diff --git a/README.md b/README.md
index 4321835..239e84e 100644
--- a/README.md
+++ b/README.md
@@ -473,8 +473,10 @@ Representative confusion-matrix files:
 ## 16. Final Figures and Visualizations
 
 All figures live in `artifacts/figures/final/` and are regenerated from the audited
-`artifacts/analysis/final__*` tables by `scripts/make_final_figures.py` (no training; every number is
-loaded from an artifact). Each figure is saved as a high-resolution PNG (220 dpi) and a vector PDF.
+`artifacts/analysis/final__*` tables by `scripts/make_final_figures.py` and
+`scripts/make_comparison_figures.py` (no training; every number is loaded from an artifact). Each
+figure is saved as a high-resolution PNG (220 dpi) **and** a matching vector PDF, and its supporting
+source table is linked where useful.
 
 **Naming and legend conventions used throughout.**
 
@@ -494,9 +496,12 @@ loaded from an artifact). Each figure is saved as a high-resolution PNG (220 dpi
 | B | Beta selection surface | none (selection, not performance) |
 | C | Dataset heterogeneity | none (data distribution) |
 | D, E | Fixed-snapshot comparison vs `beta = 0` | `beta = 0` control |
-| F | Direct comparison vs `original_protean` | `original_protean` |
-| G | Per-client comparison vs `beta = 0` | `beta = 0` control |
-| H | Paired `original_protean` vs selected beta | `original_protean` |
+| F | Direct comparison vs `original_protean` (test) | `original_protean` |
+| G | Per-client comparison vs `beta = 0` (validation) | `beta = 0` control |
+| H | Paired `original_protean` vs selected beta confusion matrices | `original_protean` |
+| I1, I2 | Per-client test comparison vs `original_protean` | `original_protean` |
+| J1, J2 | Per-class test comparison vs `original_protean` | `original_protean` |
+| K | Ranked confusion change vs `original_protean` (+ full matrix supplement) | `original_protean` |
 
 ### Fig A1 / A2 — Validation-based beta selection (support_weighted)
 
@@ -548,18 +553,23 @@ the imbalance-displacement link is mechanical under the beta formula and is not
 
 ![Support CV vs displacement](artifacts/figures/final/figE_support_cv_vs_displacement.png)
 
-### Fig F — Selected-beta PROTEAN vs ORIGINAL PROTEAN (test)
+### Fig F — Selected-beta PROTEAN vs ORIGINAL PROTEAN (test), two panels
 
-**Compared:** `original_protean` (charcoal) against `support_weighted` at the validation-selected
-beta (teal), per condition. **Level/split:** test, mean over seeds `{0,1,2}`. **Metric:** test
-rare-class macro-F1; error bars = ±1 sample std across seeds. **Observation:** signed deltas are
-`+0.0230`, `+0.0014`, `+0.0168` (positive), `−0.0086` (negative), and `−0.0007`/`−0.0010` at the two
-`beta = 0` conditions (floating-point ~0). **Limits:** no statistical significance test (three
-seeds); the largest gain is within seed noise; the weighting does not consistently outperform
-`original_protean`.
+**Compared:** `original_protean` against `support_weighted` at the validation-selected beta, per
+condition, test split. **How to read:** Panel A shows the absolute mean test rare-class macro-F1
+(±1 sample std across seeds) and is deliberately near-flat; Panel B shows the signed difference
+`proposed − original` with the exact value printed at each bar and the selected beta annotated.
+**Main result:** three positive deltas (X-IIoTID a0.50 +0.0230, X-IIoTID a0.75 +0.0014, 5G-NIDD a0.75
++0.0168), one negative delta (5G-NIDD a0.50 −0.0086), and two `beta=0` conditions that are
+effectively unchanged (−0.0007 / −0.0010). **Limitation:** no significance test (three seeds); a
+positive mean delta is not a proven improvement.
 
 ![Selected vs original test rare macro-F1](artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.png)
 
+Links: [PNG](artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.png) ·
+[PDF](artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.pdf) · source
+[`final__six_condition_master.csv`](artifacts/analysis/final__six_condition_master.csv).
+
 ### Fig G — Per-client effect of selected beta relative to the beta=0 control
 
 **Compared:** selected-beta `support_weighted` against the `beta = 0` control (not
@@ -571,17 +581,25 @@ client ids are per-seed realizations and are not pooled.
 
 ![Client effect at selected beta](artifacts/figures/final/figG_client_effect_at_selected_beta.png)
 
-### Fig H (supplemental) — Paired test confusion matrices
+### Fig H — Paired test confusion matrices (original vs beta-tuned)
 
 **Compared:** `original_protean` (left) against selected-beta `support_weighted` (right) for a valid
-paired condition (5G-NIDD, `alpha = 0.75`, `seed = 0`). **Level/split:** test; global counts over the
-evaluation split. **Metric:** raw 9×9 counts (log colour scale), identical class order and colour
-scale. **Observation:** the dominant off-diagonal error is `Benign → UDPFlood` (~56.9k vs ~57.1k) and
-is essentially unchanged by beta. **Limits:** no statistical significance is inferred from the visual
-difference; the matrices come from the same manifest, split, seed, and class order.
+paired condition (5G-NIDD, `alpha = 0.75`, `seed = 0`), test split. **How to read:** rows = actual
+class, columns = predicted class; the diagonal is correct and off-diagonal is misclassified. Raw
+counts, identical class order and colour scale; both matrices contain the same 243,177 test samples.
+**Main result:** the dominant residual error `Benign → UDPFlood` is essentially unchanged (56,899 →
+57,054, +155). The summary below the figure lists the most important changes (`HTTPFlood →
+SlowrateDoS` −547; `Benign → TCPConnectScan` −520; `SlowrateDoS → HTTPFlood` +229; diagonal
+`HTTPFlood` +547, `Benign` +288, `SlowrateDoS` −239). **Limitation:** no significance test; the
+`Benign → UDPFlood` increase is **not** an improvement.
 
 ![Paired confusion matrices](artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.png)
 
+Links: [PNG](artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.png) ·
+[PDF](artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.pdf) · source
+`results/5gnidd__alpha0.75__seed0__original_protean/test_confusion_matrix.json` and
+`results/5gnidd__alpha0.75__seed0__support_weighted__beta0.5/test_confusion_matrix.json`.
+
 
 ## 17. Deep Comparison: Original PROTEAN vs Proposed (Result)
 
@@ -609,34 +627,71 @@ Both `alpha=0.25` conditions selected `beta=0` and are numerically null. Among t
 with it, while rare-class recall usually falls (precision–recall trade-off). 5G-NIDD a0.50 is
 seed-dominated. No statistical significance is claimed.
 
-### Fig I — Paired per-client test rare-class macro-F1 (proposed vs ORIGINAL PROTEAN)
+### Fig I1 / I2 — Paired per-client test rare-class macro-F1 (proposed vs ORIGINAL PROTEAN)
 
-Each point is one (client, seed) test observation, faceted by dataset and alpha; charcoal/teal/red/gray
-and markers encode outcome and seed. At `beta>0` conditions the response is client-specific (e.g.
-X-IIoTID a0.50: 12/30 improved, 1 degraded, 17 unchanged), with individual changes larger than the
-condition mean; at `beta=0` conditions almost all observations are unchanged.
+One figure per dataset, three alpha panels. **Compared:** `original_protean` against selected-beta
+`support_weighted` on the test split, each client on its own evaluation partition. **How to read:**
+x = client id, one marker per seed (circle = seed 0, square = seed 1, triangle = seed 2),
+y = Δ rare-class macro-F1; the shaded band is the ±0.02 effectively-unchanged threshold; the strongest
+positive and negative observation in each panel is labelled `c<client> s<seed>`. **Main result:**
+X-IIoTID α=0.25 essentially unchanged (1 / 1 / 28 improved/degraded/unchanged), α=0.50 12 / 1 / 17
+(mean +0.030, range −0.257 … +0.324), α=0.75 8 / 2 / 20; 5G-NIDD α=0.25 1 / 0 / 29, α=0.50 15 / 6 / 9
+(seed-dominated), α=0.75 8 / 1 / 21. **Limitation:** client ids are per-seed realizations (id `k` is
+not the same entity across seeds); at α=0.25 the selected beta is 0 so the residual spread is
+numerical, not a weighting effect.
 
-![Paired per-client comparison](artifacts/figures/final/figI_per_client_original_vs_proposed.png)
+![Per-client comparison, X-IIoTID](artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.png)
 
-### Fig J — Per-class test F1 change (beta>0 conditions)
+![Per-client comparison, 5G-NIDD](artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.png)
 
-Signed Δ F1 per class, panels per dataset-alpha; class support is annotated. Largest gains are
-X-IIoTID `Tampering` (+0.14) and `Reconnaissance` (+0.05) and 5G-NIDD `TCPConnectScan` (+0.03) /
-`ICMPFlood` (+0.10, tiny support); largest losses are X-IIoTID `C&C` (−0.06) and 5G-NIDD `SYNFlood`
-(−0.04). Benign detection is essentially unchanged.
+Links: [I1 PNG](artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.png) ·
+[I1 PDF](artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.pdf) ·
+[I2 PNG](artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.png) ·
+[I2 PDF](artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.pdf) · source
+[`deepcmp__per_client.csv`](artifacts/analysis/deepcmp__per_client.csv).
 
-![Per-class comparison](artifacts/figures/final/figJ_per_class_original_vs_proposed.png)
+### Fig J1 / J2 — Per-class test F1 change (beta>0 conditions)
 
-### Fig K — Confusion-matrix difference (proposed − original)
+One figure per dataset, two alpha panels. **Compared:** `original_protean` against selected-beta
+`support_weighted`, test split. **How to read:** horizontal bars are the mean ΔF1 over seeds `{0,1,2}`
+(teal = increased, red = decreased) with the exact signed ΔF1 printed at each bar; test support `n≈`
+is on a second label line; both panels share one x-scale so bar lengths are comparable. **Main
+result:** X-IIoTID largest gains `Tampering` +0.1438 and `Reconnaissance` +0.0542, largest loss `C&C`
+−0.0556; 5G-NIDD largest gain `TCPConnectScan` +0.0342 / `ICMPFlood` +0.0970 (tiny support), largest
+loss `SYNFlood` −0.0426. **Limitation:** the delta is **F1**, not precision or recall (e.g. Tampering
++0.1438 F1 corresponds to +0.2120 precision and −0.0091 recall); no significance test.
+
+![Per-class comparison, X-IIoTID](artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.png)
+
+![Per-class comparison, 5G-NIDD](artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.png)
+
+Links: [J1 PNG](artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.png) ·
+[J1 PDF](artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.pdf) ·
+[J2 PNG](artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.png) ·
+[J2 PDF](artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.pdf) · source
+[`deepcmp__per_class.csv`](artifacts/analysis/deepcmp__per_class.csv).
+
+### Fig K — Ranked confusion change vs ORIGINAL PROTEAN (+ full-matrix supplement)
 
 For the paired condition 5G-NIDD, `alpha=0.75`, `seed=0` (identical manifest, split, class order,
-243,177 test samples): left = raw count difference, right = row-normalized rate difference. The
-dominant error `Benign → UDPFlood` is essentially unchanged (56,899 → 57,054, +155). The main reduced
-errors are `HTTPFlood → SlowrateDoS` (−547) and `Benign → TCPConnectScan` (−520); the main increased
-error is `SlowrateDoS → HTTPFlood` (+229). Red = positive delta, blue = negative; no significance is
-inferred.
+243,177 test samples). **How to read:** the ranked view lists the largest off-diagonal changes as raw
+test counts; the upper panel shows *more* errors after weighting, the lower panel *fewer*.
+**Main result:** the largest reduced errors are `HTTPFlood → SlowrateDoS` (−547; 759 → 212) and
+`Benign → TCPConnectScan` (−520; 2,015 → 1,495); the largest increased errors are `SlowrateDoS →
+HTTPFlood` (+229) and `Benign → UDPFlood` (+155). The dominant error `Benign → UDPFlood` is essentially
+unchanged and is **not** an improvement. **Limitation:** no significance test; the figure is a single
+paired condition, not a condition average. The full difference matrices (raw count and row-rate) are
+kept as a supplemental figure for traceability.
 
-![Confusion-matrix difference](artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.png)
+![Ranked confusion change](artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.png)
+
+![Full confusion-matrix difference, supplemental](artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.png)
+
+Links: [ranked PNG](artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.png) ·
+[ranked PDF](artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.pdf) ·
+[full-matrix PNG](artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.png) ·
+[full-matrix PDF](artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.pdf) · source
+[`deepcmp__confusion_delta.csv`](artifacts/analysis/deepcmp__confusion_delta.csv).
 
 **Per-client findings.** The proposed method redistributes benefit: within a condition some clients
 improve and others degrade, and the low-rare-support quartile is not systematically disadvantaged
@@ -792,7 +847,8 @@ Analysis outputs (`artifacts/analysis/`), by function:
 | Prototype snapshots & weights | `*__prototype_snapshot.json`, `*__prototype_weights.csv`, `*__prototype_effect*.csv` |
 
 Partition manifests: `artifacts/partitions/<dataset>__alpha<alpha>__seed<seed>.json` (18 files).
-Figures: `artifacts/figures/final/` — 12 figures (figA–figK) as PNG (220 dpi) plus matching vector PDF;
+Figures: `artifacts/figures/final/` — 15 figures (figA–figK, including the split I1/I2 and J1/J2 and
+the K full-matrix supplement) as PNG (220 dpi) plus matching vector PDF;
 regenerated reproducibly by `scripts/make_final_figures.py`.
 
 Key documents:
diff --git a/artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.pdf b/artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.pdf
index 8962651..c94028b 100644
Binary files a/artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.pdf and b/artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.pdf differ
diff --git a/artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.png b/artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.png
index c994564..9ff0e4d 100644
Binary files a/artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.png and b/artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.png differ
diff --git a/artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.pdf b/artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.pdf
index 0527396..04a6b3f 100644
Binary files a/artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.pdf and b/artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.pdf differ
diff --git a/artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.png b/artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.png
index a65f9ba..8def404 100644
Binary files a/artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.png and b/artifacts/figures/final/figH_confusion_matrix_5gnidd_a0.75_seed0.png differ
diff --git a/artifacts/figures/final/figI_per_client_original_vs_proposed.pdf b/artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.pdf
similarity index 57%
rename from artifacts/figures/final/figI_per_client_original_vs_proposed.pdf
rename to artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.pdf
index 912ba34..9f7af1c 100644
Binary files a/artifacts/figures/final/figI_per_client_original_vs_proposed.pdf and b/artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.pdf differ
diff --git a/artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.png b/artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.png
new file mode 100644
index 0000000..8602b93
Binary files /dev/null and b/artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.png differ
diff --git a/artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.pdf b/artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.pdf
new file mode 100644
index 0000000..053d602
Binary files /dev/null and b/artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.pdf differ
diff --git a/artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.png b/artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.png
new file mode 100644
index 0000000..31c4f53
Binary files /dev/null and b/artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.png differ
diff --git a/artifacts/figures/final/figI_per_client_original_vs_proposed.png b/artifacts/figures/final/figI_per_client_original_vs_proposed.png
deleted file mode 100644
index 2fbf903..0000000
Binary files a/artifacts/figures/final/figI_per_client_original_vs_proposed.png and /dev/null differ
diff --git a/artifacts/figures/final/figJ_per_class_original_vs_proposed.pdf b/artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.pdf
similarity index 60%
rename from artifacts/figures/final/figJ_per_class_original_vs_proposed.pdf
rename to artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.pdf
index fad8f1e..0b42b2a 100644
Binary files a/artifacts/figures/final/figJ_per_class_original_vs_proposed.pdf and b/artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.pdf differ
diff --git a/artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.png b/artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.png
new file mode 100644
index 0000000..90df82a
Binary files /dev/null and b/artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.png differ
diff --git a/artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.pdf b/artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.pdf
new file mode 100644
index 0000000..9a1ea04
Binary files /dev/null and b/artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.pdf differ
diff --git a/artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.png b/artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.png
new file mode 100644
index 0000000..47106ab
Binary files /dev/null and b/artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.png differ
diff --git a/artifacts/figures/final/figJ_per_class_original_vs_proposed.png b/artifacts/figures/final/figJ_per_class_original_vs_proposed.png
deleted file mode 100644
index edb1f77..0000000
Binary files a/artifacts/figures/final/figJ_per_class_original_vs_proposed.png and /dev/null differ
diff --git a/artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.pdf b/artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.pdf
index 7c0ee78..09e2c27 100644
Binary files a/artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.pdf and b/artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.pdf differ
diff --git a/artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.png b/artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.png
index f90e630..0fa071e 100644
Binary files a/artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.png and b/artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.png differ
diff --git a/artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.pdf b/artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.pdf
new file mode 100644
index 0000000..45b966b
Binary files /dev/null and b/artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.pdf differ
diff --git a/artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.png b/artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.png
new file mode 100644
index 0000000..52bbd27
Binary files /dev/null and b/artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.png differ
diff --git a/docs/research/ORIGINAL_VS_PROPOSED_DEEP_COMPARISON.md b/docs/research/ORIGINAL_VS_PROPOSED_DEEP_COMPARISON.md
index 8a4d45e..b64df7f 100644
--- a/docs/research/ORIGINAL_VS_PROPOSED_DEEP_COMPARISON.md
+++ b/docs/research/ORIGINAL_VS_PROPOSED_DEEP_COMPARISON.md
@@ -133,7 +133,9 @@ classes while giving up a little rare-class coverage.
 
 ## 6. Paired per-client results (test, vs original_protean)
 
-Figure I: `artifacts/figures/final/figI_per_client_original_vs_proposed.png`.
+Figures I1/I2 (one per dataset):
+`artifacts/figures/final/figI1_xiiotid_per_client_original_vs_proposed.png`,
+`artifacts/figures/final/figI2_5gnidd_per_client_original_vs_proposed.png`.
 Per-observation table: `deepcmp__per_client.csv` (30 observations per condition).
 
 | Condition | beta* | improved | degraded | unchanged | mean Δ | min Δ | max Δ |
@@ -159,7 +161,9 @@ Counts use the ±0.02 effectively-unchanged band on per-client test rare-class m
 
 ## 7. Per-class precision/recall/F1 changes (beta>0 conditions)
 
-Figure J: `artifacts/figures/final/figJ_per_class_original_vs_proposed.png`.
+Figures J1/J2 (one per dataset):
+`artifacts/figures/final/figJ1_xiiotid_per_class_original_vs_proposed.png`,
+`artifacts/figures/final/figJ2_5gnidd_per_class_original_vs_proposed.png`.
 Source: `deepcmp__per_class.csv` (per seed) aggregated as the mean delta over the three seeds; class
 support is the mean test support over seeds. X-IIoTID and 5G-NIDD labels are never mixed.
 
@@ -198,8 +202,11 @@ Points to note:
 
 ## 8. Confusion-matrix difference analysis
 
-Figure K shows the paired difference for **5G-NIDD, alpha=0.75, seed=0** (the same pair as Figure H):
-`artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.png`. Rows = true class, columns =
+The ranked-change figure (primary) is
+`artifacts/figures/final/figK_confusion_error_changes_5gnidd_a0.75_seed0.png`; the full difference
+matrices (raw count and row-rate) are kept as the supplemental
+`artifacts/figures/final/figK_confusion_delta_5gnidd_a0.75_seed0.png`. Both describe the paired
+difference for **5G-NIDD, alpha=0.75, seed=0** (the same pair as Figure H). Rows = true class, columns =
 predicted class; both matrices have identical class order and 243,177 test samples. Source:
 `deepcmp__confusion_delta.csv`.
 
@@ -227,8 +234,8 @@ Largest raw-count changes for this pair (proposed − original):
   are X-IIoTID `Reconnaissance → Exploitation` (−6,773), `Reconnaissance → Weaponization` (−3,398),
   `Lateral_movement → Tampering` (−1,699) and 5G-NIDD `Benign → TCPConnectScan` (−1,484); the largest
   **increased** ones are X-IIoTID `Normal → C&C` (+2,546) and `Weaponization → Exploitation` (+1,286).
-  Raw counts are support-sensitive, so these are descriptive; the row-normalized view (Figure K,
-  right panel) is the support-fair comparison.
+  Raw counts are support-sensitive, so these are descriptive; the row-normalized view in the
+  supplemental Figure K matrix (right panel) is the support-fair comparison.
 
 ## 9. Support imbalance versus client outcomes
 
diff --git a/scripts/make_comparison_figures.py b/scripts/make_comparison_figures.py
index ac46800..868ae59 100644
--- a/scripts/make_comparison_figures.py
+++ b/scripts/make_comparison_figures.py
@@ -1,13 +1,16 @@
 #!/usr/bin/env python3
 """Figures I, J, K: original_protean vs support_weighted(selected beta).
 
-Figure I  paired per-client rare-class F1 deltas (test), vs original_protean.
-Figure J  per-class test F1 change, beta>0 conditions.
-Figure K  test confusion-matrix difference (5G-NIDD, alpha=0.75, seed=0).
+Readability overhaul. Produces:
 
-Reads derived tables written by ``scripts/deep_comparison.py`` and the raw
-per-run confusion matrices. Writes PNG (220 dpi) and PDF to
-``artifacts/figures/final``. No training, no result mutation.
+  figI1 / figI2  paired per-client test rare-class F1 deltas, one figure per dataset
+  figJ1 / figJ2  per-class test F1 change, one figure per dataset (two alpha panels)
+  figK_confusion_error_changes_*   ranked reduced/increased misclassifications (primary)
+  figK_confusion_delta_*           full difference matrix (supplemental, preserved)
+
+Reads derived tables from ``scripts/deep_comparison.py`` and raw confusion
+matrices. Writes PNG (220 dpi) and PDF to ``artifacts/figures/final``.
+No training, no result mutation.
 
 Run:
     .venv/bin/python scripts/make_comparison_figures.py
@@ -17,9 +20,9 @@ from __future__ import annotations
 
 import csv
 import json
+import statistics
 from collections import defaultdict
 from pathlib import Path
-import statistics
 
 import matplotlib
 
@@ -35,11 +38,13 @@ OUT = ROOT / "artifacts" / "figures" / "final"
 
 CHARCOAL = "#3A3A3A"
 TEAL = "#0E7C7B"
+GREEN = "#1B7F3B"
 RED = "#C0392B"
 GREY = "#9A9A9A"
-DATASETS = ["xiiotid", "5gnidd"]
 DATASET_NAME = {"xiiotid": "X-IIoTID", "5gnidd": "5G-NIDD"}
 ALPHAS = [0.25, 0.50, 0.75]
+SELECTED = {"xiiotid": {0.25: 0.0, 0.5: 0.75, 0.75: 0.5},
+            "5gnidd": {0.25: 0.0, 0.5: 1.0, 0.75: 0.5}}
 TOL = 0.02
 SEED_MARKERS = {0: "o", 1: "s", 2: "^"}
 
@@ -48,10 +53,10 @@ def _style():
     plt.rcParams.update({
         "figure.facecolor": "white", "axes.facecolor": "white",
         "axes.edgecolor": "#5A5A5A", "axes.linewidth": 0.9,
-        "axes.titlesize": 11.5, "axes.titleweight": "bold",
-        "axes.labelsize": 10.5, "xtick.labelsize": 9,
-        "ytick.labelsize": 9, "legend.fontsize": 9,
-        "font.size": 10, "axes.spines.top": False, "axes.spines.right": False,
+        "axes.titlesize": 12, "axes.titleweight": "bold",
+        "axes.labelsize": 11, "xtick.labelsize": 10, "ytick.labelsize": 10,
+        "legend.fontsize": 9.5, "font.size": 10.5,
+        "axes.spines.top": False, "axes.spines.right": False,
         "grid.color": "#B0B0B0", "grid.alpha": 0.3, "grid.linestyle": ":",
         "savefig.dpi": 220, "savefig.bbox": "tight",
     })
@@ -71,102 +76,175 @@ def _read(path):
 
 
 # ---------------------------------------------------------------------------
-def figure_i():
-    rows = _read(ANALYSIS / "deepcmp__per_client.csv")
-    fig, axes = plt.subplots(2, 3, figsize=(14.0, 7.4), constrained_layout=True,
-                             sharey=True)
-    ylim = 0.47
-    for ri, ds in enumerate(DATASETS):
-        for ci, alpha in enumerate(ALPHAS):
-            ax = axes[ri][ci]
-            ax.set_axisbelow(True)
-            ax.grid(axis="y")
-            for seed in (0, 1, 2):
-                sub = [r for r in rows if r["dataset"] == ds
-                       and abs(float(r["alpha"]) - alpha) < 1e-9
-                       and int(r["seed"]) == seed]
-                xs = [int(r["client_id"]) for r in sub]
-                ys = [float(r["delta_rare_f1"]) for r in sub]
-                cols = [TEAL if y > TOL else (RED if y < -TOL else GREY) for y in ys]
-                ax.scatter(xs, ys, s=48, marker=SEED_MARKERS[seed], c=cols,
-                           edgecolor="#333333", linewidth=0.5, alpha=0.9, zorder=4)
-            ax.axhspan(-TOL, TOL, color="#DDDDDD", alpha=0.35, zorder=1)
-            ax.axhline(0.0, color="#555555", lw=1.0, zorder=2)
-            beta = {"xiiotid": {0.25: 0.0, 0.5: 0.75, 0.75: 0.5},
-                    "5gnidd": {0.25: 0.0, 0.5: 1.0, 0.75: 0.5}}[ds][alpha]
-            deltas = [float(r["delta_rare_f1"]) for r in rows
-                      if r["dataset"] == ds and abs(float(r["alpha"]) - alpha) < 1e-9]
-            ax.set_ylim(-ylim, ylim)
-            ax.set_xlim(-0.7, 9.7)
-            ax.set_xticks(range(10))
-            ax.set_title(f"{DATASET_NAME[ds]}  \u03b1={alpha:.2f}   "
-                         f"(beta={beta:g})")
-            ax.text(0.98, 0.04, f"mean \u0394={statistics.mean(deltas):+.3f}",
-                    transform=ax.transAxes, ha="right", va="bottom",
-                    fontsize=8.5, color="#333333")
-            if ri == 1:
-                ax.set_xlabel("client id (per-seed partition)")
-            if ci == 0:
-                ax.set_ylabel("test rare-class macro-F1\n\u0394 (proposed \u2212 original)")
-    handles = [plt.Line2D([0], [0], marker="o", ls="", ms=7, mfc=TEAL, mec="#333", label="improved (> +0.02)"),
-               plt.Line2D([0], [0], marker="o", ls="", ms=7, mfc=RED, mec="#333", label="degraded (< -0.02)"),
-               plt.Line2D([0], [0], marker="o", ls="", ms=7, mfc=GREY, mec="#333", label="unchanged (|\u0394| \u2264 0.02)"),
-               plt.Line2D([0], [0], marker="o", ls="", ms=7, mfc="white", mec="#333", label="seed 0"),
-               plt.Line2D([0], [0], marker="s", ls="", ms=7, mfc="white", mec="#333", label="seed 1"),
-               plt.Line2D([0], [0], marker="^", ls="", ms=7, mfc="white", mec="#333", label="seed 2")]
+# Figure I: paired per-client rare-class F1 vs original_protean (one per dataset)
+# ---------------------------------------------------------------------------
+def figure_i(dataset: str) -> None:
+    rows = [r for r in _read(ANALYSIS / "deepcmp__per_client.csv")
+            if r["dataset"] == dataset]
+    all_d = [float(r["delta_rare_f1"]) for r in rows]
+    ymax = max(abs(min(all_d)), abs(max(all_d))) * 1.12
+
+    fig, axes = plt.subplots(1, 3, figsize=(16.0, 5.6), sharey=True,
+                             constrained_layout=True)
+    for ax, alpha in zip(axes, ALPHAS):
+        ax.set_axisbelow(True)
+        ax.grid(axis="y")
+        ax.axhspan(-TOL, TOL, color="#DDDDDD", alpha=0.4, zorder=1)
+        ax.axhline(0.0, color="#444444", lw=1.1, zorder=2)
+        sub = [r for r in rows if abs(float(r["alpha"]) - alpha) < 1e-9]
+        for seed in (0, 1, 2):
+            pts = [r for r in sub if int(r["seed"]) == seed]
+            xs = [int(r["client_id"]) + (seed - 1) * 0.18 for r in pts]
+            ys = [float(r["delta_rare_f1"]) for r in pts]
+            cols = [TEAL if y > TOL else (RED if y < -TOL else GREY) for y in ys]
+            ax.scatter(xs, ys, s=62, marker=SEED_MARKERS[seed], c=cols,
+                       edgecolor="#333333", linewidth=0.6, zorder=4)
+        # annotate the strongest positive and negative observation
+        for r in (max(sub, key=lambda r: float(r["delta_rare_f1"])),
+                  min(sub, key=lambda r: float(r["delta_rare_f1"]))):
+            y = float(r["delta_rare_f1"])
+            ax.annotate(f"c{r['client_id']} s{r['seed']}", (int(r["client_id"]), y),
+                        textcoords="offset points", xytext=(6, 6 if y >= 0 else -12),
+                        fontsize=8.2, color="#333333")
+        imp = sum(1 for r in sub if float(r["delta_rare_f1"]) > TOL)
+        deg = sum(1 for r in sub if float(r["delta_rare_f1"]) < -TOL)
+        unc = len(sub) - imp - deg
+        ax.text(0.5, 0.985, f"improved {imp}  |  degraded {deg}  |  unchanged {unc}",
+                transform=ax.transAxes, ha="center", va="top", fontsize=9,
+                bbox=dict(boxstyle="round", fc="white", ec="#CCCCCC"))
+        ax.set_ylim(-ymax, ymax)
+        ax.set_xlim(-0.75, 9.75)
+        ax.set_xticks(range(10))
+        ax.set_xlabel("client id (per-seed partition)")
+        beta = SELECTED[dataset][alpha]
+        ax.set_title(f"{DATASET_NAME[dataset]}  \u03b1={alpha:.2f}   (beta={beta:g})")
+    axes[0].set_ylabel("test rare-class macro-F1\n\u0394 (proposed \u2212 ORIGINAL PROTEAN)")
+    handles = [plt.Line2D([0], [0], marker="o", ls="", ms=8, mfc=TEAL, mec="#333", label=f"improved (> +{TOL})"),
+               plt.Line2D([0], [0], marker="o", ls="", ms=8, mfc=RED, mec="#333", label=f"degraded (< \u2212{TOL})"),
+               plt.Line2D([0], [0], marker="o", ls="", ms=8, mfc=GREY, mec="#333", label=f"unchanged (|\u0394| \u2264 {TOL})"),
+               plt.Line2D([0], [0], marker="o", ls="", ms=8, mfc="white", mec="#333", label="seed 0"),
+               plt.Line2D([0], [0], marker="s", ls="", ms=8, mfc="white", mec="#333", label="seed 1"),
+               plt.Line2D([0], [0], marker="^", ls="", ms=8, mfc="white", mec="#333", label="seed 2")]
     fig.legend(handles=handles, loc="upper center", bbox_to_anchor=(0.5, 1.05),
                ncols=6, frameon=True, edgecolor="#CCCCCC")
-    fig.suptitle("Paired per-client test rare-class macro-F1: proposed (selected beta) vs ORIGINAL PROTEAN",
-                 fontsize=12.5, weight="bold", y=1.11)
-    fig.text(0.5, -0.03,
-             "Each point is one (client, seed) test observation; x = client id, marker = seed. Points compare the "
-             "support_weighted run at the validation-selected beta against original_protean on the same partition. "
-             "At \u03b1=0.25 the selected beta is 0, so the residual spread is numerical (equal-weight equivalence), not a weighting effect. "
-             "Do not treat client id across seeds as the same real-world client.",
-             ha="center", va="top", fontsize=8.4, color="#444444", linespacing=1.5)
-    _save(fig, "figI_per_client_original_vs_proposed")
+    fig.suptitle(f"Per-client test rare-class macro-F1: proposed vs ORIGINAL PROTEAN \u2014 {DATASET_NAME[dataset]}",
+                 fontsize=13, weight="bold", y=1.11)
+    fig.text(0.5, -0.04,
+             "Each point is one (client, seed) test observation on the client's own evaluation partition; same partition for both runs. "
+             "The shaded band is the \u00b10.02 effectively-unchanged threshold. Client ids are per-seed realizations (id k is not the same entity across seeds). "
+             "At \u03b1=0.25 the selected beta is 0, so the residual spread is numerical, not a weighting effect.",
+             ha="center", va="top", fontsize=8.6, color="#444444", linespacing=1.5)
+    _save(fig, f"figI{'1' if dataset == 'xiiotid' else '2'}_{dataset}_per_client_original_vs_proposed")
 
 
 # ---------------------------------------------------------------------------
-def figure_j():
-    rows = _read(ANALYSIS / "deepcmp__per_class.csv")
-    fig, axes = plt.subplots(2, 2, figsize=(13.4, 8.2), constrained_layout=True)
-    for ri, ds in enumerate(DATASETS):
-        for ci, alpha in enumerate([0.5, 0.75]):
-            ax = axes[ri][ci]
-            ax.set_axisbelow(True)
-            ax.grid(axis="x")
-            stats = defaultdict(lambda: {"f1": [], "sup": []})
-            for r in rows:
-                if r["dataset"] == ds and abs(float(r["alpha"]) - alpha) < 1e-9:
-                    stats[r["class_name"]]["f1"].append(float(r["delta_f1"]))
-                    stats[r["class_name"]]["sup"].append(int(r["support"]))
-            names = sorted(stats, key=lambda n: statistics.mean(stats[n]["f1"]))
-            ys = np.arange(len(names))
-            vals = [statistics.mean(stats[n]["f1"]) for n in names]
-            sup = [int(statistics.mean(stats[n]["sup"])) for n in names]
-            cols = [TEAL if v >= 0 else RED for v in vals]
-            ax.barh(ys, vals, color=cols, edgecolor="#333333", linewidth=0.6, zorder=3)
-            ax.axvline(0.0, color="#555555", lw=1.0, zorder=2)
-            ax.set_yticks(ys)
-            ax.set_yticklabels([f"{n}  (n\u2248{s:,})" for n, s in zip(names, sup)],
-                               fontsize=8.2)
-            beta = {"xiiotid": {0.5: 0.75, 0.75: 0.5},
-                    "5gnidd": {0.5: 1.0, 0.75: 0.5}}[ds][alpha]
-            ax.set_title(f"{DATASET_NAME[ds]}  \u03b1={alpha:.2f}   beta={beta:g}")
-            ax.set_xlabel("\u0394 F1 (proposed \u2212 original), test")
-    fig.suptitle("Per-class test F1 change for beta>0 conditions (mean over seeds {0,1,2})",
+# Figure J: per-class test F1 change (one per dataset; two alpha panels)
+# ---------------------------------------------------------------------------
+def figure_j(dataset: str) -> None:
+    rows = [r for r in _read(ANALYSIS / "deepcmp__per_class.csv")
+            if r["dataset"] == dataset]
+    alphas = [0.5, 0.75]
+    stats = {}
+    for alpha in alphas:
+        agg = defaultdict(lambda: {"f1": [], "sup": []})
+        for r in rows:
+            if abs(float(r["alpha"]) - alpha) < 1e-9:
+                agg[r["class_name"]]["f1"].append(float(r["delta_f1"]))
+                agg[r["class_name"]]["sup"].append(int(r["support"]))
+        stats[alpha] = {n: (statistics.mean(v["f1"]), int(statistics.mean(v["sup"])))
+                        for n, v in agg.items()}
+    amax = max(abs(v[0]) for a in alphas for v in stats[a].values())
+    xlim = amax * 1.35
+
+    fig, axes = plt.subplots(1, 2, figsize=(15.5, 6.4), constrained_layout=True)
+    for ax, alpha in zip(axes, alphas):
+        ax.set_axisbelow(True)
+        ax.grid(axis="x")
+        items = sorted(stats[alpha].items(), key=lambda kv: kv[1][0])
+        names = [n for n, _ in items]
+        vals = [v[0] for _, v in items]
+        sups = [v[1] for _, v in items]
+        ys = np.arange(len(names))
+        cols = [TEAL if v >= 0 else RED for v in vals]
+        bars = ax.barh(ys, vals, color=cols, edgecolor="#333333", linewidth=0.6, zorder=3)
+        ax.axvline(0.0, color="#444444", lw=1.1, zorder=2)
+        for y, v in zip(ys, vals):
+            off = amax * 0.03
+            ax.annotate(f"{v:+.4f}", (v + (off if v >= 0 else -off), y),
+                        ha="left" if v >= 0 else "right", va="center",
+                        fontsize=9.2, weight="bold",
+                        color="#1B7F3B" if v >= 0 else "#8E1F14")
+        ax.set_yticks(ys)
+        ax.set_yticklabels([f"{n}\ntest n\u2248{s:,}" for n, s in zip(names, sups)],
+                           fontsize=10)
+        ax.set_xlim(-xlim, xlim)
+        ax.set_xlabel("\u0394 F1  (proposed \u2212 ORIGINAL PROTEAN), test")
+        beta = SELECTED[dataset][alpha]
+        ax.set_title(f"{DATASET_NAME[dataset]}  \u03b1={alpha:.2f}   (beta={beta:g})")
+    fig.suptitle(f"Per-class test F1 change vs ORIGINAL PROTEAN \u2014 {DATASET_NAME[dataset]}",
+                 fontsize=13, weight="bold", y=1.04)
+    fig.text(0.5, -0.02,
+             "Bars are mean \u0394F1 over seeds {0,1,2}; both panels use the same x-scale so bar lengths are comparable. "
+             "Teal = F1 increased, red = F1 decreased and the exact signed \u0394F1 is printed at each bar. Delta is F1, NOT precision or recall. "
+             "Only beta>0 conditions are shown (\u03b1=0.25 selected beta=0). Test support is the mean over seeds. No significance test (3 seeds).",
+             ha="center", va="top", fontsize=8.6, color="#444444", linespacing=1.5)
+    _save(fig, f"figJ{'1' if dataset == 'xiiotid' else '2'}_{dataset}_per_class_original_vs_proposed")
+
+
+# ---------------------------------------------------------------------------
+# Figure K (primary): ranked error changes for 5G-NIDD alpha=0.75 seed=0
+# ---------------------------------------------------------------------------
+def figure_k_ranked(dataset="5gnidd", alpha=0.75, seed=0):
+    rows = [r for r in _read(ANALYSIS / "deepcmp__confusion_delta.csv")
+            if r["dataset"] == dataset and abs(float(r["alpha"]) - alpha) < 1e-9
+            and int(r["seed"]) == seed]
+    off = [r for r in rows if r["true_class"] != r["pred_class"]]
+    reduced = sorted(off, key=lambda r: int(r["delta"]))[:5]
+    increased = sorted(off, key=lambda r: -int(r["delta"]))[:5]
+
+    fig, (ax1, ax2) = plt.subplots(2, 1, figsize=(11.5, 7.0),
+                                   constrained_layout=True, sharex=True)
+    for ax, grp, colour, sign_label in (
+            (ax1, increased, RED, "more errors after weighting"),
+            (ax2, reduced, TEAL, "fewer errors after weighting")):
+        ax.set_axisbelow(True)
+        ax.grid(axis="x")
+        order = sorted(grp, key=lambda r: -abs(int(r["delta"])))
+        labels = [f"{r['true_class']} \u2192 {r['pred_class']}" for r in order]
+        deltas = [int(r["delta"]) for r in order]
+        ys = np.arange(len(order))
+        ax.barh(ys, deltas, color=colour, edgecolor="#333333", linewidth=0.6, zorder=3)
+        ax.axvline(0.0, color="#444444", lw=1.1, zorder=2)
+        for y, r in zip(ys, order):
+            d = int(r["delta"])
+            ax.annotate(f"{d:+d}   ({r['orig_count']} \u2192 {r['prop_count']})",
+                        (d, y), textcoords="offset points",
+                        xytext=(6 if d >= 0 else -6, 0), va="center",
+                        ha="left" if d >= 0 else "right", fontsize=9,
+                        weight="bold", color="#8E1F14" if d > 0 else "#0E5C5A")
+        ax.set_yticks(ys)
+        ax.set_yticklabels(labels, fontsize=10)
+        ax.set_title(("\u25b2  More errors after weighting" if colour == RED
+                      else "\u25bc  Fewer errors after weighting"), fontsize=10.5)
+    all_show = [int(r["delta"]) for r in increased] + [int(r["delta"]) for r in reduced]
+    ax1.set_xlim(min(all_show) * 1.5, max(all_show) * 1.7)
+    ax2.set_xlabel("change in test sample count (proposed \u2212 original;  "
+                   "negative = fewer, positive = more)")
+    dname = DATASET_NAME[dataset]
+    fig.suptitle(f"Ranked change in misclassifications vs ORIGINAL PROTEAN \u2014 "
+                 f"{dname}, \u03b1={alpha:g}, seed={seed}, test",
                  fontsize=12.5, weight="bold", y=1.04)
-    fig.text(0.5, -0.015,
-             "Only beta>0 conditions are shown; \u03b1=0.25 selected beta=0 so it carries no weighting effect. "
-             "Class support n\u2248 is the mean test support over seeds. Teal = F1 increased, red = F1 decreased. "
-             "X-IIoTID and 5G-NIDD class labels are never mixed. No significance testing (3 seeds).",
-             ha="center", va="top", fontsize=8.4, color="#444444", linespacing=1.5)
-    _save(fig, "figJ_per_class_original_vs_proposed")
+    fig.text(0.5, -0.02,
+             "Each bar is one off-diagonal (true \u2192 predicted) confusion cell; raw test counts over the same 243,177 samples. "
+             "Below zero = fewer such errors under the proposed selected-beta run; above zero = more. "
+             "The dominant error Benign \u2192 UDPFlood is ~unchanged (56,899 \u2192 57,054, +155 more) and remains the largest error. "
+             "Full difference matrices are in the supplemental figure. No statistical significance is inferred.",
+             ha="center", va="top", fontsize=8.6, color="#444444", linespacing=1.5)
+    _save(fig, f"figK_confusion_error_changes_{dataset}_a{alpha:g}_seed{seed}")
 
 
-# ---------------------------------------------------------------------------
-def figure_k(dataset="5gnidd", alpha=0.75, seed=0, beta=0.5):
+def figure_k_full(dataset="5gnidd", alpha=0.75, seed=0, beta=0.5):
+    """Supplemental full difference matrices (raw counts and row-rate)."""
     odir = RESULTS / f"{dataset}__alpha{alpha:g}__seed{seed}__original_protean"
     sdir = RESULTS / f"{dataset}__alpha{alpha:g}__seed{seed}__support_weighted__beta{beta:g}"
     names = json.loads((odir / "metadata.json").read_text())["class_names"]
@@ -187,42 +265,40 @@ def figure_k(dataset="5gnidd", alpha=0.75, seed=0, beta=0.5):
     for ax in axes:
         ax.set_xticks(range(len(names)))
         ax.set_yticks(range(len(names)))
-        ax.set_xticklabels(names, rotation=45, ha="right", fontsize=8)
-        ax.set_yticklabels(names, fontsize=8)
+        ax.set_xticklabels(names, rotation=45, ha="right", fontsize=9)
+        ax.set_yticklabels(names, fontsize=9)
         ax.set_xlabel("predicted class")
     axes[0].set_ylabel("true class")
-    axes[0].set_title("A. Difference in raw counts\n(proposed \u2212 original)")
-    axes[1].set_title("B. Difference in row-normalized rate\n(percentage points, proposed \u2212 original)")
-    # annotate the strongest cells on each panel
+    axes[0].set_title("A. Raw count difference\n(proposed \u2212 original)")
+    axes[1].set_title("B. Row-normalized rate difference\n(percentage points)")
     for ax, M, is_rate in ((axes[0], diff, False), (axes[1], diff_rate * 100, True)):
-        flat = [(abs(M[i, j]), i, j) for i in range(9) for j in range(9) if i != j]
+        flat = [(abs(M[i, j]), i, j) for i in range(len(names)) for j in range(len(names)) if i != j]
         for _, i, j in sorted(flat, reverse=True)[:6]:
             v = M[i, j]
             ax.text(j, i, f"{v:+.0f}" if not is_rate else f"{v:+.1f}",
-                    ha="center", va="center", fontsize=6.4, fontweight="bold",
-                    color="white" if abs(v) > (vmax_c if not is_rate else vmax_r) * 0.5
-                    else "#222222")
-    cb0 = fig.colorbar(im0, ax=axes[0], fraction=0.046, pad=0.02)
-    cb0.set_label("count delta")
-    cb1 = fig.colorbar(im1, ax=axes[1], fraction=0.046, pad=0.02)
-    cb1.set_label("row-rate delta (pp)")
-    dname = DATASET_NAME[dataset]
-    fig.suptitle(f"Confusion-matrix change: proposed (beta={beta:g}) minus original PROTEAN - "
-                 f"{dname}, alpha={alpha:g}, seed={seed}, test",
+                    ha="center", va="center", fontsize=6.6, fontweight="bold",
+                    color="white" if abs(v) > (vmax_c if not is_rate else vmax_r) * 0.5 else "#222222")
+    fig.colorbar(im0, ax=axes[0], fraction=0.046, pad=0.02).set_label("count delta")
+    fig.colorbar(im1, ax=axes[1], fraction=0.046, pad=0.02).set_label("row-rate delta (pp)")
+    fig.suptitle(f"Supplemental: full confusion-matrix change (proposed \u2212 original) \u2014 "
+                 f"{DATASET_NAME[dataset]}, \u03b1={alpha:g}, seed={seed}, test",
                  fontsize=12, weight="bold", y=1.05)
     fig.text(0.5, -0.02,
-             "Rows = true class, columns = predicted class; identical class order and sample counts (243,177) for both runs. "
-             "Diagonal: positive = more correct predictions; off-diagonal: positive = more errors of that (true\u2192predicted) type. "
-             "Red = positive delta (proposed has MORE samples in the cell); blue = negative delta (fewer). Values are descriptive only; no significance is inferred.",
-             ha="center", va="top", fontsize=8.4, color="#444444", linespacing=1.5)
+             "Rows = true class, columns = predicted class; identical class order and 243,177 samples. "
+             "Diagonal: positive = more correct predictions; off-diagonal: positive = more errors of that type. "
+             "Red = positive delta, blue = negative. See the ranked figure for the practical reading.",
+             ha="center", va="top", fontsize=8.6, color="#444444", linespacing=1.5)
     _save(fig, f"figK_confusion_delta_{dataset}_a{alpha:g}_seed{seed}")
 
 
 def main():
     _style()
-    figure_i()
-    figure_j()
-    figure_k()
+    figure_i("xiiotid")
+    figure_i("5gnidd")
+    figure_j("xiiotid")
+    figure_j("5gnidd")
+    figure_k_ranked()
+    figure_k_full()
 
 
 if __name__ == "__main__":
diff --git a/scripts/make_final_figures.py b/scripts/make_final_figures.py
index 9008406..1aaca7c 100644
--- a/scripts/make_final_figures.py
+++ b/scripts/make_final_figures.py
@@ -327,43 +327,69 @@ def figure_f() -> None:
     sel_e = np.array([float(master[c]["selected_rare_macro_f1_std"]) for c in conditions])
     betas = [float(master[c]["selected_beta"]) for c in conditions]
     delta = sel - orig
-
-    fig, ax = plt.subplots(figsize=(11.4, 6.0), constrained_layout=True)
-    ax.set_axisbelow(True)
-    ax.grid(axis="y")
     x = np.arange(len(conditions))
-    w = 0.36
-    ax.bar(x - w / 2, orig, w, yerr=orig_e, capsize=3.5, color=CHARCOAL,
-           edgecolor="#1A1A1A", linewidth=0.7, label="original_protean", zorder=3)
-    ax.bar(x + w / 2, sel, w, yerr=sel_e, capsize=3.5, color=TEAL,
-           edgecolor="#0A4F4E", linewidth=0.7, hatch="///",
-           label="support_weighted (validation-selected beta)", zorder=3)
+    w = 0.38
 
-    for xi, db, sb in zip(x, delta, betas):
-        if sb == 0.0:
-            colour = GREY
-        elif db > 0:
-            colour = GREEN
-        else:
-            colour = RED
-        sign = "+" if db >= 0 else "\u2212"
+    fig, (axA, axB) = plt.subplots(2, 1, figsize=(11.2, 8.6),
+                                   constrained_layout=True,
+                                   gridspec_kw={"height_ratios": [1.15, 1.0]})
+
+    # --- Panel A: absolute levels ---------------------------------------
+    axA.set_axisbelow(True)
+    axA.grid(axis="y")
+    axA.bar(x - w / 2, orig, w, yerr=orig_e, capsize=3.5, color=CHARCOAL,
+            edgecolor="#1A1A1A", linewidth=0.7, label="original_protean", zorder=3)
+    axA.bar(x + w / 2, sel, w, yerr=sel_e, capsize=3.5, color=TEAL,
+            edgecolor="#0A4F4E", linewidth=0.7, hatch="///",
+            label="proposed: support_weighted (selected beta)", zorder=3)
+    for xi, sb in zip(x, betas):
         top = max(orig[xi] + orig_e[xi], sel[xi] + sel_e[xi])
-        ax.annotate(f"{sign}{abs(db):.4f}", (xi, top + 0.012), ha="center",
-                    fontsize=9.5, weight="bold", color=colour)
-        ax.annotate(f"beta={sb:g}", (xi, top + 0.030), ha="center", fontsize=8.5,
-                    color="#444444")
-    ax.set_xticks(x)
-    ax.set_xticklabels(labels)
-    ax.set_ylabel("test rare-class macro-F1\n(mean over seeds {0,1,2})")
-    ax.set_ylim(0.5, max(max(orig + orig_e), max(sel + sel_e)) + 0.10)
-    ax.set_title("Test rare-class macro-F1: original PROTEAN vs validation-selected beta")
-    ax.legend(loc="upper left", frameon=True, edgecolor="#CCCCCC")
+        axA.annotate(f"beta={sb:g}", (xi, top + 0.006), ha="center",
+                     fontsize=9, color="#444444")
+    axA.set_xticks(x)
+    axA.set_xticklabels([])
+    axA.set_ylabel("test rare-class macro-F1\n(mean over seeds {0,1,2})")
+    axA.set_ylim(0.5, max(max(orig + orig_e), max(sel + sel_e)) + 0.075)
+    axA.set_title("A. Absolute test rare-class macro-F1")
+    axA.legend(loc="upper left", frameon=True, edgecolor="#CCCCCC")
+
+    # --- Panel B: signed difference -------------------------------------
+    axB.set_axisbelow(True)
+    axB.grid(axis="y")
+    cols = [GREY if sb == 0 else (GREEN if d > 0 else RED)
+            for d, sb in zip(delta, betas)]
+    axB.bar(x, delta, 0.55, color=cols, edgecolor="#333333", linewidth=0.6, zorder=3)
+    axB.axhline(0.0, color="#444444", lw=1.1, zorder=2)
+    for xi, d, sb in zip(x, delta, betas):
+        sign = "\u2212" if d < 0 else "+"
+        colr = "#1B7F3B" if (sb > 0 and d > 0) else ("#8E1F14" if (sb > 0 and d < 0) else "#555555")
+        axB.annotate(f"{sign}{abs(d):.4f}", (xi, d), textcoords="offset points",
+                     xytext=(0, 6 if d >= 0 else -14), ha="center", fontsize=10.5,
+                     weight="bold", color=colr)
+    axB.set_xticks(x)
+    axB.set_xticklabels(labels)
+    axB.set_ylabel("\u0394 rare-class macro-F1\n(proposed \u2212 original)")
+    m = max(abs(delta)) * 1.65
+    axB.set_ylim(-m, m)
+    axB.set_title("B. Signed difference (proposed \u2212 original)")
+    handles = [plt.Rectangle((0, 0), 1, 1, facecolor=GREEN, edgecolor="#333"),
+               plt.Rectangle((0, 0), 1, 1, facecolor=RED, edgecolor="#333"),
+               plt.Rectangle((0, 0), 1, 1, facecolor=GREY, edgecolor="#333")]
+    axB.legend(handles, ["positive (3 conditions)", "negative (1 condition)",
+                         "selected beta = 0 / ~0 (2 conditions)"],
+               title="signed delta", loc="upper right", frameon=True,
+               edgecolor="#CCCCCC", ncols=1)
+
+    fig.suptitle("Original PROTEAN vs proposed selected-beta: test rare-class macro-F1",
+                 fontsize=13, weight="bold", y=1.02)
     _footnote(fig,
-              "Bar height = mean test rare-class macro-F1 over seeds {0,1,2}; error bars = \u00b11 sample std across seeds.\n"
-              "Delta (top) = selected-beta minus original_protean; green = positive, red = negative, grey = selected beta 0\n"
-              "(floating-point ~0). No statistical significance test is applied (3 seeds). Gains are condition-dependent;\n"
+              "Panel A: bar height = mean test rare-class macro-F1 over seeds {0,1,2}; error bars = \u00b11 sample std across seeds "
+              "(the levels are close by design, so panel B is the informative view).\n"
+              "Panel B: delta = proposed \u2212 original; the exact value is printed at each bar. Green = positive (X-IIoTID a0.50, "
+              "X-IIoTID a0.75, 5G-NIDD a0.75); red = negative (5G-NIDD a0.50); grey = selected beta 0 (both a0.25, floating-point ~0).\n"
+              "No statistical significance test is applied (3 seeds). A positive mean delta does not establish significance; "
               "the proposed weighting does NOT consistently outperform original PROTEAN.",
-              y=-0.13)
+              y=-0.10)
     _save(fig, "figF_selected_vs_original_test_rare_macro_f1")
 
 
@@ -458,15 +484,34 @@ def figure_h(dataset="5gnidd", alpha=0.75, seed=0, selected_beta=0.5) -> bool:
 
     class_names = meta["class_names"]
     mats = {k: np.array(json.loads(p.read_text())) for k, p in paths.items()}
+    o, s = mats["original_protean"], mats["selected_beta"]
+    diff = s - o
+    idx = {n: i for i, n in enumerate(class_names)}
+
+    # Computed summary of the most important changes (never hard-coded).
+    off = [(int(diff[i, j]), i, j) for i in range(len(class_names))
+           for j in range(len(class_names)) if i != j]
+    reduced = sorted(off, key=lambda t: t[0])[:2]
+    increased = sorted(off, key=lambda t: -t[0])[:2]
+    diag = sorted(range(len(class_names)), key=lambda i: -abs(int(diff[i, i])))[:3]
+    lines = ["Most important ranked changes (proposed \u2212 original):"]
+    for d, i, j in reduced:
+        lines.append(f"  \u2022 {class_names[i]} \u2192 {class_names[j]}: "
+                     f"{int(o[i, j])} \u2192 {int(s[i, j])}  ({d} fewer)")
+    for d, i, j in increased:
+        note = "  \u2190 dominant residual error; NOT an improvement" \
+            if (class_names[i], class_names[j]) == ("Benign", "UDPFlood") else ""
+        lines.append(f"  \u2022 {class_names[i]} \u2192 {class_names[j]}: "
+                     f"{int(o[i, j])} \u2192 {int(s[i, j])}  (+{d} more){note}")
+    lines.append("  \u2022 diagonal (correct): " + ", ".join(
+        f"{class_names[i]} {int(diff[i, i]):+d}" for i in diag))
 
     from matplotlib.colors import LogNorm
     vmax = max(m.max() for m in mats.values())
     vmin = 1
-    fig, axes = plt.subplots(1, 2, figsize=(13.2, 6.2), constrained_layout=True)
-    titles = {
-        "original_protean": "ORIGINAL PROTEAN",
-        "selected_beta": f"SELECTED-BETA PROTEAN (beta={selected_beta:g})",
-    }
+    fig, axes = plt.subplots(1, 2, figsize=(13.6, 6.0), constrained_layout=True)
+    titles = {"original_protean": "LEFT: ORIGINAL PROTEAN",
+              "selected_beta": f"RIGHT: BETA-TUNED PROTEAN (beta={selected_beta:g})"}
     im = None
     for ax, key in zip(axes, ["original_protean", "selected_beta"]):
         m = mats[key]
@@ -474,8 +519,8 @@ def figure_h(dataset="5gnidd", alpha=0.75, seed=0, selected_beta=0.5) -> bool:
                        aspect="equal")
         ax.set_xticks(range(len(class_names)))
         ax.set_yticks(range(len(class_names)))
-        ax.set_xticklabels(class_names, rotation=45, ha="right", fontsize=8.2)
-        ax.set_yticklabels(class_names, fontsize=8.2)
+        ax.set_xticklabels(class_names, rotation=45, ha="right", fontsize=9)
+        ax.set_yticklabels(class_names, fontsize=9)
         ax.set_xlabel("predicted class")
         ax.set_title(titles[key], fontsize=11)
         thr = m.max() * 0.06
@@ -485,7 +530,7 @@ def figure_h(dataset="5gnidd", alpha=0.75, seed=0, selected_beta=0.5) -> bool:
                 if v == 0:
                     continue
                 txt = f"{v/1000:.1f}k" if v >= 10000 else str(v)
-                ax.text(j, i, txt, ha="center", va="center", fontsize=6.4,
+                ax.text(j, i, txt, ha="center", va="center", fontsize=6.8,
                         color="white" if m[i, j] > thr else "#333333")
     axes[0].set_ylabel("true class")
     cbar = fig.colorbar(im, ax=axes, fraction=0.025, pad=0.02)
@@ -494,11 +539,13 @@ def figure_h(dataset="5gnidd", alpha=0.75, seed=0, selected_beta=0.5) -> bool:
     fig.suptitle(
         f"Paired test confusion matrices - {dname}, alpha={alpha:g}, seed={seed} "
         f"(same manifest, split, seed, class order)",
-        fontsize=11.5, weight="bold", y=1.05)
+        fontsize=12, weight="bold", y=1.05)
     _footnote(fig,
-              "Left = original_protean; right = support_weighted at the validation-selected beta. Raw counts; identical class order and colour scale.\n"
-              "Dominant off-diagonal error: Benign predicted as UDPFlood, essentially unchanged by beta. No statistical significance is inferred.",
-              y=-0.20)
+              "Rows = actual class, columns = predicted class; diagonal = correct, off-diagonal = misclassified. Raw counts, "
+              "identical class order and colour scale; both matrices contain the same 243,177 test samples.\n"
+              + "\n".join(lines) +
+              "\nNo statistical significance is inferred from the visual difference.",
+              y=-0.13)
     _save(fig, f"figH_confusion_matrix_{dataset}_a{alpha:g}_seed{seed}")
     return True