research: finalize experimental synthesis

docs/handoff/SESSION_HANDOFF.md

@@ -3,30 +3,36 @@
 This is the canonical continuation point for the next agent session. It is written to be
 readable without prior chat history, terminal scrollback, or git history.
 
-Last updated: 2026-09-30 (Session 11). Session 11 type: **final 5G-NIDD beta-validation condition**
-(`alpha=0.25`), completing the planned experiment grid. Sessions 6–8 completed the X-IIoTID
-three-alpha study (`alpha=0.25/0.50/0.75`, seeds `{0,1,2}`); Session 9 ran 5G-NIDD `alpha=0.50`;
-Session 10 ran 5G-NIDD `alpha=0.75`; Session 11 ran **only** 5G-NIDD `alpha=0.25` for seeds
-`{0,1,2}` across the five frozen beta candidates plus `original_protean`, using the same
-pooled-across-seed validation-only selection rule. **No condition was run after alpha=0.25.** No
-commit or push was made.
+Last updated: 2026-09-30 (Session 12). Session 12 type: **thesis synthesis of the complete audited
+grid** — no training, no partition regeneration, no source/model/aggregation change. Session 11 ran
+the final 5G-NIDD condition (`alpha=0.25`), completing the planned grid (108 full runs + 1 smoke).
 
-> **Audit status (Session 11).** Independent audit verdict: **PASS**
-> (`docs/research/SESSION11_AUDIT.md`). `beta_star(5gnidd, alpha=0.25) = 0.00` confirmed from
-> validation only; run inventory 18/18; pairing hash-identical and reproduced from raw data;
-> rare/unseen 0 mismatches over 1,620 cells; leakage audit PASS; `beta=0` class **B**; source/data
-> integrity PASS; pytest 75 passed. **Session 11 is safe to checkpoint.**
+> **Session 12 scope and status.** Consolidated the 108-run grid into authoritative six-condition
+> tables, publication-ready figures, a final RQ1/RQ2/RQ3 synthesis, a Chapter-4 blueprint and a
+> claims register, then independently re-verified every number against the audited artifacts.
+> **No material scientific gap remains for the defined thesis scope (classification A).** No
+> training was executed, no tracked file changed, `planningawal/` and both raw datasets are
+> unchanged, `results/` is unchanged (108 non-smoke run dirs), and pytest is **75 passed**.
+> Deliverables: `docs/research/FINAL_EXPERIMENTAL_SYNTHESIS.md`,
+> `docs/thesis/CHAPTER4_RESULTS_OUTLINE.md`, `docs/thesis/CLAIMS_REGISTER.md`,
+> `artifacts/analysis/final__*.{csv,json}`, `artifacts/figures/final/figA–G*.png`.
 
-> **Headline (Session 11).** PRIMARY pooled-validation selected beta for 5G-NIDD `alpha=0.25` =
-> **0.00** (validation only; per-seed diagnostics 0.25 / 0.00 / 0.00). The pooled validation curve
-> declines from 0.761 (beta=0) to 0.722 (beta=0.75) and rises slightly to 0.738 (beta=1.0), with
-> the between-beta spread (0.039) exceeding the across-seed std (0.003–0.022). This is the first
-> condition where the selected beta is 0, i.e. the equal-`S_j` mean; the selected mode is therefore
-> numerically the baseline and shows no support-weighting benefit. `beta=0` is class **B** for all
-> three seeds. Realized heterogeneity is the highest of the three 5G-NIDD conditions
-> (16/27 `|S_j|<M` cells, 30 missing cells, client ratio mean 17.60, mean CV 1.302), and the
-> rare-class union differs across seeds (6/6/8), so this condition's test rare support is not
-> constant.
+> **Final RQ findings (Session 12).** (1) Changing beta changes contribution weights and prototype
+> positions; for a fixed snapshot displacement is monotone increasing in beta in all six
+> conditions, and realized support imbalance and displacement co-vary within each dataset. (2) The
+> pooled validation rare-class response is **not monotone** in beta and is weak relative to seed
+> variance; the direct aggregation effect and the end-to-end training effect remain distinct.
+> (3) Selected betas are condition-dependent and non-monotone in alpha:
+> X-IIoTID `{0.00, 0.75, 0.50}` and 5G-NIDD `{0.00, 1.00, 0.50}` for alpha `{0.25, 0.50, 0.75}`;
+> both datasets select beta=0 at alpha=0.25 (no weighting benefit) and beta=0.50 at alpha=0.75;
+> no universal/global optimum. All six top-two selection gaps are smaller than the selected beta's
+> across-seed std. No inferential test was added (3 seeds).
+
+> **Session 11 audit (preserved).** Verdict **PASS** (`docs/research/SESSION11_AUDIT.md`);
+> `beta_star(5gnidd, alpha=0.25) = 0.00` validation-only (per-seed 0.25/0.00/0.00); run inventory
+> 18/18; pairing hash-identical; rare/unseen 0 mismatches; leakage PASS; `beta=0` class **B**;
+> source/data integrity PASS; pytest 75 passed. Realized heterogeneity highest of the three 5G-NIDD
+> conditions (16/27 cells, 30 missing, client ratio 17.60, mean CV 1.302).
 
 ---
 
@@ -304,11 +310,13 @@ Six-condition, validation-only pooled selections:
 
 ## 15. Git status
 
-No commit or push was made. `HEAD = bbd35085063194db58e4472e70b204d4be4a3607`, branch `main`, in
-sync with `origin/main`. Only untracked docs/analysis artifacts exist (no tracked modification);
-`results/` is gitignored. `planningawal/` and both raw datasets are unchanged. `equal_available`
-absent. No test-based selection. The 18 runs are the only `results/5gnidd__alpha0.25__*`
-directories.
+No commit or push was made in Session 12. `HEAD = d30e9831e6c17fc6b28383ae4592070ccd83657e`,
+branch `main`, in sync with `origin/main`. `git diff` / `git diff --cached` are empty (no tracked
+modification); only untracked docs/analysis/figures artifacts exist. `planningawal/` and both raw
+datasets are unchanged (SHA-256 match). `src/`, `tests/`, `configs/`, `scripts/`, `pyproject.toml`
+are unchanged. `results/` is gitignored and holds exactly 108 non-smoke run directories plus 1
+`smoke`. `equal_available` absent. No test-based selection. `matplotlib 3.11.2` was installed into
+the git-ignored `.venv` only; no dependency file was modified.
 
 ## 16. Session 11 audit result
 
@@ -321,45 +329,51 @@ directories.
 - Direct-effect, client-level, cross-alpha, cross-dataset labels: PASS.
 - Source/data integrity PASS. pytest: **75 passed, 0 failed, 0 skipped**.
 
-## 17. Files created / modified this session
+## 17. Files created / modified (Session 12)
 
 Created:
 
-- `docs/research/5GNIDD_ALPHA025_BETA_VALIDATION.md`
-- `docs/research/5GNIDD_CROSS_ALPHA_ANALYSIS.md`
-- `docs/research/CROSS_DATASET_BETA_ANALYSIS.md`
-- `docs/research/SESSION11_AUDIT.md`
-- `artifacts/analysis/5gnidd__alpha0.25__*` — 18 prototype snapshots + 3 per-seed
-  effect/weights CSVs + `5gnidd__alpha0.25__multiseed__*` (validation per-seed/pooled,
-  client detail/summary/deltas, test descriptive, prototype effect pooled, beta0 equivalence,
-  runtime, selection summary, audit recomputation) + `5gnidd__alpha0.25__partition_characterization.csv`
-- `artifacts/analysis/5gnidd__cross_alpha__*` (regenerated to include alpha=0.25)
-- `artifacts/analysis/cross_dataset__matched_alpha_summary.{csv,json}`,
-  `artifacts/analysis/cross_dataset__beta_curve.csv`
-- `results/5gnidd__alpha0.25__seed{0,1,2}__*` (18 run dirs, gitignored)
+- `docs/research/FINAL_EXPERIMENTAL_SYNTHESIS.md` — final six-condition synthesis, RQ1/RQ2/RQ3
+  answers, threats to validity, contribution, unsupported claims, gap assessment.
+- `docs/thesis/CHAPTER4_RESULTS_OUTLINE.md` — precise Chapter-4 blueprint (purpose / artifact /
+  key numbers / allowed / avoided per subsection).
+- `docs/thesis/CLAIMS_REGISTER.md` — claim / evidence / status / limitation / thesis-safe wording.
+- `artifacts/analysis/final__six_condition_master.{csv,json}`
+- `artifacts/analysis/final__pooled_validation_curves.csv` (30 rows)
+- `artifacts/analysis/final__realized_heterogeneity.csv`
+- `artifacts/analysis/final__prototype_displacement_by_beta.csv`
+- `artifacts/analysis/final__prototype_displacement_by_class.csv`
+- `artifacts/analysis/final__client_effect_summary.csv` (90 rows)
+- `artifacts/analysis/final__selected_vs_original.csv`
+- `artifacts/analysis/final__selection_strength.csv`
+- `artifacts/figures/final/figA–G*.png` — 8 thesis-friendly figures (matplotlib only).
 
 Modified:
 
-- `docs/handoff/SESSION_HANDOFF.md` (this file)
-- `artifacts/analysis/5gnidd__cross_alpha__*` (extended from 2 to 3 alphas; previous 2-alpha
-  content was superseded, not silently lost — it is reproducible from the frozen manifests)
+- `docs/handoff/SESSION_HANDOFF.md` (this file).
 
-No source or test code was changed.
+No source, test, config, model or aggregation file was changed. The synthesis generator is kept
+outside the repository at `/tmp/opencode/s12/synthesize_final_analysis.py` (Session-11 convention
+for ad-hoc analysis helpers). `matplotlib 3.11.2` was installed into the git-ignored `.venv`.
 
 ## 18. Exact recommended next phase
 
-1. **Session 11 is audited and safe to checkpoint** (verdict PASS). Before checkpointing, stage
-   only the intended docs/analysis artifacts; do not commit `results/` or the raw datasets.
-2. **The planned experiment grid is COMPLETE** (two datasets × three alphas × three seeds × five
-   betas + baseline = 108 runs). Do **not** run any new condition.
-3. Remaining work is analysis/writing, not new training:
-   - optionally add an inferential comparison (e.g. paired sign/permutation test over seeds) to
-     formalise the weak selection effects — clearly labelled as post-hoc if added;
-   - freeze the final thesis tables/report from the six-condition artifacts;
-   - decide and document the `beta=0` vs `original_protean` reporting convention.
-4. Do not infer a monotone alpha→beta law; do not claim dataset causality or superiority.
+1. **The experiment grid is COMPLETE** (2 datasets × 3 alphas × 3 seeds × (5 betas + baseline) =
+   108 runs). Do **not** run any new condition; no new training is required (gap classification A).
+2. **Synthesis is complete and verified.** All six selections, selected/original test summaries,
+   heterogeneity values, displacement summaries and client counts were independently recomputed
+   from raw artifacts with zero mismatches.
+3. Remaining work is **writing and formatting**, not experiments:
+   - draft Chapter 4 from `docs/thesis/CHAPTER4_RESULTS_OUTLINE.md` using the `final__*` tables
+     and `figA–G` figures, respecting `docs/thesis/CLAIMS_REGISTER.md`;
+   - optionally add appendices with the full per-seed and per-class tables;
+   - if the examiners require inferential support, add a clearly post-hoc, explicitly
+     underpowered analysis and label it as such (not done by default).
+4. Do not infer a monotone alpha→beta law; do not claim dataset causality, superiority,
+   generalization, or a global optimum.
 5. Update `EXPERIMENT_PROTOCOL.md` only if a real ambiguity appears; update this handoff.
-6. Do not commit or push unless asked.
+6. Do not commit or push unless asked. When checkpointing, stage only intended docs/analysis/
+   figure artifacts; never commit `results/` or raw datasets.
 
 ## 19. Important commands
 
@@ -394,16 +408,15 @@ No source or test code was changed.
 ## 20. Recommended next-session prompt
 
 ```
-You are continuing the perdalam-protean research repository (Session 12).
+You are continuing the perdalam-protean research repository (Session 13).
 
 FIRST, read completely:
   1. docs/handoff/SESSION_HANDOFF.md
-  2. docs/research/SESSION11_AUDIT.md
-  3. docs/research/5GNIDD_ALPHA025_BETA_VALIDATION.md
-  4. docs/research/5GNIDD_CROSS_ALPHA_ANALYSIS.md
+  2. docs/research/FINAL_EXPERIMENTAL_SYNTHESIS.md
+  3. docs/thesis/CHAPTER4_RESULTS_OUTLINE.md
+  4. docs/thesis/CLAIMS_REGISTER.md
   5. docs/research/CROSS_DATASET_BETA_ANALYSIS.md
-  6. docs/research/XIIOTID_CROSS_ALPHA_ANALYSIS.md
-  7. docs/research/EXPERIMENT_PROTOCOL.md
+  6. docs/research/EXPERIMENT_PROTOCOL.md
 
 Critical rules:
   - planningawal/ is immutable.
@@ -412,14 +425,16 @@ Critical rules:
   - Do NOT add an executable equal_available mode.
   - Beta selection uses validation only; PRIMARY is pooled across clients and seeds.
   - The planned 2-dataset x 3-alpha experiment grid is COMPLETE. Do NOT run new conditions.
+  - Scientific gap classification is A (NO MATERIAL GAP); no new experiment is required.
 
-Context: Session 11 ran the final condition 5G-NIDD alpha=0.25 (18 runs, ~19.11 h), audited PASS,
-and selected beta_star=0.00 (validation only; per-seed 0.25/0.00/0.00). The six-condition pooled
-selections are 5G-NIDD 0.25/0.50/0.75 = 0.00/1.00/0.50 and X-IIoTID = 0.00/0.75/0.50; no universal
-or monotone beta. beta=0 equivalence class B. pytest 75 passed.
+Context: Session 12 completed the thesis synthesis with no training. Six pooled selections:
+X-IIoTID 0.25/0.50/0.75 = 0.00/0.75/0.50 and 5G-NIDD = 0.00/1.00/0.50; no universal/monotone beta;
+all top-two gaps below selected-beta across-seed std; beta=0 equivalence class B everywhere.
+Authoritative artifacts: artifacts/analysis/final__*.{csv,json}; figures artifacts/figures/final/.
+All synthesis numbers independently verified; pytest 75 passed; no tracked file changed.
 
-Task: no new training. Consolidate the final thesis-ready six-condition tables/report from the
-existing audited artifacts, optionally add a clearly post-hoc inferential comparison of the weak
-pooled selections, and keep all limitations (S_j assumption, three seeds, no global optimum)
-explicit. Use .venv/bin/python. Do not commit or push unless asked.
+Task: write/format thesis chapters from the existing synthesis. Use the final__* tables and
+figA-G figures, obey CLAIMS_REGISTER.md wording, keep all limitations explicit, and do not run
+training or add inferential tests by default. Use .venv/bin/python. Do not commit or push unless
+asked.
 ```

docs/research/FINAL_EXPERIMENTAL_SYNTHESIS.md

@@ -0,0 +1,491 @@
+# Final Experimental Synthesis
+
+Session 12 synthesis of the complete PROTEAN support-weighted prototype aggregation study.
+This document is **read-only with respect to experiments**: every number is loaded from the
+audited machine-readable artifacts under `artifacts/analysis/`. No training was run, no
+partition was regenerated, no source/model/aggregation code was modified, and no beta was
+re-selected from test metrics.
+
+- Repository HEAD at synthesis: `d30e9831e6c17fc6b28383ae4592070ccd83657e` (branch `main`, in sync with `origin/main`).
+- Experimental grid: **2 datasets × 3 alpha levels × 3 seeds × (5 betas + original baseline) = 108 full runs** (plus one unrelated `smoke` directory, excluded).
+- Tests: `75 passed, 0 failed, 0 skipped`.
+- Beta candidates (frozen): `{0.00, 0.25, 0.50, 0.75, 1.00}`.
+- Selection rule (frozen): maximize **pooled validation per-client-mean rare-class macro-F1**, validation only; per-seed selection is diagnostic only; test is never used for selection.
+- `planningawal/` and both raw datasets unchanged (re-verified).
+
+Terminology used throughout:
+
+- **Measured**: directly read or recomputed from an audited artifact.
+- **Descriptive interpretation**: a neutral reading of measured values.
+- **Unsupported / not established**: a claim that the design cannot defend (causal, universal, superiority).
+
+---
+
+## 1. Scope and protocol
+
+The thesis modifies **only** the server-side prototype aggregation of PROTEAN (Chennoufi et al.,
+2025). Each contributing client `i` provides a class-`j` local prototype `C_ij` (local class
+embedding mean) and a class-specific training support `N_ij`. The candidate aggregation is
+
+```
+Cbar_j(beta) = ( sum_{i in S_j} N_ij^beta * C_ij ) / ( sum_{i in S_j} N_ij^beta )
+S_j = { i | N_ij > 0 and a valid local prototype C_ij exists }
+```
+
+`S_j` is an explicit **reproduction assumption** (no verified official PROTEAN code is
+available); `N_ij` is class-specific **sample support, not prototype quality**. The frozen
+`original_protean` mode uses the equal-mean reference. `beta=0` is arithmetically the equal mean
+over the shared `S_j` but is **not bitwise identical** to the baseline (floating-point
+divergence; equivalence class **B** in every condition).
+
+Protocol (frozen, `docs/research/EXPERIMENT_PROTOCOL.md`): datasets X-IIoTID (`class2`,
+benign `Normal`, K=9) and 5G-NIDD (`Attack Type`, benign `Benign`, K=9); 10 clients; symmetric
+Dirichlet non-IID with `alpha ∈ {0.25, 0.50, 0.75}`; seeds `{0,1,2}`; stratified train/validation/test
+= 70/10/20 (validation is a project decision; the paper uses 80/20); 10 rounds; 3 local epochs;
+batch 256; Adam; `lr=1e-3`, `mu=0.1`, `lambda=1.0` (project decisions). All compared configurations
+within a `(dataset, alpha, seed)` reuse exactly the same manifest, split and initialization
+(verified by content and partition hashes).
+
+---
+
+## 2. Completed experimental grid
+
+| dataset | alpha | seeds | betas + baseline | runs | audit |
+|---|---|---|---|---|---|
+| X-IIoTID | 0.25 | 0,1,2 | 5 + 1 | 18 | PASS WITH MINOR CORRECTIONS (Session 7) |
+| X-IIoTID | 0.50 | 0,1,2 | 5 + 1 | 18 | PASS WITH MINOR CORRECTIONS (Session 6) |
+| X-IIoTID | 0.75 | 0,1,2 | 5 + 1 | 18 | PASS (Session 8) |
+| 5G-NIDD | 0.25 | 0,1,2 | 5 + 1 | 18 | PASS (Session 11) |
+| 5G-NIDD | 0.50 | 0,1,2 | 5 + 1 | 18 | PASS (Session 9) |
+| 5G-NIDD | 0.75 | 0,1,2 | 5 + 1 | 18 | PASS (Session 10) |
+| **total** | | | | **108** | |
+
+---
+
+## 3. Six-condition master result
+
+Authoritative table: `artifacts/analysis/final__six_condition_master.csv` / `.json`.
+
+| dataset | alpha | beta* | pooled val (std) | sel test rare-F1 | orig test rare-F1 | sel rare-recall | orig rare-recall | sel macro-F1 | sel acc | sel wF1 |
+|---|---|---|---|---|---|---|---|---|---|---|
+| X-IIoTID | 0.25 | **0.00** | 0.5917 (0.0895) | 0.6321 | 0.6328 | 0.8213 | 0.8220 | 0.6813 | 0.8740 | 0.9069 |
+| X-IIoTID | 0.50 | **0.75** | 0.5794 (0.0751) | 0.6655 | 0.6425 | 0.8626 | 0.8770 | 0.7312 | 0.9064 | 0.9336 |
+| X-IIoTID | 0.75 | **0.50** | 0.4767 (0.0466) | 0.6052 | 0.6039 | 0.8488 | 0.8565 | 0.7576 | 0.9281 | 0.9452 |
+| 5G-NIDD | 0.25 | **0.00** | 0.7615 (0.0219) | 0.8412 | 0.8423 | 0.9612 | 0.9606 | 0.8166 | 0.7302 | 0.7045 |
+| 5G-NIDD | 0.50 | **1.00** | 0.6924 (0.1391) | 0.8720 | 0.8805 | 0.9671 | 0.9729 | 0.8386 | 0.7433 | 0.7182 |
+| 5G-NIDD | 0.75 | **0.50** | 0.8091 (0.0587) | 0.8967 | 0.8799 | 0.9597 | 0.9612 | 0.8531 | 0.7523 | 0.7275 |
+
+All selected betas are validation-only; `test_used_for_selection = false` in every run;
+`beta0_equivalence_class = B` in every condition. The selected beta takes values
+`{0.00, 0.50, 1.00}` across the six conditions: **not universal, not monotone, not a function of
+alpha alone**.
+
+Selected-beta test metrics are **post-selection descriptive summaries**, not selection inputs.
+
+---
+
+## 4. Realized non-IID characteristics
+
+Dirichlet `alpha` is *not* a realized-heterogeneity measure. Authoritative realized table:
+`artifacts/analysis/final__realized_heterogeneity.csv`.
+
+| dataset | alpha | `|S_j|<M` / 27 | missing client-class cells | client-size ratio (mean / max) | mean support CV | max support CV | max per-class max/min |
+|---|---|---|---|---|---|---|---|
+| X-IIoTID | 0.25 | 19 | 38 | 14.75 / 20.84 | 1.442 | 2.134 | 122,800 |
+| X-IIoTID | 0.50 | 7 | 9 | 7.41 / 9.89 | 1.183 | 2.095 | 12,215 |
+| X-IIoTID | 0.75 | 3 | 4 | 4.67 / 5.07 | 0.881 | 1.322 | 1,903 |
+| 5G-NIDD | 0.25 | 16 | 30 | 17.60 / 27.83 | 1.302 | 2.139 | 185,504 |
+| 5G-NIDD | 0.50 | 4 | 4 | 9.59 / 12.72 | 1.225 | 1.964 | 88,943 |
+| 5G-NIDD | 0.75 | 4 | 5 | 6.31 / 11.36 | 0.916 | 1.265 | 7,642.5 |
+
+Measured facts:
+
+- Realized support imbalance, missing-cell frequency and client-size ratio **decrease as alpha
+  rises** within each dataset (alpha orders heterogeneity only on average).
+- Per-seed realized imbalance is **not monotone** even at fixed alpha: for 5G-NIDD alpha=0.50,
+  seed 0 has client-size ratio 11.63 while alpha=0.25 seed 0 has 9.93.
+- All 10 clients are eligible in every condition (exactly two rare attack classes each).
+- Rare-class **union differs across seeds** in several conditions (e.g. 5G-NIDD alpha=0.25:
+  union rare 6/6/8, union unseen 6/5/5), so test rare support is not constant across seeds and
+  absolute cross-seed metric levels are not directly comparable.
+- The `S_j` reproduction assumption is materially exercised in 19/27 (X-IIoTID alpha=0.25),
+  16/27 (5G-NIDD alpha=0.25) and ≤7/27 cells elsewhere.
+
+**Interpretation**: `alpha` is a design knob; the realized partition (support CV, missing cells,
+client-size ratio) is the variable that actually differs across conditions.
+
+---
+
+## 5. Pooled beta-selection behavior
+
+Authoritative curve table: `artifacts/analysis/final__pooled_validation_curves.csv` (30 rows).
+Figure: `artifacts/figures/final/figA_*_pooled_validation_beta_curve.png`.
+
+Pooled validation rare-class macro-F1 (selected in bold):
+
+| beta | X-IIoTID a0.25 | X-IIoTID a0.50 | X-IIoTID a0.75 | 5G-NIDD a0.25 | 5G-NIDD a0.50 | 5G-NIDD a0.75 |
+|---|---|---|---|---|---|---|
+| 0.00 | **0.5917** | 0.5444 | 0.4717 | **0.7615** | 0.6872 | 0.7781 |
+| 0.25 | 0.5887 | 0.5554 | 0.4744 | 0.7493 | 0.6451 | 0.7905 |
+| 0.50 | 0.5859 | 0.5673 | **0.4767** | 0.7257 | 0.6747 | **0.8091** |
+| 0.75 | 0.5795 | **0.5794** | 0.4725 | 0.7222 | 0.6884 | 0.7915 |
+| 1.00 | 0.5321 | 0.5735 | 0.4686 | 0.7376 | **0.6924** | 0.8043 |
+
+Measured facts:
+
+- Response is **not monotone** in beta in any condition:
+  X-IIoTID alpha=0.25 decreases monotonically (0.5917→0.5321);
+  X-IIoTID alpha=0.50 peaks at 0.75; X-IIoTID alpha=0.75 peaks at 0.50;
+  5G-NIDD alpha=0.25 falls to a minimum at 0.75 then rises at 1.00;
+  5G-NIDD alpha=0.50 peaks at 1.00; 5G-NIDD alpha=0.75 peaks at 0.50.
+- Between-beta spreads (0.008 – 0.060) are of the same order as, or smaller than, across-seed
+  sample std (0.003 – 0.139). See section 13.
+
+---
+
+## 6. Selected-beta test results (descriptive only)
+
+Table: `artifacts/analysis/final__selected_vs_original.csv`.
+Figure: `artifacts/figures/final/figF_selected_vs_original_test_rare_macro_f1.png`.
+
+Descriptive difference `selected beta − original_protean` (mean over 3 seeds), **not** a test of
+statistical superiority:
+
+| dataset | alpha | beta* | Δ rare-F1 | Δ rare-recall | Δ macro-F1 | Δ accuracy | Δ wF1 |
+|---|---|---|---|---|---|---|---|
+| X-IIoTID | 0.25 | 0.00 | −0.0007 | −0.0007 | −0.0007 | −0.0014 | −0.0008 |
+| X-IIoTID | 0.50 | 0.75 | **+0.0230** | −0.0144 | +0.0222 | +0.0069 | +0.0089 |
+| X-IIoTID | 0.75 | 0.50 | +0.0014 | −0.0077 | +0.0058 | +0.0075 | +0.0061 |
+| 5G-NIDD | 0.25 | 0.00 | −0.0010 | +0.0007 | −0.0028 | −0.0007 | −0.0010 |
+| 5G-NIDD | 0.50 | 1.00 | −0.0086 | −0.0058 | −0.0038 | −0.0007 | −0.0005 |
+| 5G-NIDD | 0.75 | 0.50 | +0.0168 | −0.0015 | +0.0117 | +0.0013 | +0.0014 |
+
+Measured facts:
+
+- The largest positive descriptive rare-F1 difference (+0.0230, X-IIoTID alpha=0.50) is still
+  smaller than that condition's between-seed sample std (0.0267) and than the original_protean
+  cross-seed std (0.0287).
+- At both alpha=0.25 conditions the selected beta is 0, so the "selected" run **is** the beta=0
+  run; the tiny differences from `original_protean` are floating-point divergence (class B), not
+  a weighting effect.
+- Selected rare-recall is ≤ original rare-recall in 5 of 6 conditions: emphasis on high-support
+  classes trades a little rare-class recall for rare-class precision.
+- 5G-NIDD test weighted-F1 sits below accuracy (hard, high-support Benign/UDPFlood); X-IIoTID
+  weighted-F1 exceeds accuracy. This is a dataset characteristic, not a beta effect.
+
+---
+
+## 7. Direct prototype displacement
+
+Authoritative tables: `final__prototype_displacement_by_beta.csv`,
+`final__prototype_displacement_by_class.csv`.
+Figures: `figD_prototype_displacement_beta0_vs_beta1.png`,
+`figE_support_cv_vs_displacement.png`.
+
+Mean `||Cbar_j(beta) − Cbar_j(0)||` across classes, fixed round-0 snapshots (no retraining):
+
+| dataset | alpha | β=0.25 | β=0.50 | β=0.75 | β=1.00 |
+|---|---|---|---|---|---|
+| X-IIoTID | 0.25 | 1.476 | 2.403 | 3.003 | **3.422** |
+| X-IIoTID | 0.50 | 0.826 | 1.374 | 1.780 | **2.099** |
+| X-IIoTID | 0.75 | 0.477 | 0.829 | 1.097 | **1.305** |
+| 5G-NIDD | 0.25 | 1.295 | 2.055 | 2.559 | **2.919** |
+| 5G-NIDD | 0.50 | 0.953 | 1.567 | 2.003 | **2.340** |
+| 5G-NIDD | 0.75 | 0.497 | 0.875 | 1.163 | **1.385** |
+
+Measured facts:
+
+- For a fixed snapshot, displacement is **monotone increasing in beta** in all six conditions
+  (mechanically expected: larger beta shifts weight toward high-`N_ij` clients).
+- Displacement at beta=1 **decreases as alpha rises** in both datasets, tracking realized
+  support imbalance.
+- Displacement is a **representation-aggregation** metric only; it is not detection quality.
+
+---
+
+## 8. Client-level effects
+
+Authoritative table: `artifacts/analysis/final__client_effect_summary.csv` (90 rows:
+6 conditions × 3 seeds × 5 betas, `delta` = validation per-client rare-class macro-F1 relative
+to beta=0). Figure: `figG_client_effect_at_selected_beta.png`.
+
+Aggregate at **beta=1.00** (mean of per-client mean delta across 3 seeds):
+
+| dataset | alpha | mean Δ | improve / degrade / unchanged (clients, summed over seeds) |
+|---|---|---|---|
+| X-IIoTID | 0.25 | **−0.0596** | 12 / 15 / 3 |
+| X-IIoTID | 0.50 | +0.0291 | 19 / 9 / 2 |
+| X-IIoTID | 0.75 | −0.0031 | 14 / 15 / 1 |
+| 5G-NIDD | 0.25 | **−0.0238** | 10 / 11 / 9 |
+| 5G-NIDD | 0.50 | +0.0051 | 16 / 10 / 4 |
+| 5G-NIDD | 0.75 | +0.0262 | 14 / 10 / 6 |
+
+Measured facts:
+
+- Support emphasis at beta=1 is most damaging at alpha=0.25 in both datasets (most negative
+  mean delta, most client degradation), consistent with those conditions selecting beta=0.
+- Within a single condition, some clients improve while others degrade at every beta
+  (both datasets, all alphas). `N_ij` vs ΔF1 is non-monotonic: the worst-degrading client is
+  not always the lowest-support client.
+- Client identities are per-seed realizations and are **not pooled** across seeds.
+
+---
+
+## 9. RQ1 answer
+
+> **RQ1.** How are class-support imbalance, differences between local prototypes, and rare-class
+> detection performance related in PROTEAN?
+
+Evidence chain (all three alphas, both datasets):
+
+*Mechanically established.*
+- Changing beta changes the normalized contribution weights `N_ij^beta / Σ_k N_kj^beta` and
+  therefore the position of the global prototype `Cbar_j`. For a fixed snapshot, mean
+  displacement grows monotonically with beta in all six conditions.
+
+*Empirically observed (descriptive).*
+- Larger realized support imbalance (higher support CV, more missing cells, higher client-size
+  ratio) tends to accompany larger beta=1 displacement, within both datasets:
+  X-IIoTID (CV, disp) = (1.442, 3.422) → (1.183, 2.099) → (0.881, 1.305);
+  5G-NIDD = (1.302, 2.919) → (1.225, 2.340) → (0.916, 1.385).
+- Imbalance also co-varies with selection behavior: both alpha=0.25 conditions are the most
+  imbalanced and both selected beta=0.
+
+*Not established.*
+- Larger displacement does **not** automatically imply better rare-class detection: at beta=1
+  (maximal displacement) rare-class performance is worse than at the selected beta in most
+  conditions, and validation response is non-monotone in beta.
+- The imbalance → displacement link is mechanical under the beta formula; the link from
+  displacement to detection is confounded by end-to-end retraining and cannot be isolated with
+  this design.
+
+RQ1 can therefore be answered **descriptively**: support imbalance and prototype displacement
+co-vary, but displacement is not a proxy for rare-class detection quality.
+
+---
+
+## 10. RQ2 answer
+
+> **RQ2.** How does changing beta affect the global prototype and rare-class detection
+> performance under different non-IID conditions?
+
+*Direct aggregation effect (fixed snapshot, no retraining).*
+- Displacement from beta=0 is monotone increasing in beta in every condition; magnitude differs
+  by condition (largest at alpha=0.25).
+- This is a pure weighting/geometry effect and is measured independently of training.
+
+*End-to-end training effect (iterative).*
+- Pooled validation rare-class macro-F1 is **not monotone** in beta. Preference varies by
+  condition: high beta is preferred at 5G-NIDD alpha=0.50 (beta=1.0), moderate beta at both
+  alpha=0.75 conditions (beta=0.5), and the smallest beta at both alpha=0.25 conditions
+  (beta=0.0).
+- Weighting helps some clients and hurts others within the same condition (section 8), so the
+  condition-level preference is an average over heterogeneous client responses.
+
+The two effects must not be merged: the geometry changes monotonically, the task-level response
+does not.
+
+---
+
+## 11. RQ3 answer
+
+> **RQ3.** Does the validation-selected beta change across non-IID levels, and what is the
+> effect on overall and per-client performance?
+
+Authoritative selected betas (validation only):
+
+| dataset | alpha=0.25 | alpha=0.50 | alpha=0.75 |
+|---|---|---|---|
+| X-IIoTID | 0.00 | 0.75 | 0.50 |
+| 5G-NIDD | 0.00 | 1.00 | 0.50 |
+
+Established conclusions:
+
+- The selected beta **changes by condition**: it is not a single universal value and not a
+  monotone function of alpha.
+- Both datasets select beta=0 at alpha=0.25 and beta=0.50 at alpha=0.75; they differ at
+  alpha=0.50 (X-IIoTID 0.75 vs 5G-NIDD 1.00).
+- Only five discrete candidates were tested; no global optimum is claimed, and each value is
+  "best among the tested candidates under pooled validation for that condition".
+
+Overall effect (section 6): selected-beta test rare-F1 differences are small and positive in
+three conditions, small and negative in three; the largest positive difference is smaller than
+between-seed variability. Per-client effect (section 8): the selected beta improves the
+validation rare-class metric for a majority of clients in four conditions, but at both alpha=0.25
+conditions (where the selected beta is 0) there is trivially no weighting change, and at higher
+beta the effect is client-dependent.
+
+---
+
+## 12. Cross-dataset synthesis
+
+- **Agreement**: matched-alpha selections coincide at alpha=0.25 (both 0.00) and alpha=0.75
+  (both 0.50); they differ at alpha=0.50 (0.75 vs 1.00).
+- **Displacement ordering is consistent**: beta=1 displacement falls with alpha in both datasets,
+  and ranking of conditions by displacement mirrors ranking by realized support CV.
+- **Rare-class objects are not comparable across datasets**: 5G-NIDD rare classes are low-volume
+  flood/scan types; X-IIoTID rare classes are low-volume behavioural categories. Absolute metric
+  levels must not be compared across datasets; only within-condition differences are meaningful.
+- **Metric asymmetry**: 5G-NIDD weighted-F1 < accuracy; X-IIoTID weighted-F1 > accuracy.
+- No cross-dataset causality, no dataset superiority, no dataset-independent beta rule.
+
+---
+
+## 13. Selection uncertainty
+
+Table: `artifacts/analysis/final__selection_strength.csv`.
+
+| dataset | alpha | beta* | best | 2nd best | abs gap | rel gap | top-2 gap / selected std | per-seed betas |
+|---|---|---|---|---|---|---|---|---|
+| X-IIoTID | 0.25 | 0.00 | 0.5917 | 0.5887 (0.25) | 0.0030 | 0.51% | 0.034 | 0.75 / 0.5 / 0.0 |
+| X-IIoTID | 0.50 | 0.75 | 0.5794 | 0.5735 (1.0) | 0.0059 | 1.02% | 0.078 | 0.5 / 1.0 / 0.75 |
+| X-IIoTID | 0.75 | 0.50 | 0.4767 | 0.4744 (0.25) | 0.0024 | 0.50% | 0.051 | 0.5 / 0.5 / 0.0 |
+| 5G-NIDD | 0.25 | 0.00 | 0.7615 | 0.7493 (0.25) | 0.0122 | 1.60% | 0.556 | 0.25 / 0.0 / 0.0 |
+| 5G-NIDD | 0.50 | 1.00 | 0.6924 | 0.6884 (0.75) | 0.0040 | 0.58% | 0.029 | 1.0 / 0.0 / 1.0 |
+| 5G-NIDD | 0.75 | 0.50 | 0.8091 | 0.8043 (1.0) | 0.0048 | 0.59% | 0.081 | 0.5 / 1.0 / 1.0 |
+
+Quantitative basis for the (descriptive) classification: the top-two gap is compared with the
+selected beta's across-seed sample std. **In all six conditions the top-two gap is smaller than
+the selected-beta sample std** (ratios 0.03 – 0.56), so every selection is classified as
+*small numerical separation relative to seed variability*. No tie-break was ever exercised. The
+per-seed diagnostic beta agrees with the pooled selection in only 1–2 of 3 seeds per condition.
+
+No inferential test (t-test, ANOVA, Wilcoxon, Friedman, regression) is applied. With three
+project-defined seeds per condition, such tests would be underpowered and would overstate the
+evidence; mean, sample std, per-seed values and the top-two gap are reported instead.
+
+---
+
+## 14. Threats to validity
+
+### Internal validity
+- **`S_j` reproduction assumption**: the participant set / denominator rule is not verified
+  against official code and is materially exercised in up to 19/27 cells. All results are
+  conditional on it.
+- **beta=0 floating-point divergence**: beta=0 is semantically the equal mean but not bitwise
+  identical to `original_protean`; selected-vs-original differences at alpha=0.25 are this
+  divergence, not an effect. Equivalence class B everywhere.
+- **Frozen implementation**: aggregation and training code unchanged across all 108 runs;
+  the same manifest/split/init is reused within each `(dataset, alpha, seed)`.
+- **Paired partitions/seeds**: verified by content, partition and split hashes.
+- **Validation/test separation**: beta selected from validation only; `test_used_for_selection`
+  is false in all runs; test read only after selection.
+
+### Construct validity
+- `N_ij` is **sample support, not prototype quality**; a larger `N_ij` does not mean a better
+  prototype.
+- Rare-class definition is operational (two smallest positive-support attack classes per client
+  from the training split); rare and unseen are never merged.
+- Prototype displacement is a **representation** metric, not detection quality.
+
+### External validity
+- Only two IDS datasets (X-IIoTID, 5G-NIDD), 10 clients, three alpha levels, three
+  project-defined seeds, one PROTEAN architecture and training configuration.
+- Rare-class compositions are dataset-specific and not comparable across datasets.
+
+### Conclusion validity
+- Small seed count (3); between-beta spreads are comparable to across-seed std.
+- Discrete beta grid `{0,0.25,0.5,0.75,1.0}`; the true response between candidates is unobserved.
+- Weak numerical separation at all six conditions (section 13).
+- No inferential significance testing was performed or is claimed.
+
+---
+
+## 15. Contribution supported by evidence
+
+1. A **controlled, beta-parameterized support-weighting** operator applied to the server-side
+   prototype aggregation of PROTEAN, with an explicit participant rule and missing-class
+   semantics.
+2. **Condition-wise, validation-based beta selection** (no test leakage) across six paired
+   conditions.
+3. A **paired multi-seed analysis** across two IDS datasets and three non-IID levels, with
+   identical partitions within each comparison cell.
+4. A **direct linkage** between realized support imbalance, weight redistribution, prototype
+   displacement, rare-class detection and per-client consequences — separated into direct
+   aggregation effects (fixed snapshots) and end-to-end training effects.
+5. Evidence of **when stronger support weighting helps, has little effect, or degrades**
+   rare-class behavior, including the identification that both evaluated alpha=0.25 conditions
+   select beta=0 (no weighting benefit) and that response is not monotone in beta.
+
+**Novelty caveat.** Sample-count prototype weighting is already discussed in related work, e.g.
+**FedProto** (Tan et al., 2022). The contribution cannot rest on using `N_ij` or adding beta
+alone; it rests on the controlled parameterization, the condition-wise validation selection, the
+paired multi-seed setup, and the resulting behavioral findings.
+
+---
+
+## 16. Claims NOT supported by evidence
+
+The following must not appear as established results:
+
+- "Support weighting improves rare-class detection." (Mixed and condition-dependent; largest
+  positive difference is within seed noise.)
+- "Selected-beta beats original PROTEAN." (Only small descriptive differences; no significance
+  test; at alpha=0.25 the difference is floating-point divergence.)
+- "alpha causally determines the optimal beta." (Only two datasets, three seeds; selection is
+  not monotone in alpha.)
+- "A larger `N_ij` means a better prototype / greater prototype quality."
+- "Larger prototype displacement means better (or worse) rare-class detection."
+- "beta=0 reproduces original PROTEAN" (only semantically equivalent under the shared `S_j`
+  assumption; class B divergence).
+- "There is a universal / globally optimal beta."
+- "The method generalizes" or "dataset X is superior".
+- Any metric interpreted as improvement without the qualifier "numerical/descriptive".
+
+---
+
+## 17. Recommended thesis conclusions
+
+1. Under the validated design, support weighting is **condition-dependent**: the best tested beta
+   differs across non-IID conditions, is not monotone in alpha, and no single beta dominates.
+2. Changing beta **monotonically displaces** global prototypes at a fixed snapshot, while the
+   rare-class detection response is **non-monotone** and weak relative to seed variability.
+3. Realized support imbalance and prototype displacement **co-vary**; displacement alone is not
+   evidence of detection quality.
+4. Weighting **redistributes benefit across clients**; within a condition some clients gain and
+   others lose.
+5. In both evaluated datasets the most imbalanced condition (alpha=0.25) selected beta=0, i.e.
+   equal weighting over the shared participant set, and realized the largest prototype
+   displacement — a descriptive pattern that motivates broader validation, not a rule.
+
+---
+
+## 18. Scientific-gap assessment
+
+**Classification: A — NO MATERIAL GAP** for the defined thesis scope.
+
+- RQ1, RQ2 and RQ3 are all answerable descriptively from the 108-run grid and the existing
+  audited artifacts.
+- The remaining limitations (S_j assumption, three seeds, discrete grid, two datasets, no
+  inferential power) are **scope/validity** boundaries, not missing experiments that prevent the
+  stated research questions from being answered.
+- A **causal** isolation of the displacement→detection link (e.g. evaluating recomputed global
+  prototypes on trained embeddings at a fixed snapshot, or interpolating beta in prototype space)
+  would strengthen RQ1/RQ2, but it is not required by the defined RQs and would change the design;
+  it is therefore out of scope rather than a blocking gap.
+- No new training is required or recommended.
+
+---
+
+## Artifact index (Session 12 deliverables)
+
+| artifact | path |
+|---|---|
+| master table | `artifacts/analysis/final__six_condition_master.{csv,json}` |
+| pooled validation curves | `artifacts/analysis/final__pooled_validation_curves.csv` |
+| realized heterogeneity | `artifacts/analysis/final__realized_heterogeneity.csv` |
+| displacement by beta | `artifacts/analysis/final__prototype_displacement_by_beta.csv` |
+| displacement by class (beta=1) | `artifacts/analysis/final__prototype_displacement_by_class.csv` |
+| client-level effect | `artifacts/analysis/final__client_effect_summary.csv` |
+| selected vs original | `artifacts/analysis/final__selected_vs_original.csv` |
+| selection strength | `artifacts/analysis/final__selection_strength.csv` |
+| figures | `artifacts/figures/final/figA–G*.png` |
+| Chapter 4 outline | `docs/thesis/CHAPTER4_RESULTS_OUTLINE.md` |
+| claims register | `docs/thesis/CLAIMS_REGISTER.md` |
+
+**Environment note.** Figures require `matplotlib`, which is not in the frozen `research`
+dependency set. It was installed into the git-ignored `.venv` only
+(`uv pip install --python .venv/bin/python matplotlib`, matplotlib 3.11.2); no tracked
+dependency file was modified. The synthesis generator is `scripts`-style analysis code kept
+outside the repository at `/tmp/opencode/s12/synthesize_final_analysis.py`, consistent with the
+Session-11 convention for ad-hoc analysis helpers.

docs/thesis/CHAPTER4_RESULTS_OUTLINE.md

@@ -0,0 +1,148 @@
+# Chapter 4 — Results and Discussion (Blueprint)
+
+This is a **blueprint**, not a finished chapter. Each subsection states its purpose, the exact
+table/figure to use, the key numbers to quote (loaded from audited artifacts), what may be
+interpreted, and what must not be interpreted. All numbers trace to
+`docs/research/FINAL_EXPERIMENTAL_SYNTHESIS.md` and the `artifacts/analysis/final__*` files.
+
+Global wording rules: prefer "numerical/descriptive difference", "best among the tested
+candidates under pooled validation", "co-varies"; never "improvement", "optimal", "causal",
+"superior", or "generalizes" unless explicitly qualified.
+
+---
+
+## 4.1 Experimental setup recap
+
+- **Purpose**: fix the protocol so later numbers are reproducible and comparable.
+- **Use**: text; table of grid (Section 2 of synthesis); `docs/research/EXPERIMENT_PROTOCOL.md`.
+- **Key numbers**: 2 datasets × 3 alpha × 3 seeds × (5 betas + `original_protean`) = **108 runs**;
+  betas `{0, 0.25, 0.5, 0.75, 1}`; 10 clients; 10 rounds; 3 local epochs; batch 256; Adam lr=1e-3,
+  mu=0.1, lambda=1.0; train/val/test 70/10/20; validation-only selection.
+- **Allowed**: state that `S_j` is a reproduction assumption; state that beta=0 is not bitwise
+  identical to the baseline.
+- **Avoid**: describing the design as new training beyond the frozen grid; implying the paper's
+  split was 70/10/20.
+
+## 4.2 Realized non-IID characteristics
+
+- **Purpose**: show that alpha is a design knob and realized heterogeneity is the measured
+  variable.
+- **Use**: Table `final__realized_heterogeneity.csv`; Figure `figC_realized_support_cv.png`.
+- **Key numbers**: support CV 1.442/1.183/0.881 (X-IIoTID) and 1.302/1.225/0.916 (5G-NIDD) for
+  alpha 0.25/0.50/0.75; `|S_j|<M` 19/7/3 and 16/4/4 of 27; missing cells 38/9/4 and 30/4/5;
+  client-size ratio 14.75/7.41/4.67 and 17.60/9.59/6.31.
+- **Allowed**: realized imbalance generally decreases as alpha rises; per-seed imbalance is not
+  monotone at fixed alpha; rare-class union differs across seeds.
+- **Avoid**: treating Dirichlet alpha itself as the observed heterogeneity; comparing rare-class
+  metrics across datasets.
+
+## 4.3 Beta tuning results
+
+- **Purpose**: present the pooled validation curves and the six selections.
+- **Use**: Table `final__pooled_validation_curves.csv`; Figures
+  `figA_xiiotid_pooled_validation_beta_curve.png`,
+  `figA_5gnidd_pooled_validation_beta_curve.png`, `figB_selected_beta_by_condition.png`;
+  selection-strength table `final__selection_strength.csv`.
+- **Key numbers**: selections X-IIoTID 0.00 / 0.75 / 0.50 and 5G-NIDD 0.00 / 1.00 / 0.50 for
+  alpha 0.25 / 0.50 / 0.75; response non-monotone in beta; all top-two gaps < selected-beta
+  across-seed std (relative gaps 0.50% – 1.60%).
+- **Allowed**: selected beta changes by condition; selection is not monotone in alpha; no global
+  optimum; separation is small relative to seed variability.
+- **Avoid**: calling the selected beta "optimal"; drawing a continuous optimum; reporting
+  per-seed diagnostic betas as the primary result.
+
+## 4.4 Effect on global prototypes
+
+- **Purpose**: separate the direct (fixed-snapshot) aggregation effect from training.
+- **Use**: Table `final__prototype_displacement_by_beta.csv`; Figure
+  `figD_prototype_displacement_beta0_vs_beta1.png`; per-class file
+  `final__prototype_displacement_by_class.csv`.
+- **Key numbers**: mean `||Cbar(1)−Cbar(0)||` = 3.422 / 2.099 / 1.305 (X-IIoTID) and
+  2.919 / 2.340 / 1.385 (5G-NIDD); monotone increasing in beta within each condition.
+- **Allowed**: displacement grows monotonically with beta for a fixed snapshot; magnitude differs
+  by realized condition; larger beta moves the global prototype toward high-support clients.
+- **Avoid**: interpreting displacement as improved/degraded detection; implying a causal link
+  from displacement to rare-class performance.
+
+## 4.5 Rare-class detection results
+
+- **Purpose**: report selected-beta test metrics against `original_protean` descriptively.
+- **Use**: master table `final__six_condition_master.csv`; table
+  `final__selected_vs_original.csv`; Figure
+  `figF_selected_vs_original_test_rare_macro_f1.png`.
+- **Key numbers**: Δ test rare-F1 = −0.0007 / +0.0230 / +0.0014 (X-IIoTID) and
+  −0.0010 / −0.0086 / +0.0168 (5G-NIDD); largest positive difference (+0.0230) below that
+  condition's between-seed std (0.0267); alpha=0.25 differences are floating-point divergence.
+- **Allowed**: differences are numerical/descriptive only; mixed sign; rare-recall is usually
+  slightly lower under weighting.
+- **Avoid**: significance markers; "beats baseline"; merging rare and unseen; comparing absolute
+  levels across conditions or datasets.
+
+## 4.6 Per-client effects
+
+- **Purpose**: show redistribution of benefit across clients.
+- **Use**: Table `final__client_effect_summary.csv`; Figure
+  `figG_client_effect_at_selected_beta.png`.
+- **Key numbers**: beta=1 mean Δ = −0.0596 / +0.0291 / −0.0031 (X-IIoTID) and
+  −0.0238 / +0.0051 / +0.0262 (5G-NIDD); improve/degrade/unchanged counts per condition.
+- **Allowed**: within a condition some clients gain and others lose; most damaging at alpha=0.25;
+  `N_ij` vs ΔF1 is non-monotonic.
+- **Avoid**: pooling client identities across seeds; claiming weighting helps "all" or
+  "low-support" clients.
+
+## 4.7 Cross-dataset analysis
+
+- **Purpose**: compare the two datasets at matched alpha without over-claiming.
+- **Use**: synthesis Section 12; cross-dataset artifacts `cross_dataset__matched_alpha_summary.csv`.
+- **Key numbers**: matched selections agree at alpha=0.25 (0.00) and alpha=0.75 (0.50), differ at
+  alpha=0.50 (0.75 vs 1.00).
+- **Allowed**: descriptive agreement/disagreement; both datasets' displacement falls with alpha.
+- **Avoid**: cross-dataset metric comparison; dataset causality; universal beta rule; claiming a
+  dataset-independent law.
+
+## 4.8 RQ1 discussion
+
+- **Purpose**: answer the support-imbalance ↔ prototype-difference ↔ rare-class relation.
+- **Use**: synthesis Section 9; sections 4.2, 4.4, 4.5.
+- **Key numbers**: imbalance ranking and displacement ranking coincide within each dataset.
+- **Allowed**: mechanical weight/prototype effect is established; imbalance and displacement
+  co-vary descriptively; displacement is not detection quality.
+- **Avoid**: causal mediation claims; "displacement improves detection".
+
+## 4.9 RQ2 discussion
+
+- **Purpose**: answer how beta affects global prototypes and rare-class performance.
+- **Use**: synthesis Section 10; sections 4.3, 4.4, 4.6.
+- **Key numbers**: monotone displacement vs non-monotone validation response.
+- **Allowed**: distinguish direct aggregation effect from end-to-end training effect; weighting
+  helps some clients and harms others.
+- **Avoid**: treating the monotonically growing displacement as a performance trend.
+
+## 4.10 RQ3 discussion
+
+- **Purpose**: answer whether the selected beta changes across non-IID levels and its effect.
+- **Use**: synthesis Section 11; sections 4.3, 4.5, 4.6.
+- **Key numbers**: selected betas `{0.00, 0.75, 0.50}` (X-IIoTID) and `{0.00, 1.00, 0.50}`
+  (5G-NIDD); only five candidate betas tested.
+- **Allowed**: selection changes by condition; not monotone in alpha; no universal beta; both
+  datasets select 0 at alpha=0.25 and 0.5 at alpha=0.75.
+- **Avoid**: "alpha causally determines beta"; "global optimum"; extrapolating to untested betas.
+
+## 4.11 Threats / limitations
+
+- **Purpose**: bound the conclusions honestly.
+- **Use**: synthesis Section 14.
+- **Key numbers**: S_j assumption material up to 19/27 cells; three seeds; discrete grid; two
+  datasets; no inferential test.
+- **Allowed**: internal/construct/external/conclusion validity bullets as written.
+- **Avoid**: omitting the reproduction assumption; implying statistical significance.
+
+## 4.12 Chapter summary
+
+- **Purpose**: one-paragraph takeaway.
+- **Use**: synthesis Section 17.
+- **Key numbers**: condition-dependent beta, non-monotone response, displacement co-varies with
+  imbalance, client-level heterogeneity, both alpha=0.25 conditions select beta=0.
+- **Allowed**: descriptive summary; explicit statement that no global optimum or universal beta
+  was found and that only tested candidates were compared.
+- **Avoid**: any claim of improvement, causality, superiority or generalization.

docs/thesis/CLAIMS_REGISTER.md

@@ -0,0 +1,149 @@
+# Thesis Claims Register
+
+Every claim considered for Chapter 4/5, with its evidence, status, limitation and a
+thesis-safe wording. Statuses:
+
+- **supported** — directly established by audited machine-readable artifacts or by the frozen
+  operator definition.
+- **descriptive only** — observable in the artifacts but not causally or inferentially established.
+- **unsupported** — the design cannot defend the claim; must not appear as a result.
+
+Artifacts referenced live under `artifacts/analysis/final__*`, `artifacts/analysis/*__multiseed__*`,
+and `docs/research/*`. Numbers are cross-checked in
+`docs/research/FINAL_EXPERIMENTAL_SYNTHESIS.md`.
+
+---
+
+## 1. Beta changes prototype positions
+
+- **Status**: supported (mechanically established).
+- **Evidence**: `final__prototype_displacement_by_beta.csv`; frozen aggregator definition
+  `Cbar_j(beta) = Σ N_ij^beta C_ij / Σ N_ij^beta`.
+- **Limitation**: conditional on the shared `S_j` reproduction assumption; fixed round-0 snapshot.
+- **Safe wording**: "Changing beta changes the normalized contribution weights and therefore the
+  position of each global prototype."
+
+## 2. Prototype displacement increases with beta
+
+- **Status**: supported (monotone at fixed snapshot).
+- **Evidence**: `final__prototype_displacement_by_beta.csv` — mean displacement increases from
+  beta=0.25 to beta=1.00 in all six conditions.
+- **Limitation**: measured on fixed round-0 snapshots; no retraining; not a detection metric.
+- **Safe wording**: "For a fixed snapshot, mean `||Cbar(beta)−Cbar(0)||` increases monotonically
+  with beta in every evaluated condition."
+
+## 3. Realized imbalance and prototype displacement co-vary
+
+- **Status**: descriptive only.
+- **Evidence**: `final__realized_heterogeneity.csv` and `final__prototype_displacement_by_beta.csv`;
+  both datasets show support CV and beta=1 displacement falling together as alpha rises.
+- **Limitation**: six conditions, three seeds; correlation not quantified with inferential
+  statistics; displacement is a representation metric.
+- **Safe wording**: "Within both datasets, conditions with larger realized support imbalance also
+  exhibit larger beta=1 prototype displacement; this is a descriptive co-variation."
+
+## 4. Beta selection changes by condition
+
+- **Status**: supported (measured, validation-only).
+- **Evidence**: `final__six_condition_master.csv`; `final__selection_strength.csv`.
+- **Limitation**: only five discrete candidates; selection is "best among tested candidates".
+- **Safe wording**: "The validation-selected beta differs across conditions and is not monotone in
+  alpha; it is best among the tested candidates for each condition."
+
+## 5. Beta is not universal
+
+- **Status**: supported.
+- **Evidence**: selected betas `{0.00,0.75,0.50}` (X-IIoTID) and `{0.00,1.00,0.50}` (5G-NIDD).
+- **Limitation**: only two datasets and three alpha levels.
+- **Safe wording**: "The selected beta takes three distinct values across the six conditions and
+  is not a single universal value; no global optimum is claimed."
+
+## 6. Beta=0 selected at alpha=0.25 on both datasets
+
+- **Status**: supported (measured).
+- **Evidence**: `final__six_condition_master.csv` (X-IIoTID and 5G-NIDD alpha=0.25 both select 0.00).
+- **Limitation**: two datasets, three seeds; descriptive pattern, not a rule. Because the selected
+  beta is 0, these conditions cannot demonstrate any support-weighting benefit.
+- **Safe wording**: "In both evaluated datasets the alpha=0.25 condition selected beta=0 among the
+  tested candidates; these conditions also showed the strongest realized support imbalance and the
+  largest prototype displacement."
+
+## 7. Selected beta beats original PROTEAN
+
+- **Status**: unsupported.
+- **Evidence**: `final__selected_vs_original.csv` — differences are mixed in sign and small; the
+  largest positive rare-F1 difference (+0.0230, X-IIoTID alpha=0.50) is below that condition's
+  between-seed std (0.0267); alpha=0.25 differences are floating-point divergence (class B).
+- **Limitation**: no significance test; three seeds; selected beta is determined post hoc per
+  condition.
+- **Safe wording**: "Selected-beta test metrics show only small, mixed-sign numerical differences
+  from `original_protean`; no statistical superiority is established."
+
+## 8. Support weighting improves rare-class detection
+
+- **Status**: unsupported.
+- **Evidence**: `final__selected_vs_original.csv`, `final__client_effect_summary.csv` — response is
+  condition-dependent, non-monotone in beta, and some clients degrade at every beta; at beta=1 the
+  mean client delta is negative at both alpha=0.25 conditions and at X-IIoTID alpha=0.75.
+- **Limitation**: end-to-end results confound the weighting with retraining; small seed count.
+- **Safe wording**: "Support weighting changes rare-class behavior condition-dependently: it helps
+  a majority of clients in some conditions and degrades most clients at the most imbalanced
+  conditions; no uniform improvement is demonstrated."
+
+## 9. Alpha causally determines the optimal beta
+
+- **Status**: unsupported.
+- **Evidence**: `final__selection_strength.csv` — selection is not monotone in alpha; selected beta
+  is `{0.00,0.75,0.50}` and `{0.00,1.00,0.50}`, differing at alpha=0.50.
+- **Limitation**: only three alpha levels, two datasets, three seeds; no trend test performed.
+- **Safe wording**: "The selected beta changes with non-IID level but is not a monotone function of
+  alpha; no causal role of alpha is established."
+
+## 10. Beta=1 represents prototype quality
+
+- **Status**: unsupported.
+- **Evidence**: spec and protocol define `N_ij` as class-specific sample support, not prototype
+  quality; beta=1 weights by support only.
+- **Limitation**: no ground-truth prototype-quality measure exists in the design.
+- **Safe wording**: "`N_ij` is class-specific sample support; a larger `N_ij` does not imply a
+  higher-quality prototype, and beta weights support, not quality."
+
+## 11. Prototype displacement implies detection quality
+
+- **Status**: unsupported.
+- **Evidence**: `final__prototype_displacement_by_beta.csv` vs `final__pooled_validation_curves.csv`
+  — displacement is monotone in beta while detection response is non-monotone; at beta=1 detection
+  is usually worse than at the selected beta.
+- **Limitation**: displacement measured on untreated round-0 embeddings; detection measured
+  end-to-end.
+- **Safe wording**: "Prototype displacement is a representation-aggregation metric; it is not
+  interpreted as improved or degraded detection."
+
+## 12. Beta=0 reproduces original PROTEAN
+
+- **Status**: unsupported (conditional equivalence only).
+- **Evidence**: `*__multiseed__beta0_equivalence.csv` — same manifest/snapshot/support but
+  independent-aggregation differences ~2e-16 and non-zero history/test metric differences;
+  classification **B** in every condition.
+- **Limitation**: conditional on the shared `S_j` reproduction assumption; no official code.
+- **Safe wording**: "Under the shared participant-set assumption, beta=0 is semantically the equal
+  mean over `S_j` and is numerically close to `original_protean`; it is not a proof that beta=0
+  reproduces the original method."
+
+## 13. Rare-class and unseen-class performance can be merged
+
+- **Status**: unsupported.
+- **Evidence**: protocol defines rare (`N_ij>0`, two smallest positive-support attack classes) and
+  unseen (`N_ij=0`) as distinct categories from the training split.
+- **Limitation**: none; definitional.
+- **Safe wording**: "Rare-class and unseen-class results are reported separately and are never
+  merged into one metric."
+
+## 14. Results generalize beyond the tested setting
+
+- **Status**: unsupported.
+- **Evidence**: two datasets, 10 clients, three alpha levels, three project-defined seeds, one
+  architecture/training configuration.
+- **Limitation**: external validity.
+- **Safe wording**: "Findings are specific to the two evaluated IDS datasets and the frozen
+  PROTEAN configuration; no generalization is claimed."