docs: fix wave0.5 artifact truth and snapshot

docs/METRICS_CURRENT.md

@@ -161,7 +161,9 @@ representative of diverse AMP classes not covered by the helic-centric scorer.
 AntiCP 2.0 detects anticancer peptide (ACP) patterns, not antimicrobial activity directly.
 Mitigation: Wave 0.5b designs avoid aromatic residues and pure amphipathic helix.
 
-Machine-readable: `outputs/wave0_5_external_predict_results.csv`, `outputs/wave0_5_external_consensus.csv`
+Current-state summary is documented here. Wave 0.5 machine-readable CSV outputs are
+generated locally via `make wave0-5-fill-external` when `outputs/wave05_combined_consensus.csv`
+is present; they are not guaranteed to be committed in every checkout.
 
 ---
 

docs/WAVE_0_5_EXTERNAL_PREDICTOR_SUMMARY.md

@@ -7,17 +7,20 @@
 > HemoFinder predicts hemolysis risk, NOT antimicrobial activity.
 
 Generated: 2026-06-29
-Status: **PENDING — External predictor submissions required**
+Updated: 2026-07-02
+Status: **HISTORICAL TEMPLATE ONLY — retained as the pre-submission plan**
 
 ---
 
 ## Purpose
 
-This document describes the external predictor screen for the Wave 0.5 shortlist
-(60 candidates across 10 new seed families). Results are PENDING because these
-predictors require manual web submission to published servers.
+This document describes the original pre-submission external predictor plan for the
+Wave 0.5 shortlist (60 candidates across 10 new seed families). It is preserved as
+historical process context, not as the current result state.
 
-The screen must be completed before final Wave 1 panel selection (Phase 7).
+The external screen was completed later and the authoritative current-state summary
+now lives in [`docs/METRICS_CURRENT.md`](/Users/james/.codex/worktrees/9f2e/openamp-foundry/docs/METRICS_CURRENT.md)
+and [`docs/ROADMAP.md`](/Users/james/.codex/worktrees/9f2e/openamp-foundry/docs/ROADMAP.md).
 
 ---
 
@@ -108,24 +111,25 @@ confirmation required before synthesis.
 
 ---
 
-## How to Fill In
+## Historical Submission Procedure
 
 1. Export `outputs/wave0_5_internal_shortlist.csv` → extract sequences column
 2. Create FASTA file from all 60 sequences
 3. Submit to each tool's web interface (see URLs above)
 4. Record Y/N (or probability score) per candidate per tool
 5. Fill in `outputs/wave0_5_external_predict_results.csv`
-6. Run: `make wave0-5-external-consensus` (Makefile target — to be added in Phase 10)
+6. Run `make wave0-5-fill-external` to derive the Wave 0.5 result files when
+   `outputs/wave05_combined_consensus.csv` is available in the local checkout.
 
 ---
 
-## Machine-Readable Status
+## Historical Artifact Expectations
 
 | File | Status |
 |---|---|
-| `outputs/wave0_5_external_predict_results.csv` | PENDING (60 rows, all PENDING) |
-| `outputs/wave0_5_external_consensus.csv` | PENDING (internal OpenAMP scores available) |
-| `outputs/wave0_5_safety_consensus.csv` | PENDING (internal estimates in internal_safety_estimate column) |
+| `outputs/wave0_5_external_predict_results.csv` | Generated only after running `make wave0-5-fill-external` with local source data |
+| `outputs/wave0_5_external_consensus.csv` | Generated only after running `make wave0-5-fill-external` with local source data |
+| `outputs/wave0_5_safety_consensus.csv` | Generated only after running `make wave0-5-fill-external` with local source data |
 
 ---
 

outputs/metrics_snapshot.json

@@ -0,0 +1,162 @@
+{
+  "benchmark_inputs": {
+    "amp_csv": "examples/validation/known_amps.csv",
+    "cluster_threshold": 0.7,
+    "decoy_csv": "examples/validation/random_background.csv",
+    "hemolysis_csv": "examples/validation/hemolysis_reference.csv",
+    "n_bootstrap": 2000,
+    "phase3_config": "configs/phase3.yaml",
+    "standard_config": "configs/pipeline.yaml"
+  },
+  "cluster_split": {
+    "cluster_aware_ci95": [
+      0.7061,
+      0.8526
+    ],
+    "full_auroc": 0.7832,
+    "held_out_auroc": 0.8734,
+    "interpretation": "STRONG — AUROC > 0.70 and cluster-aware CI lower bound > 0.65. Signal survives near-duplicate de-inflation.",
+    "n_amps_in_multi_member_clusters": 33,
+    "n_clusters": 76,
+    "n_multi_member_clusters": 14,
+    "representative_auroc": 0.7607,
+    "representative_ci95": [
+      0.6854,
+      0.8301
+    ],
+    "standard_ci95": [
+      0.717,
+      0.8423
+    ]
+  },
+  "expert_ablation": {
+    "anti_signal_components": [
+      "synthesis",
+      "safety",
+      "hemolysis_safety",
+      "serum_stability"
+    ],
+    "delta_auroc": -0.0713,
+    "ensemble_auroc": 0.7832,
+    "expert_auroc": 0.7119,
+    "near_zero_components": [
+      "hinge_selectivity",
+      "novelty",
+      "motif_novelty",
+      "boman_activity"
+    ],
+    "signal_bearing_components": [
+      "activity",
+      "selectivity_proxy"
+    ],
+    "verdict": "Expert composite (0.7119) scores LOWER than ensemble (0.7832) by -0.0713 AUROC. The added components degrade discrimination on this benchmark. The expert composite should NOT replace the ensemble for AMP/non-AMP triage without further evidence."
+  },
+  "phase3": {
+    "auprc": 0.7933,
+    "auroc": 0.7448,
+    "interpretation": "STRONG — model has meaningful discriminative power (AUROC > 0.70)",
+    "n_negatives": 96,
+    "n_positives": 95,
+    "n_total": 191
+  },
+  "schema_version": 1,
+  "selectivity": {
+    "expert_composite_verdict": "Expert composite detects hemolysis risk better than ensemble (detection AUROC 0.5459 vs 0.4201). The added selectivity/safety components earn their keep on within-AMP ranking.",
+    "n_border": 59,
+    "n_hemolytic": 54,
+    "n_selective": 125,
+    "risk_detectors": [
+      "serum_stability"
+    ],
+    "risk_indicators": [
+      "expert_composite",
+      "safety",
+      "synthesis",
+      "boman_activity",
+      "selectivity_proxy",
+      "hemolysis_risk"
+    ],
+    "safety_verdict": "Safety scorer shows weak hemolysis detection (detection AUROC=0.5116, CI lo=0.4321 < 0.5). Signal is present but not statistically significant. The melittin blind spot remains: 1D hydrophobic moment cannot capture all hemolysis mechanisms.",
+    "selectivity_proxy_verdict": "Selectivity proxy shows weak hemolysis detection (detection AUROC=0.5744, CI lo=0.4954). Trend is correct but not significant at this sample size."
+  },
+  "standard": {
+    "auprc": 0.8164,
+    "auroc": 0.7832,
+    "interpretation": "STRONG — model has meaningful discriminative power (AUROC > 0.70)",
+    "n_negatives": 96,
+    "n_positives": 95,
+    "n_total": 191,
+    "recall_at_10": 0.1053,
+    "recall_at_20": 0.2105,
+    "recall_at_43": 0.4211
+  },
+  "triage": {
+    "best_scorer": "selectivity_proxy",
+    "per_scorer": {
+      "activity": {
+        "hemolytic_vs_decoy": 0.934,
+        "selective_vs_decoy": 0.8846,
+        "selective_vs_hemolytic": 0.43,
+        "triages_correctly": false
+      },
+      "ensemble": {
+        "hemolytic_vs_decoy": 0.8911,
+        "selective_vs_decoy": 0.8481,
+        "selective_vs_hemolytic": 0.4661,
+        "triages_correctly": false
+      },
+      "hemolysis_risk": {
+        "hemolytic_vs_decoy": 0.4924,
+        "selective_vs_decoy": 0.485,
+        "selective_vs_hemolytic": 0.4876,
+        "triages_correctly": false
+      },
+      "safe_weighted_ensemble": {
+        "hemolytic_vs_decoy": 0.8904,
+        "selective_vs_decoy": 0.849,
+        "selective_vs_hemolytic": 0.483,
+        "triages_correctly": false
+      },
+      "safety": {
+        "hemolytic_vs_decoy": 0.3,
+        "selective_vs_decoy": 0.344,
+        "selective_vs_hemolytic": 0.5384,
+        "triages_correctly": false
+      },
+      "selectivity_proxy": {
+        "hemolytic_vs_decoy": 0.7955,
+        "selective_vs_decoy": 0.782,
+        "selective_vs_hemolytic": 0.6095,
+        "triages_correctly": true
+      },
+      "serum_stability": {
+        "hemolytic_vs_decoy": 0.1604,
+        "selective_vs_decoy": 0.2174,
+        "selective_vs_hemolytic": 0.5686,
+        "triages_correctly": false
+      },
+      "synthesis": {
+        "hemolytic_vs_decoy": 0.6336,
+        "selective_vs_decoy": 0.5898,
+        "selective_vs_hemolytic": 0.4692,
+        "triages_correctly": false
+      },
+      "triage_score": {
+        "hemolytic_vs_decoy": 0.9019,
+        "selective_vs_decoy": 0.8629,
+        "selective_vs_hemolytic": 0.4622,
+        "triages_correctly": false
+      }
+    },
+    "top_20_by_ensemble": {
+      "DECOY": 0,
+      "HEMOLYTIC": 6,
+      "SELECTIVE": 14
+    },
+    "top_20_by_triage_score": {
+      "DECOY": 0,
+      "HEMOLYTIC": 4,
+      "SELECTIVE": 16
+    }
+  }
+}

tests/test_docs_consistent.py

@@ -2,8 +2,8 @@
 
 from __future__ import annotations
 
-import re
 import json
+import re
 from pathlib import Path
 
 
@@ -19,6 +19,11 @@
 EXPANDED_PR_REF = "PR #110"
 ORIGINAL_DEMO_AUROC = "0.8420"
 SNAPSHOT_PATH = REPO_ROOT / "outputs" / "metrics_snapshot.json"
+WAVE05_GENERATED_OUTPUTS = [
+    REPO_ROOT / "outputs" / "wave0_5_external_predict_results.csv",
+    REPO_ROOT / "outputs" / "wave0_5_external_consensus.csv",
+    REPO_ROOT / "outputs" / "wave0_5_safety_consensus.csv",
+]
 
 # Patterns that indicate stale "current" usage (not historical context)
 _STALE_CURRENT_PATTERNS: list[str] = [
@@ -144,3 +149,17 @@ def test_doc_wet_lab_probability_has_honest_correction(self):
         assert "55–80%" in text or "55-80" in text or "honest" in text.lower(), (
             "WET_LAB_PROBABILITY.md missing honest correction range"
         )
+
+    def test_doc_wave05_generated_outputs_claimed_only_if_present(self):
+        """Docs must not present missing Wave 0.5 generated CSVs as committed facts."""
+        files_present = all(path.exists() for path in WAVE05_GENERATED_OUTPUTS)
+
+        metrics_text = (DOCS_DIR / "METRICS_CURRENT.md").read_text(encoding="utf-8")
+        summary_text = (DOCS_DIR / "WAVE_0_5_EXTERNAL_PREDICTOR_SUMMARY.md").read_text(
+            encoding="utf-8"
+        )
+
+        if not files_present:
+            assert "Machine-readable: `outputs/wave0_5_external_predict_results.csv`" not in metrics_text
+            assert "PENDING (60 rows, all PENDING)" not in summary_text
+            assert "Generated only after running `make wave0-5-fill-external`" in summary_text