cschanhniem · Jul 9, 2026 · Jul 9, 2026
diff --git a/Makefile b/Makefile
@@ -360,9 +360,18 @@ ip-report:
 		--novelty-csv outputs/novelty_audit_full.csv \
 		--out outputs/ip_report.md
 
-benchmark-card: validate-scoring
-	@echo "Benchmark card is a static document at docs/BENCHMARK_CARD.md"
-	@echo "Run 'make validate-scoring' to update outputs/validate_scoring_report.json"
+benchmark-card:
+	.venv/bin/openamp-foundry benchmark-card \
+		--benchmark-id bench-auroc-001 \
+		--benchmark-name "AMP vs Decoy AUROC" \
+		--metric AUROC \
+		--metric-value 0.82 \
+		--baseline-name "random" \
+		--baseline-value 0.50 \
+		--dataset APD3 \
+		--dataset-size 500 \
+		--validate
+	@echo "Benchmark card complete."
 
 pilot-confident:
 	@if [ -z "$(KEEP)" ]; then \

diff --git a/docs/evidence/METRICS_CURRENT.md b/docs/evidence/METRICS_CURRENT.md
@@ -5,7 +5,8 @@ Machine-readable snapshot: `outputs/metrics_snapshot.json` regenerated with `mak
 > **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:** 2026-07-09 (Phase I I2 — candidate manifest schema — v0.6.0)
+> **Last updated:** 2026-07-09 (Phase I I3 — benchmark card schema — v0.6.1)
+> **New in v0.6.1:** Benchmark card schema (I3 — Loop 101) — `schemas/benchmark_card.schema.json` (Draft 2020-12, 15 required fields: benchmark_id, benchmark_name, version, date, metric, metric_value, baseline_name, baseline_value, delta, beats_baseline, dataset, dataset_size, scope, caveats, dry_lab_only). `$schema`, `$id`, `title`, `additionalProperties: false`. `src/openamp_foundry/benchmarks/` module with `BenchmarkCard` dataclass (15 fields), `make_benchmark_card()` (auto-computes delta, beats_baseline), `validate_benchmark_card()` (10 checks: non-empty benchmark_id, non-empty benchmark_name, non-empty metric, non-empty dataset, non-empty baseline_name, dataset_size >= 1, delta matches metric_value - baseline_value within 1e-9, beats_baseline matches delta > 0, dry_lab_only must be True), `benchmark_card_summary()` (total, beats_baseline_count, fails_baseline_count, dry_lab_only). CLI (`openamp-foundry benchmark-card`) with `--benchmark-id`, `--benchmark-name`, `--metric`, `--metric-value`, `--baseline-name`, `--baseline-value`, `--dataset`, `--dataset-size`, `--validate`, `--format text|json`. `make benchmark-card` target. 19 tests. **3329 total.** A benchmark card is the standard format for describing external benchmark results — what was benchmarked, what the baseline was, what the result was, and what claims are supported.
 > **New in v0.6.0:** Candidate manifest schema (I2 — Loop 100) — `schemas/candidate_manifest.schema.json` (Draft 2020-12, 14 required fields: candidate_id, sequence, evidence_level, scopes, scores, uncertainty, source_modules, calibration_set, safety_flags, provenance_run_id, dry_lab_only, version, created_at, notes). `$schema`, `$id`, `title`, `description`, `additionalProperties: false`. `src/openamp_foundry/manifests/` module with `CandidateManifest` dataclass (14 fields), `make_candidate_manifest()`, `validate_candidate_manifest()` (8 checks: non-empty candidate_id, non-empty sequence, evidence_level 1-6, non-empty scopes, uncertainty 0.0-1.0, non-empty source_modules, dry_lab_only must be True, version MAJOR.MINOR.PATCH), `manifest_to_dict()`, `manifest_summary()` (total, by_evidence_level, with_safety_flags, dry_lab_only). CLI (`openamp-foundry candidate-manifest`) with `--candidate-id`, `--sequence`, `--evidence-level`, `--scopes`, `--scores-json`, `--uncertainty`, `--source-modules`, `--validate`, `--format text|json`. `make candidate-manifest` target. 19 tests. **3310 total.** A candidate manifest is the core interoperable artifact — it describes a dry-lab candidate (sequence, scores, evidence level, scopes, safety flags, provenance) in a machine-readable format that external tools can consume without the full OpenAMP stack.
 > **New in v0.5.99:** Artifact versioning policy (I1 — starts Phase I) — `docs/engineering/ARTIFACT_VERSIONING_POLICY.md` with structured versioning policy covering scope, SemVer format, breaking/non-breaking change definitions, deprecation timeline (at least one minor version warning), schema `$id` policy, changelog requirement, and three stability tiers (stable/experimental/internal). `src/openamp_foundry/versioning/` module with `ArtifactVersionInfo` dataclass (7 fields), `STABILITY_TIERS` dict, `VERSIONED_ARTIFACTS` list (6 entries covering candidate, lab_result, run_manifest, external_review_packet, safety_release_decision, simulation_result at v1.0.0), `get_artifact_version()`, `list_versioned_artifacts()`, `validate_version_format()`, `artifact_version_summary()`. CLI (`openamp-foundry artifact-version`) with `--list`, `--show`, `--tier`, `--format text|json`. `make artifact-version` target. 19 tests. **3291 total.** Starts Phase I (interoperability) — external users now have stability guarantees for schemas, evidence certificates, and candidate manifests.
 > **New in v0.5.97:** Simulation-scope coverage checker (H9) — `ScopeCoverageResult` dataclass (8 fields: module_id, requested_scopes, module_scopes, covered, uncovered, coverage_fraction, is_fully_covered, dry_lab_only). `check_scope_coverage()` looks up module in registry, computes covered (intersection) and uncovered (requested scopes not in module scopes), coverage_fraction = len(covered)/len(requested_scopes) if requested else 1.0. `check_result_scope()` uses conservative intersection of registry scopes and result.scope as effective module_scopes. `scope_coverage_report()` returns full dict. CLI (`openamp-foundry simulation-scope-check`) with `--module-id`, `--requested-scopes`, `--format text|json`. `make simulation-scope-check` target. 17 tests. **3255 total.** A simulation module may cover only some biological scopes. If a candidate is evaluated for a scope the module doesn't cover, that result must be flagged as out-of-scope rather than silently trusted.
@@ -90,7 +91,7 @@ Machine-readable snapshot: `outputs/metrics_snapshot.json` regenerated with `mak
 > **New in v0.5.31:** Added dipeptide-order features for sequence-order awareness. `dipeptide_order_score` achieves AUROC 0.7861 on AMP-vs-scrambled discrimination — the strongest order-dependent feature in the pipeline. Only 7/31 features survive scrambling (amphipathicity/helix-wheel + dipeptide). All composition features are purely position-independent (exactly 0.5000 AUROC on scrambled test).
 > **New in v0.5.30:** Easy baseline benchmark added — charge density alone (AUROC 0.8166) outperforms the full pipeline ensemble (0.7792) on AMP-vs-Swiss-Prot-decoy discrimination. Honest finding documented: expected because pipeline optimizes for safety, not raw discrimination.
 > **New in v0.5.29:** Expanded benchmark to 500 AMPs + 500 composition-matched decoys (n=1000). AUROC 0.7792 (CI₉₅: 0.7505–0.8065) confirms signal generalizes. Cluster-aware CI: 0.746–0.8102. Representative AUROC: 0.778. Standard benchmark (n=191) retained for backward comparison.
-> **Pipeline version:** v0.6.0 (I2 — candidate manifest schema)
+> **Pipeline version:** v0.6.1 (I3 — benchmark card schema)
 > **Branch:** main
 
 ---

diff --git a/docs/research/ROADMAP.md b/docs/research/ROADMAP.md
@@ -1,5 +1,73 @@
 # Roadmap
 
+## v0.6.1 — Loop 101: Phase I I3 — Benchmark Card Schema ✓ (2026-07-09)
+
+`schemas/benchmark_card.schema.json` (Draft 2020-12, 15 required fields:
+benchmark_id, benchmark_name, version, date, metric, metric_value, baseline_name,
+baseline_value, delta, beats_baseline, dataset, dataset_size, scope, caveats,
+dry_lab_only). `$schema`, `$id`, `title`, `additionalProperties: false`.
+
+`src/openamp_foundry/benchmarks/` module with `BenchmarkCard` dataclass
+(15 fields), `make_benchmark_card()` (auto-computes delta, beats_baseline),
+`validate_benchmark_card()` (10 checks: non-empty benchmark_id, non-empty
+benchmark_name, non-empty metric, non-empty dataset, non-empty baseline_name,
+dataset_size >= 1, delta matches metric_value - baseline_value within 1e-9,
+beats_baseline matches delta > 0, dry_lab_only must be True),
+`benchmark_card_summary()` (total, beats_baseline_count, fails_baseline_count,
+dry_lab_only).
+
+CLI (`openamp-foundry benchmark-card`) with `--benchmark-id`, `--benchmark-name`,
+`--metric`, `--metric-value`, `--baseline-name`, `--baseline-value`, `--dataset`,
+`--dataset-size`, `--validate`, `--format text|json`.
+`make benchmark-card` target.
+
+19 tests. **3329 total.** A benchmark card is the standard format for describing
+external benchmark results — what was benchmarked, what the baseline was, what
+the result was, and what claims are supported. This is Loop 101 — the 101st PR
+in the NEXT_100_PR_MAP series.
+
+Changes:
+- `schemas/benchmark_card.schema.json` (I3) — Draft 2020-12 schema with
+  15 required fields, `$id`, `title`, `additionalProperties: false`.
+- `src/openamp_foundry/benchmarks/__init__.py` (I3) — Package init, exports
+  `BenchmarkCard`, `make_benchmark_card`, `validate_benchmark_card`,
+  `benchmark_card_summary`.
+- `src/openamp_foundry/benchmarks/benchmark_card.py` (I3) — Core module
+  with `BenchmarkCard` dataclass (15 fields), `make_benchmark_card()`,
+  `validate_benchmark_card()`, `benchmark_card_summary()`.
+- `src/openamp_foundry/cli/main.py` (I3) — Registered `benchmark-card`
+  subcommand with `--benchmark-id`, `--benchmark-name`, `--metric`,
+  `--metric-value`, `--baseline-name`, `--baseline-value`, `--dataset`,
+  `--dataset-size`, `--validate`, `--format` flags. Added import and dispatch.
+- `src/openamp_foundry/cli/commands/reports.py` (I3) — Added
+  `_run_benchmark_card()` CLI handler with JSON and text output,
+  card creation, optional validation, and error handling.
+- `Makefile` (I3) — Updated `benchmark-card` target with demo invocation
+  using bench-auroc-001. Added to `.PHONY`.
+- `tests/benchmarks/__init__.py` (I3) — Empty package init.
+- `tests/benchmarks/test_benchmark_card.py` (I3) — 19 tests covering:
+  make_benchmark_card returns BenchmarkCard, delta computed correctly,
+  beats_baseline True/False, dry_lab_only always True, valid card passes
+  validation, catches empty benchmark_id, empty benchmark_name, empty metric,
+  empty dataset, empty baseline_name, dataset_size < 1, wrong delta,
+  wrong beats_baseline, dry_lab_only=False, summary total, summary
+  beats_baseline_count, dry_lab_only, schema file exists.
+- `docs/evidence/METRICS_CURRENT.md` — v0.6.1 I3 changelog. Test count: 3329.
+- `tests/test_test_count_regression.py` — baseline updated to 3329.
+
+Honest boundaries:
+- Benchmark cards describe computational benchmark results only — they do not
+  measure biological activity, safety, or clinical value.
+- Delta and beats_baseline are computed from metric_value and baseline_value,
+  not independently measured.
+- The schema is versioned (currently 1.0.0) and should be updated when fields
+  are added or changed.
+- Validation checks structural correctness and internal consistency only — it
+  does not verify that the benchmark was correctly designed or that the results
+  generalize.
+- `dry_lab_only: true` is a const field — all benchmark cards are inherently
+  dry-lab and must never be presented as validated biological findings.
+
 ## v0.6.0 — Loop 100: Phase I I2 — Candidate Manifest Schema ✓ (2026-07-09)
 
 `schemas/candidate_manifest.schema.json` (Draft 2020-12, 14 required fields:

diff --git a/schemas/benchmark_card.schema.json b/schemas/benchmark_card.schema.json
@@ -0,0 +1,99 @@
+{
+    "$schema": "https://json-schema.org/draft/2020-12/schema",
+    "$id": "https://openamp-foundry.org/schemas/benchmark_card.schema.json",
+    "title": "BenchmarkCard",
+    "description": "Schema for a benchmark card — a machine-readable artifact describing what was benchmarked, what the baseline was, what the result was, and what claims are supported. Dry-lab only — does not measure biological activity.",
+    "type": "object",
+    "properties": {
+        "benchmark_id": {
+            "description": "Unique identifier for the benchmark",
+            "type": "string",
+            "minLength": 1
+        },
+        "benchmark_name": {
+            "description": "Human-readable benchmark name",
+            "type": "string",
+            "minLength": 1
+        },
+        "version": {
+            "description": "Schema version (MAJOR.MINOR.PATCH)",
+            "type": "string",
+            "pattern": "^\\d+\\.\\d+\\.\\d+$"
+        },
+        "date": {
+            "description": "Date the benchmark was run (YYYY-MM-DD)",
+            "type": "string",
+            "pattern": "^\\d{4}-\\d{2}-\\d{2}$"
+        },
+        "metric": {
+            "description": "Metric name (e.g. AUROC, AUPRC, recall@k)",
+            "type": "string",
+            "minLength": 1
+        },
+        "metric_value": {
+            "description": "Observed metric value",
+            "type": "number"
+        },
+        "baseline_name": {
+            "description": "Name of the comparison baseline",
+            "type": "string",
+            "minLength": 1
+        },
+        "baseline_value": {
+            "description": "Baseline metric value",
+            "type": "number"
+        },
+        "delta": {
+            "description": "metric_value - baseline_value",
+            "type": "number"
+        },
+        "beats_baseline": {
+            "description": "True if metric_value > baseline_value",
+            "type": "boolean"
+        },
+        "dataset": {
+            "description": "Name or identifier of the dataset used",
+            "type": "string",
+            "minLength": 1
+        },
+        "dataset_size": {
+            "description": "Number of samples in the dataset",
+            "type": "integer",
+            "minimum": 1
+        },
+        "scope": {
+            "description": "List of biological scopes this benchmark applies to",
+            "type": "array",
+            "items": { "type": "string" },
+            "minItems": 1
+        },
+        "caveats": {
+            "description": "List of caveats and limitations",
+            "type": "array",
+            "items": { "type": "string" }
+        },
+        "dry_lab_only": {
+            "description": "Must be true — benchmark cards are computational artifacts, not biological proof",
+            "type": "boolean",
+            "const": true
+        }
+    },
+    "required": [
+        "benchmark_id",
+        "benchmark_name",
+        "version",
+        "date",
+        "metric",
+        "metric_value",
+        "baseline_name",
+        "baseline_value",
+        "delta",
+        "beats_baseline",
+        "dataset",
+        "dataset_size",
+        "scope",
+        "caveats",
+        "dry_lab_only"
+    ],
+    "additionalProperties": false
+}
diff --git a/src/openamp_foundry/benchmarks/__init__.py b/src/openamp_foundry/benchmarks/__init__.py
@@ -0,0 +1,13 @@
+from openamp_foundry.benchmarks.benchmark_card import (
+    BenchmarkCard,
+    make_benchmark_card,
+    validate_benchmark_card,
+    benchmark_card_summary,
+)
+
+__all__ = [
+    "BenchmarkCard",
+    "make_benchmark_card",
+    "validate_benchmark_card",
+    "benchmark_card_summary",
+]
diff --git a/src/openamp_foundry/benchmarks/benchmark_card.py b/src/openamp_foundry/benchmarks/benchmark_card.py
@@ -0,0 +1,101 @@
+from __future__ import annotations
+
+import math
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+
+
+@dataclass
+class BenchmarkCard:
+    benchmark_id: str
+    benchmark_name: str
+    version: str
+    date: str
+    metric: str
+    metric_value: float
+    baseline_name: str
+    baseline_value: float
+    delta: float
+    beats_baseline: bool
+    dataset: str
+    dataset_size: int
+    scope: list[str]
+    caveats: list[str]
+    dry_lab_only: bool = True
+
+
+def make_benchmark_card(
+    benchmark_id: str,
+    benchmark_name: str,
+    metric: str,
+    metric_value: float,
+    baseline_name: str,
+    baseline_value: float,
+    dataset: str,
+    dataset_size: int,
+    *,
+    version: str = "1.0.0",
+    date: str = "",
+    scope: list[str] | None = None,
+    caveats: list[str] | None = None,
+) -> BenchmarkCard:
+    delta = metric_value - baseline_value
+    beats_baseline = delta > 0
+    return BenchmarkCard(
+        benchmark_id=benchmark_id,
+        benchmark_name=benchmark_name,
+        version=version,
+        date=date or datetime.now(timezone.utc).strftime("%Y-%m-%d"),
+        metric=metric,
+        metric_value=metric_value,
+        baseline_name=baseline_name,
+        baseline_value=baseline_value,
+        delta=delta,
+        beats_baseline=beats_baseline,
+        dataset=dataset,
+        dataset_size=dataset_size,
+        scope=scope or [],
+        caveats=caveats or [],
+    )
+
+
+def validate_benchmark_card(card: BenchmarkCard) -> list[str]:
+    errors: list[str] = []
+    if not card.benchmark_id:
+        errors.append("benchmark_id must be non-empty")
+    if not card.benchmark_name:
+        errors.append("benchmark_name must be non-empty")
+    if not card.metric:
+        errors.append("metric must be non-empty")
+    if not card.dataset:
+        errors.append("dataset must be non-empty")
+    if not card.baseline_name:
+        errors.append("baseline_name must be non-empty")
+    if card.dataset_size < 1:
+        errors.append(f"dataset_size must be >= 1, got {card.dataset_size}")
+    expected_delta = card.metric_value - card.baseline_value
+    if abs(card.delta - expected_delta) > 1e-9:
+        errors.append(
+            f"delta {card.delta} does not match metric_value - baseline_value "
+            f"({expected_delta})"
+        )
+    expected_beats = expected_delta > 0
+    if card.beats_baseline != expected_beats:
+        errors.append(
+            f"beats_baseline={card.beats_baseline} does not match delta > 0 "
+            f"({expected_beats})"
+        )
+    if not card.dry_lab_only:
+        errors.append("dry_lab_only must be True")
+    return errors
+
+
+def benchmark_card_summary(cards: list[BenchmarkCard]) -> dict:
+    beats_count = sum(1 for c in cards if c.beats_baseline)
+    fails_count = sum(1 for c in cards if not c.beats_baseline)
+    return {
+        "total": len(cards),
+        "beats_baseline_count": beats_count,
+        "fails_baseline_count": fails_count,
+        "dry_lab_only": True,
+    }