
--- a/explainers/hiring-pipeline-adverse-impact.md
+++ b/explainers/hiring-pipeline-adverse-impact.md
@@ -64,7 +64,6 @@ alive = np.ones(len(df), dtype=bool)
 def ratio_with_interval(passed, reached, n_boot=500):
     """Women's selection rate divided by everyone else's, among the people who reached the stage, with a bootstrap interval."""
     w, m = reached & (women == 1), reached & (women == 0)
-    rate = lambda mask, rows: passed[rows][mask[rows]].mean()
     point = passed[w].mean() / passed[m].mean()
     idx_w, idx_m = np.flatnonzero(w), np.flatnonzero(m)
     boots = [passed[rng.choice(idx_w, len(idx_w))].mean() / passed[rng.choice(idx_m, len(idx_m))].mean() for _ in range(n_boot)]
diff --git a/explainers/leave-one-group-out-evaluation.md b/explainers/leave-one-group-out-evaluation.md
index 0994fb8..9c1707b 100644
--- a/explainers/leave-one-group-out-evaluation.md
+++ b/explainers/leave-one-group-out-evaluation.md
@@ -50,7 +50,6 @@ The overall reading: on this data a forest generalizes to a held-out group about
 Run from the repo root (`PYTHONPATH=. python3 explainers/code/leave-one-group-out-evaluation.py`; that file is generated from the block below).
 
 ```python
-import numpy as np
 import pandas as pd
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness
 from faircode.models import build_model
diff --git a/explainers/monitoring-fairness-in-production.md b/explainers/monitoring-fairness-in-production.md
index 8e58cae..acf92fb 100644
--- a/explainers/monitoring-fairness-in-production.md
+++ b/explainers/monitoring-fairness-in-production.md
@@ -90,7 +90,7 @@ def batches_until_alert(batch, window, drift, drift_at=20, n_batches=60, z=3):
             return b - drift_at + 1 if b >= drift_at else -1       # -1 = a false alarm before the drift
     return None
 
-print(f"\nalert rule: the gap over the last W batches is more than 3 standard errors from the launch value")
+print("\nalert rule: the gap over the last W batches is more than 3 standard errors from the launch value")
 print(f"{'batch size':>10} {'window':>7} {'drift':>6} {'median batches to alert':>25} {'never alerted':>14} {'false alarms before the drift':>31}")
 for batch, window in ((200, 3), (500, 3), (1000, 3), (2000, 3), (1000, 1), (1000, 6)):
     for drift in (0.03, 0.06):
diff --git a/explainers/small-group-shrinkage.md b/explainers/small-group-shrinkage.md
index cc429f9..138ead1 100644
--- a/explainers/small-group-shrinkage.md
+++ b/explainers/small-group-shrinkage.md
@@ -92,7 +92,7 @@ for _ in range(300):
     small_raw.append(np.sqrt(np.mean((raw[small] - truth[small]) ** 2))); small_shrunk.append(np.sqrt(np.mean((shrunk[small] - truth[small]) ** 2)))
     estimates.append((raw[worst], shrunk[worst]))
 
-print(f"\nerror (RMSE against the true cell accuracy) from a 600-row audit, averaged over 300 audits")
+print("\nerror (RMSE against the true cell accuracy) from a 600-row audit, averaged over 300 audits")
 print(f"  all {len(cells)} cells:        raw {np.mean(raw_err):.3f} | shrunk {np.mean(shrunk_err):.3f}")
 print(f"  the smallest {small.sum()} cells:   raw {np.mean(small_raw):.3f} | shrunk {np.mean(small_shrunk):.3f}")
 mean_raw, mean_shrunk = np.mean(estimates, axis=0)
diff --git a/explainers/survivorship-bias.md b/explainers/survivorship-bias.md
index 6cdeb33..3b72a47 100644
--- a/explainers/survivorship-bias.md
+++ b/explainers/survivorship-bias.md
@@ -46,7 +46,6 @@ The size of the distortion depends on how hard the gate is. A gate that keeps on
 Run from the repo root (`PYTHONPATH=. python3 explainers/code/survivorship-bias.py`; that file is generated from the block below). The audit has the outcome for everyone, so we can compare against it; with real survivorship data you cannot, and the check is to ask what the gate was and model it.
 
 ```python
-import numpy as np
 from sklearn.linear_model import LogisticRegression
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness
 
diff --git a/explainers/utilization-as-a-label.md b/explainers/utilization-as-a-label.md
index 90d0f3c..0bd8c60 100644
--- a/explainers/utilization-as-a-label.md
+++ b/explainers/utilization-as-a-label.md
@@ -58,7 +58,6 @@ That is a finding. It says that for this label the Obermeyer mechanism is not vi
 Run from the repo root (`PYTHONPATH=. python3 explainers/code/utilization-as-a-label.py`; that file is generated from the block below). `data.df` holds the raw rows, so columns the harness does not encode can still be used to define strata.
 
 ```python
-import numpy as np
 import pandas as pd
 from faircode.benchmark import load_audit        # same loader, encoding and split as the benchmark harness