diff --git a/docs/BENCHMARK_CARD.md b/docs/BENCHMARK_CARD.md
index fc486497..a7d2c36d 100644
--- a/docs/BENCHMARK_CARD.md
+++ b/docs/BENCHMARK_CARD.md
@@ -1,7 +1,7 @@
 # Benchmark Card — OpenAMP Foundry v0.5.x
 
 > **Purpose:** Single-page summary of benchmark methodology, data, and metrics.
-> **Last updated:** 2026-07-01 (cluster-split benchmark added)
+> **Last updated:** 2026-07-01 (hemolysis benchmark expanded to 238 peptides)
 
 ---
 
@@ -39,6 +39,32 @@
 | Bootstrap | 2000 resamples | 2000 resamples | — |
 | **Interpretation** | **STRONG** | **STRONG** | Below random (expected) |
 
+## Within-AMP Selectivity Benchmark
+
+> Tests whether scorers can distinguish hemolytic AMPs (HC50 < 25 µg/ml) from
+> selective AMPs (HC50 >= 100 µg/ml). Run: `make bench-selectivity`
+
+| Set | Count | Description |
+|-----|:-----:|-------------|
+| Hemolytic | **54** | HC50 < 25 µg/ml (literature + DBAASP v3) |
+| Selective | **125** | HC50 >= 100 µg/ml |
+| Border | **59** | 25 <= HC50 < 100 (excluded from binary AUROC) |
+| **Total** | **238** | Expanded from 42 in v0.5.11 using DBAASP data |
+
+| Score | Detection AUROC | CI₉₅ | Significant? |
+|-------|:--------------:|:----:|:------------:|
+| hemolysis_risk | 0.5650 | 0.47–0.66 | No (direction correct) |
+| safety | 0.5116 | 0.43–0.60 | No |
+| selectivity_proxy | 0.5744 | 0.50–0.66 | Borderline |
+| ensemble | 0.4201 | 0.33–0.51 | No (wrong direction) |
+| expert_composite | 0.5459 | 0.46–0.63 | No |
+| serum_stability | 0.5873 | 0.51–0.66 | Yes (risk detector) |
+
+**Key finding:** The hemolysis risk scorer's original AUROC=0.9218 (CI 0.82-0.99)
+on n=35 was small-sample inflation. On the expanded n=179, no scorer achieves
+strong hemolysis detection. Hemolysis remains unpredictable by 1D
+physicochemical features and must be assayed experimentally.
+
 ## Method
 
 AUROC computed via Wilcoxon-Mann-Whitney statistic (concordant-pair enumeration).
diff --git a/docs/METRICS_CURRENT.md b/docs/METRICS_CURRENT.md
index f95d41e4..bcbfcef6 100644
--- a/docs/METRICS_CURRENT.md
+++ b/docs/METRICS_CURRENT.md
@@ -3,7 +3,7 @@
 > **Purpose:** One authoritative table of current pipeline metrics. If any doc disagrees
 > with this file, this file wins. Updated whenever benchmark/benchmark config changes.
 >
-> **Last updated:** 2026-07-01 (cluster-split benchmark added)
+> **Last updated:** 2026-07-01 (hemolysis benchmark expanded to 238 peptides)
 > **Pipeline version:** v0.5.x
 > **Branch:** main
 
@@ -312,56 +312,66 @@ upon as a hemolysis predictor.
 > hemolysis detection (AUROC=0.3844). A dedicated hemolysis risk scorer was
 > built from empirically-validated components identified in that benchmark.
 >
+> **v0.5.11 correction:** The original 42-peptide reference set (14 hemolytic
+> vs 21 selective, n=35) produced detection AUROC=0.9218 (CI 0.82-0.99). This
+> was **small-sample inflation**. Expansion to 238 peptides using DBAASP human
+> erythrocyte data (54 hemolytic vs 125 selective, n=179) dropped the detection
+> AUROC to 0.5650 (CI 0.47-0.66) — direction correct but NOT statistically
+> significant. The scorer retains weak directional signal but should not be
+> trusted as a standalone hemolysis detector.
+>
 > Run: `make bench-selectivity` (hemolysis_risk column in the output)
 
 **Module:** `src/openamp_foundry/scoring/hemolysis.py`
 
-**Components** (detection AUROC from leave-one-out on n=14 hemolytic vs n=21 selective):
+**Components** (individual AUROC from original n=14 vs n=21; may not replicate on expanded set):
 
-| Component | Individual AUROC | Weight | Signal source |
-|-----------|:----------------:|:------:|---------------|
+| Component | Individual AUROC (n=35) | Weight | Signal source |
+|-----------|:-----------------------:|:------:|---------------|
 | Synthesis difficulty (1 - synth_feasibility) | 0.8027 | 0.30 | Incidental: hemolytic AMPs harder to synthesize |
 | Aromatic fraction (F/W/Y density) | 0.8299 | 0.30 | Trp/Phe intercalation in both membrane types |
 | Cationic-on-hydrophobic-face fraction | 0.7585 | 0.20 | Poor amphipathic face segregation |
 | Cysteine fraction | 0.7500 | 0.20 | Beta-sheet defensin/protegrin class |
 
-**Combined performance:**
+**Combined performance (expanded n=179):**
 
-| Metric | Value | Notes |
-|--------|-------|-------|
-| **Detection AUROC** | **0.9218** | Bootstrap CI₉₅: 0.82-0.99 (n_bootstrap=2000) |
-| Mean hemolytic risk | 0.4064 | n=14 |
-| Mean selective risk | 0.1501 | n=21 |
-| Risk threshold > 0.40 | Enriched in hemolytic | 10/14 hemolytic vs 2/21 selective |
-| Safety scorer (comparison) | 0.3844 | FAILS — all 14 hemolytic AMPs score safety >= 0.94 |
+| Metric | Original (n=35) | Expanded (n=179) | Notes |
+|--------|:---------------:|:-----------------:|-------|
+| **Detection AUROC** | **0.9218** | **0.5650** | Small-sample inflation corrected |
+| CI₉₅ lower bound | 0.82 | 0.47 | No longer > 0.5 — not significant |
+| CI₉₅ upper bound | 0.99 | 0.66 | |
+| Mean hemolytic risk | 0.4064 (n=14) | 0.2042 (n=54) | Direction still correct |
+| Mean selective risk | 0.1501 (n=21) | 0.1535 (n=125) | |
+| Safety scorer detection | 0.3844 | 0.5116 | Safety also improves slightly with more data |
 
-**Expert composite integration:**
+**Expert composite integration (expanded n=179):**
 
-| Metric | Before (v0.5.9) | After (v0.5.10) |
-|--------|:---------------:|:----------------:|
-| Expert composite detection AUROC | 0.5119 | 0.6429 |
-| Expert composite CI lo | 0.3129 | 0.4490 |
-| Ensemble detection AUROC | 0.3486 | 0.3486 (unchanged) |
+| Metric | Before (v0.5.9, n=35) | After (v0.5.10, n=35) | Expanded (n=179) |
+|--------|:---------------------:|:---------------------:|:-----------------:|
+| Expert composite detection AUROC | 0.5119 | 0.6429 | 0.5459 |
+| Expert composite CI lo | 0.3129 | 0.4490 | 0.4562 |
+| Ensemble detection AUROC | 0.3486 | 0.3486 | 0.4201 |
 
-**Expert ablation (AMP-vs-decoy):**
+**Expert ablation (AMP-vs-decoy, unchanged):**
 
 | Metric | Value | Classification |
 |--------|:-----:|:--------------:|
 | hemolysis_safety AUROC | 0.3285 | **Anti-signal** (above_random = -0.1715) |
 | Expert composite AUROC | 0.7119 | Down from 0.7360 (delta -0.0713 vs ensemble) |
 
-**Key finding:** The dedicated hemolysis risk scorer is the first pipeline score
-with a statistically significant hemolysis detection signal (CI lo = 0.82 > 0.5).
-It complements (not replaces) the safety scorer, which retains its role for
-AMP-vs-decoy discrimination. The hemolysis_safety component is anti-signal on
-AMP-vs-decoy (AUROC=0.3285) — expected, since real AMPs have higher hemolysis
-risk than random decoys. This confirms the component measures a within-AMP
-property, not an AMP-vs-non-AMP property.
+**Key finding (corrected):** The hemolysis risk scorer's original detection
+AUROC=0.9218 on n=35 was small-sample inflation. On the expanded n=179
+reference set, detection AUROC=0.5650 (CI 0.47-0.66) — direction is correct
+(hemolytic > selective on average) but not statistically significant. The
+scorer should NOT be described as a "statistically significant hemolysis
+detector." It provides weak directional signal that may be useful as one
+factor in a composite but cannot be relied upon for hemolysis triage. Hemolysis
+must still be assayed experimentally for every candidate.
 
-**Honest limitation:** The reference set is small (n=35). The CI is wide.
-Melittin's risk score (0.13) remains modest because its bent-helix hemolysis
-mechanism is not fully captured by 1D features. Hemolysis must still be assayed
-experimentally for every candidate.
+**Honest limitation:** The expanded reference set (n=179) provides a more honest
+estimate, but HC50 values are approximate literature values with high inter-assay
+variability. Melittin's risk score (0.13) remains modest because its bent-helix
+hemolysis mechanism is not fully captured by 1D features.
 
 ---
 
@@ -369,7 +379,7 @@ experimentally for every candidate.
 
 | Metric | Value |
 |--------|-------|
-| Total tests | 1471 |
+| Total tests | 1471+ |
 | Coverage (branch) | 99% (6 CLI guard lines only) |
 | Source modules at 100% | All pipeline, QC, scoring modules |
 
@@ -398,6 +408,7 @@ experimentally for every candidate.
 | 2026-06-29 | External predictor results filled from wave05_combined_consensus.csv; all 7 gates PASS | OpenAMP Wave 0.5 |
 | 2026-06-29 | Wave 0.5 scaffold diversification — 24-candidate Wave 1 panel across 14 families | OpenAMP Wave 0.5 |
 | 2026-07-01 | Expert ablation benchmark added: expert composite AUROC 0.736 vs ensemble 0.7832 (delta −0.0472); 3 components anti-signal; ensemble remains primary gate | OpenAMP loop |
+| 2026-07-01 | **Hemolysis benchmark expanded:** 42 -> 238 peptides using DBAASP human erythrocyte data (54 hemolytic vs 125 selective, n=179 binary). Hemolysis risk scorer detection AUROC drops 0.9218 -> 0.5650 (CI 0.47-0.66) — original performance was small-sample inflation. Direction correct, not significant. Safety scorer detection improves 0.3844 -> 0.5116 (still not significant). 196 new peptides from DBAASP v3. | OpenAMP loop |
 | 2026-07-01 | Dedicated hemolysis risk scorer: 4-component score (synth+aromatic+face+cys) achieves detection AUROC=0.9218 (CI: 0.82-0.99); integrated into expert composite (detection 0.5119→0.6429); safety scorer unchanged; 1471 tests | OpenAMP loop |
 | 2026-07-01 | Within-AMP selectivity benchmark added: safety scorer FAILS hemolysis detection (AUROC=0.3844); synthesis is only significant risk detector (AUROC=0.8027); expert composite better than ensemble but not significant (0.5119 vs 0.3486) | OpenAMP loop |
 | 2026-06-29 | Initial — expanded benchmark (PR #110) | OpenAMP CI |
diff --git a/docs/ROADMAP.md b/docs/ROADMAP.md
index ec916ad9..48435aa6 100644
--- a/docs/ROADMAP.md
+++ b/docs/ROADMAP.md
@@ -146,12 +146,27 @@ Implemented during the pre-wet-lab improvement loop (PRs #31–#54):
 - `make bench-cluster-split` Makefile target added (was missing despite CLI command existing)
 - 14 new tests in `test_expert_ablation.py`; 1435 total tests
 
+## v0.5.11 — Hemolysis Benchmark Expansion (DBAASP) ✓ (2026-07-01)
+
+- Expanded hemolysis reference from 42 to 238 peptides using DBAASP v3 human erythrocyte data
+- 196 new peptides with 50% hemolysis (HC50) values, converted to µg/ml using peptide MW
+- Binary task: 54 hemolytic (HC50 < 25 µg/ml) vs 125 selective (HC50 >= 100 µg/ml) = 179 total (was 14 vs 21 = 35)
+- **Critical honesty correction:** hemolysis risk scorer detection AUROC drops from 0.9218 (CI 0.82-0.99) to 0.5650 (CI 0.47-0.66)
+  - Original n=35 performance was **small-sample inflation**
+  - Direction still correct (hemolytic mean=0.204 > selective mean=0.154) but NOT statistically significant
+  - Safety scorer detection improves slightly: 0.3844 → 0.5116 (still not significant)
+  - Selectivity proxy: detection AUROC=0.5744 (CI 0.50-0.66) — borderline significant
+- Extraction script: `scripts/expand_hemolysis_benchmark.py`
+- Tests updated: `test_selectivity_benchmark.py` test renamed from `test_hemolysis_risk_is_significant_detector` to `test_hemolysis_risk_direction_correct`
+- Docs updated: METRICS_CURRENT.md, hemolysis.py, expert.py comments corrected to reflect expanded-set reality
+- **Lesson:** A 0.92 AUROC on n=35 is not robust. The project's adversarial benchmark philosophy caught this.
+
 ## v0.5.10 — Dedicated Hemolysis Risk Scorer ✓ (2026-07-01)
 
 - `scoring/hemolysis.py` — standalone 4-component hemolysis risk score (synth difficulty + aromatic density + face cationic leakage + cysteine content)
-- Combined detection AUROC=0.9218 (CI₉₅: 0.82-0.99) — first statistically significant hemolysis detector
+- Combined detection AUROC=0.9218 (CI₉₅: 0.82-0.99) on original 42-peptide reference — **CORRECTED in v0.5.11: small-sample inflation; expanded n=179 gives 0.5650 (CI 0.47-0.66)**
 - Integrated into selectivity benchmark as risk-direction score; integrated into expert composite as `hemolysis_safety` (weight 0.10)
-- Expert composite hemolysis detection improved: 0.5119 → 0.6429 (not significant, but trend correct)
+- Expert composite hemolysis detection improved: 0.5119 → 0.6429 on n=35 — **expanded n=179 gives 0.5459 (CI 0.46-0.63)**
 - Safety scorer UNCHANGED — standard AUROC remains 0.7832; hemolysis risk is complementary, not replacement
 - Expert ablation: hemolysis_safety is anti-signal on AMP-vs-decoy (AUROC=0.3285) — confirms it measures within-AMP property
 - 15 new tests in `test_hemolysis_risk.py`; 3 new tests in `test_selectivity_benchmark.py`; 1471 total tests
diff --git a/examples/validation/hemolysis_reference.csv b/examples/validation/hemolysis_reference.csv
index eed25f60..66f96f13 100644
--- a/examples/validation/hemolysis_reference.csv
+++ b/examples/validation/hemolysis_reference.csv
@@ -41,3 +41,199 @@ REF-ESC-001,GIFSKLAGKKIKNLLISGLKG,esculentin,100,SELECTIVE,Simmaco_1993
 REF-BOM-001,IIGPVLGLVGSALGGLLKKI,bombinin_H,100,SELECTIVE,Halverson_1990
 REF-TMPG-001,ILSKIFSLLKKVLP,temporin_G,100,SELECTIVE,Simmaco_1996
 REF-POLY1-001,RRWCFRVCYRGFCYRKCR,polyphemusin_I,10,MODERATE,Miyata_1989
+DBAASP-37,GMASKAGAIAGKIAKVALKAL,dbaasp_hemolysis,1.2,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-102,ALWKTLLKKVLKAAAKAALKAVLVGANA,dbaasp_hemolysis,1.4,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1797,GIGAVLWVLTTGLPALISWIKRKRQQ,dbaasp_hemolysis,2.6,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-3173,FFGWLIKGAIHAGKAIHGLIHRRRH,dbaasp_hemolysis,2.9,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-3174,FFGWLIRGAIHAGKAIHGLIHRRRH,dbaasp_hemolysis,2.9,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1790,GIGAVLAVLTTGLPALISWIKRKRQQ,dbaasp_hemolysis,3.0,HEMOLYTIC,DBAASP_v3_2meas
+DBAASP-113,KLLAKAALKWLLKALKAA,dbaasp_hemolysis,3.3,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1793,GIGAVLGVLTTGLPALISWIKRKRQQ,dbaasp_hemolysis,3.4,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-111,KAIAKSILKWIKSIAKAI,dbaasp_hemolysis,3.6,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-115,KALKKLLAKWLAAAKALL,dbaasp_hemolysis,3.9,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1792,GIGAVLEVLTTGLPALISWIKRKRQQ,dbaasp_hemolysis,4.4,HEMOLYTIC,DBAASP_v3_2meas
+DBAASP-112,KALAALLKKWAKLLAALK,dbaasp_hemolysis,4.9,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-2718,YKLLKKLLKKLKKLLKKL,dbaasp_hemolysis,6.1,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-2719,YKLLKWLLLKLKALLEKL,dbaasp_hemolysis,6.3,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1999,KWARLWRWFRITRWLWYIK,dbaasp_hemolysis,7.0,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1791,GIGAVLDVLTTGLPALISWIKRKRQQ,dbaasp_hemolysis,7.3,HEMOLYTIC,DBAASP_v3_2meas
+DBAASP-948,FWGALAKGALKLIPSLFSSFSKKD,dbaasp_hemolysis,8.6,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-970,GLFGKLIKKFLRKAISYAVKKARGKH,dbaasp_hemolysis,8.9,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1418,FALALKALKKLLKKLKKLAKKAL,dbaasp_hemolysis,10.0,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1468,FLSLALAALPKFLCLVFKKC,dbaasp_hemolysis,11.1,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1470,FLPLLLAGLPKLLCLFFKKC,dbaasp_hemolysis,11.4,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-2201,FLPILAGLAAKIVPKLFCLATKKC,dbaasp_hemolysis,12.8,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-3136,GLLKLIKTLL,dbaasp_hemolysis,13.0,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1469,FLSLALAALPKLFCLIFKKC,dbaasp_hemolysis,15.7,HEMOLYTIC,DBAASP_v3_3meas
+DBAASP-1001,FLGLLFHGVHHVGKWIHGLIHGHH,dbaasp_hemolysis,16.0,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1465,FIGAIARLLSKIF,dbaasp_hemolysis,17.1,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-536,GLFGKLIKKFGRKAISYAVKKARGKH,dbaasp_hemolysis,17.4,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1240,INWLKLGKAIIDAL,dbaasp_hemolysis,18.0,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-117,KALAKALAKLWKALAKAA,dbaasp_hemolysis,18.7,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1030,KLALKLALKALKAALKLA,dbaasp_hemolysis,18.8,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-11,RVKRVWPLVIRTVIAGYNLYRAIKKK,dbaasp_hemolysis,18.9,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-2156,AIGSILGALAKGLPTLISWIKNR,dbaasp_hemolysis,19.1,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1421,FAKLLAKLAKKLL,dbaasp_hemolysis,19.2,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-116,KITLKLAIKAWKLALKAA,dbaasp_hemolysis,19.8,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-1422,FAKLLAKALKKLL,dbaasp_hemolysis,20.0,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-103,KLALKLALKAWKAALKLA,dbaasp_hemolysis,21.5,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-2453,GLPVCGETCFGGTCNTPGCSCETWPVCSRN,dbaasp_hemolysis,23.2,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-3140,GLLKWIKTLL,dbaasp_hemolysis,23.7,HEMOLYTIC,DBAASP_v3_2meas
+DBAASP-2712,YKLLKLLLPKLKGLLFKL,dbaasp_hemolysis,24.0,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-3102,LLKKLLKKLLKKLLKK,dbaasp_hemolysis,24.0,HEMOLYTIC,DBAASP_v3_1meas
+DBAASP-3100,FFRRFFRRFFRRFFRR,dbaasp_hemolysis,26.0,MODERATE,DBAASP_v3_1meas
+DBAASP-107,ALWKTLLKKVLKAAAK,dbaasp_hemolysis,26.7,MODERATE,DBAASP_v3_2meas
+DBAASP-2209,FLGSIVGALASALPSLISKIRN,dbaasp_hemolysis,26.7,MODERATE,DBAASP_v3_1meas
+DBAASP-2714,YKLLKLLLPKLKGLLIKL,dbaasp_hemolysis,28.0,MODERATE,DBAASP_v3_1meas
+DBAASP-2452,GIPCGESCVWIPCITSAIGCSCKSKVCYRN,dbaasp_hemolysis,28.3,MODERATE,DBAASP_v3_1meas
+DBAASP-3137,GLLKIIKTLL,dbaasp_hemolysis,29.0,MODERATE,DBAASP_v3_1meas
+DBAASP-1106,GEILCNLCTGLINTLENLLTTKRKRQQ,dbaasp_hemolysis,30.7,MODERATE,DBAASP_v3_1meas
+DBAASP-3150,GLLKVIKVLL,dbaasp_hemolysis,32.0,MODERATE,DBAASP_v3_1meas
+DBAASP-2204,FLSGIVGMLGKLF,dbaasp_hemolysis,34.5,MODERATE,DBAASP_v3_1meas
+DBAASP-2370,FFHHIFRGIVHVGKTIHRLVTG,dbaasp_hemolysis,38.6,MODERATE,DBAASP_v3_2meas
+DBAASP-911,WKLFKKILKWL,dbaasp_hemolysis,39.1,MODERATE,DBAASP_v3_1meas
+DBAASP-3326,ILGIITSLLKSL,dbaasp_hemolysis,39.4,MODERATE,DBAASP_v3_1meas
+DBAASP-2255,FFGKVLKLIRKIF,dbaasp_hemolysis,40.2,MODERATE,DBAASP_v3_2meas
+DBAASP-701,GIWDTIKSMGKVFAGLILQNL,dbaasp_hemolysis,41.5,MODERATE,DBAASP_v3_1meas
+DBAASP-691,KIAGKIAKIAGKIAKIAGKIA,dbaasp_hemolysis,42.0,MODERATE,DBAASP_v3_1meas
+DBAASP-2000,KWAKKWKWFAKAAWKWYKK,dbaasp_hemolysis,42.0,MODERATE,DBAASP_v3_1meas
+DBAASP-3104,FLSIIAKVLGSLF,dbaasp_hemolysis,42.2,MODERATE,DBAASP_v3_1meas
+DBAASP-2159,ILGKIWEGIKSLF,dbaasp_hemolysis,43.6,MODERATE,DBAASP_v3_1meas
+DBAASP-3105,FLPLVTMLLGKLF,dbaasp_hemolysis,44.8,MODERATE,DBAASP_v3_1meas
+DBAASP-700,GIWDTIKSMGKVFAGAILQNL,dbaasp_hemolysis,45.2,MODERATE,DBAASP_v3_1meas
+DBAASP-912,WKLFKKILKFL,dbaasp_hemolysis,46.8,MODERATE,DBAASP_v3_1meas
+DBAASP-1889,FLPIVAGLAANFLPKIVCKITKKC,dbaasp_hemolysis,49.2,MODERATE,DBAASP_v3_1meas
+DBAASP-1890,FLSTLLNVASNVVPTLICKITKKC,dbaasp_hemolysis,49.5,MODERATE,DBAASP_v3_1meas
+DBAASP-1892,FLPMLAGLAANFLPKIVCKITKKC,dbaasp_hemolysis,49.8,MODERATE,DBAASP_v3_1meas
+DBAASP-167,KWKLFKKIGIGAVLKVLTTGLPALKLTK,dbaasp_hemolysis,50.0,MODERATE,DBAASP_v3_1meas
+DBAASP-2298,GLLGLLGSVVSHVVPAIVGHF,dbaasp_hemolysis,51.8,MODERATE,DBAASP_v3_1meas
+DBAASP-1978,FLAGLIGGLAKML,dbaasp_hemolysis,52.1,MODERATE,DBAASP_v3_1meas
+DBAASP-2713,YKLLKLLLPKLKPLLFKL,dbaasp_hemolysis,55.0,MODERATE,DBAASP_v3_1meas
+DBAASP-2508,GFMKYIGPLIPHAVKAIKKLI,dbaasp_hemolysis,58.4,MODERATE,DBAASP_v3_1meas
+DBAASP-3135,GLLKVIKTLL,dbaasp_hemolysis,60.0,MODERATE,DBAASP_v3_1meas
+DBAASP-913,WKLFKKILKYL,dbaasp_hemolysis,62.2,MODERATE,DBAASP_v3_1meas
+DBAASP-999,GFWGKLFKLGLHGIGLLHLHL,dbaasp_hemolysis,64.0,MODERATE,DBAASP_v3_1meas
+DBAASP-1320,GLFDVIKKVASVIGGL,dbaasp_hemolysis,64.6,MODERATE,DBAASP_v3_1meas
+DBAASP-2715,YKLLKLLLPKLKPLLIKL,dbaasp_hemolysis,69.0,MODERATE,DBAASP_v3_1meas
+DBAASP-2264,VSRRRRRRGGRRRR,dbaasp_hemolysis,69.9,MODERATE,DBAASP_v3_1meas
+DBAASP-3138,GLLKFIKTLL,dbaasp_hemolysis,77.0,MODERATE,DBAASP_v3_1meas
+DBAASP-2203,NFLGTLVNLAKKIL,dbaasp_hemolysis,77.2,MODERATE,DBAASP_v3_1meas
+DBAASP-121,GIGKFIHAAKKFGKLFIGEIMNS,dbaasp_hemolysis,80.2,MODERATE,DBAASP_v3_1meas
+DBAASP-3143,GLLKRIKLLL,dbaasp_hemolysis,81.0,MODERATE,DBAASP_v3_1meas
+DBAASP-2198,GLFGKILGVGKKVLCGLSGMC,dbaasp_hemolysis,83.2,MODERATE,DBAASP_v3_1meas
+DBAASP-108,ALWKTLLKKVLKA,dbaasp_hemolysis,86.2,MODERATE,DBAASP_v3_2meas
+DBAASP-3379,LNWGALLKHLLK,dbaasp_hemolysis,88.6,MODERATE,DBAASP_v3_1meas
+DBAASP-1460,FLPVILPVIGKLLSGIL,dbaasp_hemolysis,89.6,MODERATE,DBAASP_v3_1meas
+DBAASP-125,GIGKFLHTLKTFGKKWVGEIMNS,dbaasp_hemolysis,90.7,MODERATE,DBAASP_v3_1meas
+DBAASP-704,GIWKTIKSMGKVFAGAILQNL,dbaasp_hemolysis,91.0,MODERATE,DBAASP_v3_1meas
+DBAASP-1477,FLIGMTQGLICLITRKC,dbaasp_hemolysis,91.7,MODERATE,DBAASP_v3_1meas
+DBAASP-2162,ILGKIWKGIKSLF,dbaasp_hemolysis,92.2,MODERATE,DBAASP_v3_3meas
+DBAASP-497,KRLFKKLKFSLRKY,dbaasp_hemolysis,94.6,MODERATE,DBAASP_v3_4meas
+DBAASP-3124,GLLKRIKALL,dbaasp_hemolysis,97.0,MODERATE,DBAASP_v3_1meas
+DBAASP-3145,GLLKRIKFLL,dbaasp_hemolysis,98.0,MODERATE,DBAASP_v3_1meas
+DBAASP-3147,GLLKRIKWLL,dbaasp_hemolysis,98.0,MODERATE,DBAASP_v3_1meas
+DBAASP-1891,FLSTLLNVASKVVPTLFCKITKKC,dbaasp_hemolysis,99.5,MODERATE,DBAASP_v3_1meas
+DBAASP-646,FLWALFKVASK,dbaasp_hemolysis,100.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3177,SLLSLIRKLIT,dbaasp_hemolysis,100.5,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2335,HFLGTLVNLAKKIL,dbaasp_hemolysis,101.8,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1127,FLPIVGKLLSGLL,dbaasp_hemolysis,102.7,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2300,GFKDLLKGAAKALVKTVLF,dbaasp_hemolysis,111.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3287,GKWMSLLKHILK,dbaasp_hemolysis,113.4,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3144,GLLKRIKILL,dbaasp_hemolysis,119.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-932,GLLALISWIKRKRQQ,dbaasp_hemolysis,120.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-498,LLLFLLKKRKKRKY,dbaasp_hemolysis,120.1,SELECTIVE,DBAASP_v3_4meas
+DBAASP-3377,LNWGAFLKHFFK,dbaasp_hemolysis,122.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2253,IKWKKLLRAAKRIL,dbaasp_hemolysis,130.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1129,FLPIIGKLLSGLL,dbaasp_hemolysis,131.4,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3123,GLLKRIATLL,dbaasp_hemolysis,132.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-496,KRLFKELLFSLRKY,dbaasp_hemolysis,134.4,SELECTIVE,DBAASP_v3_3meas
+DBAASP-2292,GFLGPLLKLGLKGAAKLLPQLLPSRQQ,dbaasp_hemolysis,142.9,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2291,GFLGPLLKLGLKGVAKVIPHLIPSRQQ,dbaasp_hemolysis,144.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-123,GIGKFIHSVKKWGKTFIGEIMNS,dbaasp_hemolysis,144.4,SELECTIVE,DBAASP_v3_1meas
+DBAASP-904,WKLFKKILKVL,dbaasp_hemolysis,147.2,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2504,GFMKYIGPLIPHAVKAISDLI,dbaasp_hemolysis,148.4,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2290,GVWTTILGGLKKFAKGGLEALTNPK,dbaasp_hemolysis,156.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2294,FLGALLGPLMNLLQ,dbaasp_hemolysis,157.5,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3121,GLLKAIKTLL,dbaasp_hemolysis,161.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3365,LNWKAILKHIIK,dbaasp_hemolysis,162.4,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3103,FLSSIGKILGNLL,dbaasp_hemolysis,164.9,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3360,LNWGAILKHIIK,dbaasp_hemolysis,167.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1783,GIGKFLHAAKKFAKAFVAEIMNS,dbaasp_hemolysis,175.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2304,GFRDVLKGAAKAFVKTVAGHIAN,dbaasp_hemolysis,189.6,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2001,KFAKKFKWFAKAAFKFFKK,dbaasp_hemolysis,195.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-109,KLALKAALKAWKAAAKLA,dbaasp_hemolysis,199.7,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1419,FAKKLAKLAKKLAKLAL,dbaasp_hemolysis,208.7,SELECTIVE,DBAASP_v3_1meas
+DBAASP-14,GIWDTIKSMGKVFAGKILQNL,dbaasp_hemolysis,208.8,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1883,FLPIVGKLLSGLF,dbaasp_hemolysis,210.5,SELECTIVE,DBAASP_v3_1meas
+DBAASP-126,GIGHFLHKVKSFGKSWIGEIMNS,dbaasp_hemolysis,211.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2401,SMLSVLKNLGKVGLGFVACKINKKC,dbaasp_hemolysis,212.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2407,ILGKLLSTAAGLLSNL,dbaasp_hemolysis,213.8,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3382,LNWGAWLKHWWK,dbaasp_hemolysis,214.5,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3232,FFFLSRIF,dbaasp_hemolysis,215.2,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1882,FFPIVGKLLSGLF,dbaasp_hemolysis,215.6,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1478,GLFDVVKGVLKGVGKNVAGSLLEQLKCKLSGGC,dbaasp_hemolysis,216.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1795,GIGAVLLVLTTGLPALISWIKRKRQQ,dbaasp_hemolysis,219.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3346,ILPILSLIGGLL,dbaasp_hemolysis,219.9,SELECTIVE,DBAASP_v3_1meas
+DBAASP-699,GIWKTIKSMGKVFAGKILQNL,dbaasp_hemolysis,221.6,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3151,KLLKFIKTLL,dbaasp_hemolysis,231.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3153,KLLKVIKKLL,dbaasp_hemolysis,231.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2593,GIGKFLKKAKKFGKAFVKMKK,dbaasp_hemolysis,238.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2400,SMLSVLKNLGKVGLGFVACKIKKQC,dbaasp_hemolysis,239.9,SELECTIVE,DBAASP_v3_1meas
+DBAASP-952,RRFPWWWPFRR,dbaasp_hemolysis,240.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1887,GLWNTIKEAGKKFAINVLDKIRCGIAGGCKT,dbaasp_hemolysis,247.9,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3157,GLLGPLLKIAAKVGSNLL,dbaasp_hemolysis,248.8,SELECTIVE,DBAASP_v3_1meas
+DBAASP-914,LKLFKKILKVL,dbaasp_hemolysis,255.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1893,GILDKLKEFGISAARGVAQSLLNTASCKLAKTC,dbaasp_hemolysis,255.6,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2289,GFWTTAAEGLKKFAKAGLASILNPK,dbaasp_hemolysis,275.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2507,GFMKYIGPLIPHAVKAISKLI,dbaasp_hemolysis,275.6,SELECTIVE,DBAASP_v3_2meas
+DBAASP-915,KKLFKKILKFL,dbaasp_hemolysis,285.4,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2398,SMLSVLKKLGKVGLGFVACKINKQC,dbaasp_hemolysis,293.2,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1886,SLFSIFKTAAKFVGKNLLKQAGKAGLETLACKAKNEC,dbaasp_hemolysis,294.7,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3142,GLLKRIKVLL,dbaasp_hemolysis,310.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3149,KLLKRIKTLL,dbaasp_hemolysis,312.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2509,GFMKYIKPLIPHAVKAISKLI,dbaasp_hemolysis,319.7,SELECTIVE,DBAASP_v3_2meas
+DBAASP-2395,SMLKVLKNLGKVGLGFVACKINKQC,dbaasp_hemolysis,323.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2399,SMLSVLKNLGKVGLGFVKCKINKQC,dbaasp_hemolysis,325.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2404,GLKEIFKAGLGSLVKGIAAHVAS,dbaasp_hemolysis,328.6,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2210,GIMSLFKGVLKTAGKHVAGSLVDQLKCKITGGC,dbaasp_hemolysis,336.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2367,IWLTALKFLGKHAAKHLAKQQLSKL,dbaasp_hemolysis,355.5,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3065,ILPWKWKWWPWRR,dbaasp_hemolysis,358.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3152,KLLKRIKKLL,dbaasp_hemolysis,369.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1010,SMLSVLKNLGKVGLGFVACKINKQC,dbaasp_hemolysis,371.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-160,VRRFPWWWPFLRR,dbaasp_hemolysis,380.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3146,GLLKRIKNLL,dbaasp_hemolysis,381.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2196,GLFTLIKGAAKLIGKTVAKEAGKTGLELMACKITNQC,dbaasp_hemolysis,382.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1888,FLSTALKVAANVVPTLFCKITKKC,dbaasp_hemolysis,389.4,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2293,GFLGSLLKTGLKVGSNLL,dbaasp_hemolysis,399.7,SELECTIVE,DBAASP_v3_1meas
+DBAASP-874,AGLQFPVGRVHRLLRK,dbaasp_hemolysis,400.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-878,FPVGRVHRLLRK,dbaasp_hemolysis,400.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1748,ILKWKWPWWPWRR,dbaasp_hemolysis,400.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-933,GLSALISWIKRKRQQ,dbaasp_hemolysis,403.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2207,AVDLAKIANKVLSSLF,dbaasp_hemolysis,422.2,SELECTIVE,DBAASP_v3_1meas
+DBAASP-910,KKLFKKILKVL,dbaasp_hemolysis,453.5,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2194,GLLDTFKNLALNAAKSAGVSVLNSLSCKLSKTC,dbaasp_hemolysis,471.5,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2403,GLKEIFKAGLGSLVKGIAAHVAN,dbaasp_hemolysis,481.6,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2510,GFMKYIKPLIPHAVKAIKKLI,dbaasp_hemolysis,481.8,SELECTIVE,DBAASP_v3_2meas
+DBAASP-3068,ILPWKWPWWKWRR,dbaasp_hemolysis,490.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2282,GVIKSVLKGVAKTVALGML,dbaasp_hemolysis,527.6,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1998,KWASLWNWFNITNWLWYIK,dbaasp_hemolysis,550.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3069,ILKWKWKWWPWRR,dbaasp_hemolysis,560.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2197,GLLSGILGAGKNIVCGLSGLC,dbaasp_hemolysis,573.8,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2199,GVLGTVKNLLIGAGKSAAQSVLKTLSCKLSNDC,dbaasp_hemolysis,592.1,SELECTIVE,DBAASP_v3_1meas
+DBAASP-937,GLPALILWIKRKRQQ,dbaasp_hemolysis,620.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-120,GIGKFIHSAKKFGKLFVGEIMNS,dbaasp_hemolysis,652.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-122,GIGKFIHSAKRFGRAWVGEIMNS,dbaasp_hemolysis,776.2,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1977,SLWENFKNAGKQFILNILDKIRCRVAGGCRT,dbaasp_hemolysis,781.4,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2018,KILRGVCKKIMRTFLRRISKDILTGKK,dbaasp_hemolysis,900.0,SELECTIVE,DBAASP_v3_2meas
+DBAASP-3355,GLNALKKVFQGIHEAIKLINNHVQ,dbaasp_hemolysis,966.6,SELECTIVE,DBAASP_v3_2meas
+DBAASP-894,AKKVFKRLEKLFSKIWNWK,dbaasp_hemolysis,974.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2193,GLFSKFAGKGIKNLIFKGVKHIGKEVGMDVIRTGIDVAGCKIKGEC,dbaasp_hemolysis,1021.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1639,GLLRASSVWGRKYYVDLAGCAKA,dbaasp_hemolysis,1242.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-118,GIGKFLHSAKKFGKAWVGEIMNS,dbaasp_hemolysis,1275.4,SELECTIVE,DBAASP_v3_1meas
+DBAASP-930,LLPALISWIKRKRQQ,dbaasp_hemolysis,1450.0,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1970,GAFGDLLKGVAKEAGMKLLNMAQCKLSGKC,dbaasp_hemolysis,1602.8,SELECTIVE,DBAASP_v3_1meas
+DBAASP-1971,SILDKIKNVALGVARGAGTGILKALLCKLDKSC,dbaasp_hemolysis,1617.5,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3141,GLLKKIKTLL,dbaasp_hemolysis,1636.8,SELECTIVE,DBAASP_v3_1meas
+DBAASP-124,GIAKFGKAAAHFGKKWVGELMNS,dbaasp_hemolysis,1713.3,SELECTIVE,DBAASP_v3_1meas
+DBAASP-2281,GVIIDTLKGAAKTVAAELLRKAHCKLTNSC,dbaasp_hemolysis,2187.8,SELECTIVE,DBAASP_v3_1meas
+DBAASP-3116,GLLKRIKTLL,dbaasp_hemolysis,2779.5,SELECTIVE,DBAASP_v3_1meas
diff --git a/scripts/expand_hemolysis_benchmark.py b/scripts/expand_hemolysis_benchmark.py
new file mode 100644
index 00000000..a7036550
--- /dev/null
+++ b/scripts/expand_hemolysis_benchmark.py
@@ -0,0 +1,210 @@
+"""Expand the hemolysis reference set using DBAASP hemolytic activity data.
+
+The within-AMP selectivity benchmark (PR #113/#114) runs on 42 peptides:
+14 hemolytic (HC50 < 25 ug/ml) vs 21 selective (HC50 >= 100 ug/ml) = 35
+for the binary AUROC task. The dedicated hemolysis risk scorer's detection
+AUROC=0.9218 (CI 0.82-0.99) is based on this small set.
+
+DBAASP (data/novelty_db/hemolytic-and-cytotoxic-activities.csv) contains
+hemolytic activity measurements for ~880 unique peptides against human
+erythrocytes. This script extracts 50%-hemolysis (HC50) values, converts
+all measurements to ug/ml using peptide molecular weight, deduplicates by
+sequence (median of multiple measurements), and appends new peptides to
+examples/validation/hemolysis_reference.csv.
+
+Unit conversion: ug/ml = uM * MW(kDa) = uM * MW(Da) / 1000
+  MW computed from amino acid residue masses (average isotopic masses,
+  water subtracted for peptide bonds, water added for terminal groups).
+
+Classification thresholds (unchanged from existing reference):
+  HEMOLYTIC: HC50 < 25 ug/ml
+  MODERATE:  25 <= HC50 < 100 ug/ml
+  SELECTIVE: HC50 >= 100 ug/ml
+
+Usage: python scripts/expand_hemolysis_benchmark.py
+"""
+
+import csv
+import re
+from collections import defaultdict
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parent.parent
+DBAASP_CSV = REPO_ROOT / "data" / "novelty_db" / "hemolytic-and-cytotoxic-activities.csv"
+OUTPUT_CSV = REPO_ROOT / "examples" / "validation" / "hemolysis_reference.csv"
+
+# Amino acid residue masses (average, monoisotopic would also work for relative ranking).
+# Values from standard biochemical tables (e.g., Sigma-Aldrich residue mass chart).
+# Includes water (18.02) in each entry; peptide bond subtracts one water.
+_AA_MW: dict[str, float] = {
+    "A": 89.09, "R": 174.20, "N": 132.12, "D": 133.10, "C": 121.16,
+    "Q": 146.15, "E": 147.13, "G": 75.03, "H": 155.16, "I": 131.17,
+    "L": 131.17, "K": 146.19, "M": 149.21, "F": 165.19, "P": 115.13,
+    "S": 105.09, "T": 119.12, "W": 204.23, "Y": 181.19, "V": 117.15,
+}
+_H2O = 18.02
+
+
+def peptide_mw(seq: str) -> float:
+    """Molecular weight of a peptide in Daltons.
+
+    Sum of amino acid masses minus water for each peptide bond,
+    plus one water for the intact termini (already included in residue masses).
+    """
+    return sum(_AA_MW[aa] for aa in seq) - (len(seq) - 1) * _H2O
+
+
+def parse_concentration(raw: str) -> tuple[float, bool] | None:
+    """Extract numeric value from a concentration string.
+
+    Returns (value, is_greater_than) or None if unparseable.
+    Handles formats like "6", "0.5±0.2", ">200", "39±1.5".
+    """
+    s = raw.strip()
+    is_gt = ">" in s
+    s = s.replace("±", " ").replace(">", "").replace("<", "").strip()
+    m = re.match(r"([\d.]+)", s)
+    if m:
+        return float(m.group(1)), is_gt
+    return None
+
+
+def classify_hemolysis(hc50_ugml: float) -> str:
+    """Classify a peptide by HC50 in ug/ml."""
+    if hc50_ugml < 25:
+        return "HEMOLYTIC"
+    elif hc50_ugml < 100:
+        return "MODERATE"
+    else:
+        return "SELECTIVE"
+
+
+def load_existing() -> tuple[set[str], list[dict]]:
+    """Load existing hemolysis reference entries."""
+    existing_seqs: set[str] = set()
+    existing_rows: list[dict] = []
+    with open(OUTPUT_CSV, newline="", encoding="utf-8") as f:
+        for row in csv.DictReader(f):
+            seq = row["sequence"].strip().upper()
+            existing_seqs.add(seq)
+            existing_rows.append(row)
+    return existing_seqs, existing_rows
+
+
+def extract_dbaasp() -> dict[str, tuple[str, float, int]]:
+    """Extract HC50 values from DBAASP hemolysis data.
+
+    Returns: {sequence: (peptide_id, median_hc50_ugml, n_measurements)}
+    Only includes human erythrocyte 50% hemolysis measurements.
+    """
+    raw: dict[str, list[float]] = defaultdict(list)
+    pid_map: dict[str, str] = {}
+
+    with open(DBAASP_CSV, newline="", encoding="utf-8") as f:
+        reader = csv.DictReader(f)
+        for row in reader:
+            seq = row.get("Peptide Sequence", "").strip().upper()
+            pid = row.get("Peptide ID", "").strip()
+            target = row.get("Target Cell", "").strip()
+            measure = row.get("Activity Measure for Lysis", "").strip()
+            conc_raw = row.get("Peptide Concentration", "").strip()
+            unit = row.get("Unit", "").strip()
+
+            if not seq:
+                continue
+            if "human erythrocyte" not in target.lower():
+                continue
+            if "50%" not in measure.lower() and "hc50" not in measure.lower():
+                continue
+            if not all(aa in "ACDEFGHIKLMNPQRSTVWY" for aa in seq):
+                continue
+            if len(seq) < 5 or len(seq) > 60:
+                continue
+
+            parsed = parse_concentration(conc_raw)
+            if parsed is None:
+                continue
+            val, is_gt = parsed
+            # Skip "greater than" values — they are lower bounds on HC50, not exact
+            if is_gt:
+                continue
+
+            # Convert to ug/ml
+            mw = peptide_mw(seq)
+            if unit == "µM":
+                hc50_ugml = val * mw / 1000.0
+            elif unit in ("µg/ml", "ug/ml", "ug/mL"):
+                hc50_ugml = val
+            else:
+                continue
+
+            raw[seq].append(hc50_ugml)
+            if seq not in pid_map:
+                pid_map[seq] = pid
+
+    # Deduplicate: take median of multiple measurements
+    result: dict[str, tuple[str, float, int]] = {}
+    for seq, vals in raw.items():
+        median = sorted(vals)[len(vals) // 2]
+        result[seq] = (pid_map[seq], round(median, 1), len(vals))
+
+    return result
+
+
+def main() -> None:
+    existing_seqs, existing_rows = load_existing()
+    print(f"Existing hemolysis reference: {len(existing_rows)} peptides")
+
+    dbaasp = extract_dbaasp()
+    print(f"DBAASP extracted (human erythrocyte HC50, standard AAs, 5-60aa): {len(dbaasp)}")
+
+    # Remove sequences already in the reference
+    new_seqs = {s: v for s, v in dbaasp.items() if s not in existing_seqs}
+    print(f"New unique peptides (after dedup vs existing): {len(new_seqs)}")
+
+    # Classify and build rows
+    new_rows: list[dict] = []
+    for seq, (pid, hc50, n_meas) in sorted(new_seqs.items(), key=lambda x: x[1][1]):
+        cls = classify_hemolysis(hc50)
+        new_rows.append({
+            "id": f"DBAASP-{pid}",
+            "sequence": seq,
+            "family": "dbaasp_hemolysis",
+            "hc50_ugml": str(hc50),
+            "hemolysis_class": cls,
+            "reference": f"DBAASP_v3_{n_meas}meas",
+        })
+
+    # Count by class
+    from collections import Counter
+    new_classes = Counter(r["hemolysis_class"] for r in new_rows)
+    print(f"New peptides by class: {dict(new_classes)}")
+
+    # Append to CSV
+    with open(OUTPUT_CSV, "a", newline="", encoding="utf-8") as f:
+        writer = csv.DictWriter(
+            f,
+            fieldnames=["id", "sequence", "family", "hc50_ugml", "hemolysis_class", "reference"],
+        )
+        for row in new_rows:
+            writer.writerow(row)
+
+    print(f"Appended {len(new_rows)} peptides to {OUTPUT_CSV}")
+
+    # Summary after merge
+    all_seqs, all_rows = load_existing()
+    all_classes = Counter(r["hemolysis_class"] for r in all_rows)
+    hemo = sum(1 for r in all_rows if float(r["hc50_ugml"]) < 25)
+    sel = sum(1 for r in all_rows if float(r["hc50_ugml"]) >= 100)
+    border = sum(1 for r in all_rows if 25 <= float(r["hc50_ugml"]) < 100)
+    print(f"\nMerged reference: {len(all_rows)} peptides")
+    print(f"  HEMOLYTIC (HC50<25): {hemo}")
+    print(f"  MODERATE (25-100): {border}")
+    print(f"  SELECTIVE (HC50>=100): {sel}")
+    print(f"  Binary task: {hemo} hemolytic vs {sel} selective = {hemo + sel} total")
+    print(f"  Previous binary task: 14 vs 21 = 35")
+    print(f"  Improvement: {hemo + sel} / 35 = {(hemo + sel) / 35:.1f}x")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/src/openamp_foundry/scoring/expert.py b/src/openamp_foundry/scoring/expert.py
index 3cd20f41..65d62335 100644
--- a/src/openamp_foundry/scoring/expert.py
+++ b/src/openamp_foundry/scoring/expert.py
@@ -203,7 +203,7 @@ def kmer_prior_art(sequence: str, kmer_index: set[str], k: int = 5) -> dict:
     "serum_stability":    0.05,   # proteolytic longevity (informational, low weight)
     "hinge_selectivity":  0.08,   # central helix-hinge bonus (expert motif)
     "motif_novelty":      0.10,   # k-mer prior-art (local novelty beyond global identity)
-    "hemolysis_safety":   0.10,   # dedicated hemolysis risk triage (detection AUROC=0.92)
+    "hemolysis_safety":   0.10,   # dedicated hemolysis risk triage (expanded n=179: AUROC=0.565, CI 0.47-0.66, not significant)
 }
 
 
diff --git a/src/openamp_foundry/scoring/hemolysis.py b/src/openamp_foundry/scoring/hemolysis.py
index 8c9c35f5..89746159 100644
--- a/src/openamp_foundry/scoring/hemolysis.py
+++ b/src/openamp_foundry/scoring/hemolysis.py
@@ -40,14 +40,19 @@
      aromatic fraction by construction (chymotrypsin cleaves after F/W/Y)
      and is excluded to avoid double-counting.
 
-Combined 4-component weighted score achieves detection AUROC=0.8912
-(bootstrap CI₉₅: 0.77-0.97), a substantial improvement over the safety
-scorer's 0.3844. The CI lower bound (0.77) clears the 0.5 significance
-threshold with margin — the first pipeline score to do so on hemolysis.
+Combined 4-component weighted score achieved detection AUROC=0.8912
+(bootstrap CI₉₅: 0.77-0.97) on the original 42-peptide reference set.
+However, expansion to 238 peptides (v0.5.11, DBAASP data) revealed this
+was small-sample inflation: on n=179 (54 hemolytic vs 125 selective),
+detection AUROC drops to 0.5650 (CI 0.47-0.66) — direction correct but
+NOT statistically significant. The scorer retains weak directional signal
+but should not be trusted as a standalone hemolysis detector.
 
 IMPORTANT LIMITATIONS:
-  - The reference set is small (14 hemolytic vs 21 selective, n=35).
-    The CI is wide. A larger reference set would tighten confidence.
+  - The original reference set was small (14 hemolytic vs 21 selective, n=35).
+    The expanded set (54 hemolytic vs 125 selective, n=179) revealed that the
+    original AUROC=0.92 was small-sample inflation. On the expanded set,
+    detection AUROC=0.565 (CI 0.47-0.66) — direction correct but not significant.
   - HC50 values are approximate literature values with high inter-assay
     variability (RBC source, buffer, incubation time, concentration range).
   - This is a triage signal, NOT a hemolysis predictor. Every candidate
diff --git a/tests/test_hemolysis_risk.py b/tests/test_hemolysis_risk.py
index 6c8133ac..e4b947ec 100644
--- a/tests/test_hemolysis_risk.py
+++ b/tests/test_hemolysis_risk.py
@@ -6,7 +6,13 @@
 
 Key finding being tested: the safety scorer fails hemolysis detection
 (AUROC=0.3844), but the dedicated hemolysis risk scorer achieves
-detection AUROC=0.9218 (CI=[0.82, 0.99]) on the same reference set.
+detection AUROC=0.9218 (CI=[0.82, 0.99]) on the original 42-peptide reference set.
+
+IMPORTANT: The expanded 238-peptide reference set (v0.5.11, DBAASP data) revealed
+this was small-sample inflation. On n=179 (54 hemolytic vs 125 selective), the
+detection AUROC drops to 0.5650 (CI 0.47-0.66) — direction correct but no longer
+statistically significant. The scorer retains weak directional signal but should
+NOT be trusted as a hemolysis detector without further improvement.
 """
 from __future__ import annotations
 
diff --git a/tests/test_selectivity_benchmark.py b/tests/test_selectivity_benchmark.py
index 7e1b2475..072eb555 100644
--- a/tests/test_selectivity_benchmark.py
+++ b/tests/test_selectivity_benchmark.py
@@ -7,6 +7,13 @@
 AMP-vs-decoy discrimination (where safety/synthesis were anti-signal); this tests
 within-AMP selectivity, the task those scorers were designed for.
 
+Reference set expanded from 42 to 238 peptides (54 hemolytic vs 125 selective)
+in v0.5.11 using DBAASP human erythrocyte hemolysis data. The expanded set
+revealed that the hemolysis risk scorer's original detection AUROC=0.9218
+(CI 0.82-0.99) on n=35 was small-sample inflation; on n=179 the detection
+AUROC drops to 0.5650 (CI 0.47-0.66) — no longer statistically significant.
+This is exactly the kind of self-deception the project is designed to catch.
+
 All results are computational. No biological activity is implied.
 """
 from __future__ import annotations
@@ -85,10 +92,26 @@ def test_expert_composite_verdict_present(self, result):
     def test_hemolysis_risk_verdict_present(self, result):
         assert "hemolysis risk" in result["hemolysis_risk_verdict"].lower()
 
-    def test_hemolysis_risk_is_significant_detector(self, result):
-        """The dedicated hemolysis risk scorer should be a significant risk detector."""
-        assert "hemolysis_risk" in result["risk_detectors"]
-        assert result["per_score_auroc"]["hemolysis_risk"]["detection_ci95_lo"] > 0.5
+    def test_hemolysis_risk_direction_correct(self, result):
+        """The hemolysis risk scorer should rank hemolytic AMPs higher than selective on average.
+
+        On the expanded reference set (n=179 binary), the detection AUROC is 0.5650
+        (CI 0.47-0.66) — direction is correct but not statistically significant.
+        The original n=35 AUROC=0.9218 was small-sample inflation, corrected by
+        the DBAASP expansion. This test verifies the direction survives expansion.
+        """
+        info = result["per_score_auroc"]["hemolysis_risk"]
+        # Direction: hemolytic should score higher (more risk) than selective
+        assert info["mean_hemolytic"] is not None
+        assert info["mean_selective"] is not None
+        assert info["mean_hemolytic"] > info["mean_selective"], (
+            f'Hemolysis risk mean_hemolytic={info["mean_hemolytic"]} should exceed '
+            f'mean_selective={info["mean_selective"]} - direction must be correct'
+        )
+        # Detection AUROC should be above 0.5 (direction correct)
+        assert info["hemolysis_detection_auroc"] > 0.50, (
+            f'Detection AUROC={info["hemolysis_detection_auroc"]} should be > 0.50'
+        )
 
     def test_hemolysis_risk_scores_present(self, result):
         """Per-peptide hemolysis risk scores should be available for inspection."""