diff --git a/docs/evidence/DIAGRAM_ARTIFACT_DEPENDENCY.md b/docs/evidence/DIAGRAM_ARTIFACT_DEPENDENCY.md
new file mode 100644
index 00000000..90c39162
--- /dev/null
+++ b/docs/evidence/DIAGRAM_ARTIFACT_DEPENDENCY.md
@@ -0,0 +1,23 @@
+# Artifact Dependency Graph
+
+Shows how artifacts depend on each other.
+
+```mermaid
+flowchart TD
+  A[Candidate CSV] --> B[Ranked JSONL]
+  A --> C[Evidence certs]
+  B --> D[Batch report]
+  B --> E[Run manifest]
+  C --> F[Lab batch pack]
+  F --> G[Lab results]
+  G --> H[Calibration intake]
+  H --> I[Gate verdict]
+  I -->|pass| J[Weight proposal]
+  I -->|fail| K[Blocked]
+```
+
+## Rules
+- Artifacts higher in the graph are inputs.
+- Arrows show the direction of data flow.
+- Dotted lines indicate optional dependencies.
+- The graph should be updated when new artifact types are added.
diff --git a/docs/evidence/GUIDE_ISSUE_ARTIFACT_TRACE.md b/docs/evidence/GUIDE_ISSUE_ARTIFACT_TRACE.md
new file mode 100644
index 00000000..88c84b2d
--- /dev/null
+++ b/docs/evidence/GUIDE_ISSUE_ARTIFACT_TRACE.md
@@ -0,0 +1,19 @@
+# Issue-to-Artifact Trace Field
+
+How issues trace to their resulting artifacts.
+
+## Format
+Issues should include a `## Artifacts` section when resolved:
+```markdown
+## Artifacts
+- PR: #NNN
+- Doc: docs/evidence/EXAMPLE.md
+- Script: scripts/example.py
+- Test: tests/test_example.py
+```
+
+## Rules
+- Every closed issue should trace to at least one artifact.
+- Artifacts can be PRs, docs, scripts, tests, or configs.
+- The trace helps future contributors understand why something exists.
+- If an issue doesn't produce artifacts (investigation only), note that.
diff --git a/docs/evidence/GUIDE_LINEAGE_RENDERER.md b/docs/evidence/GUIDE_LINEAGE_RENDERER.md
new file mode 100644
index 00000000..12e8b0f3
--- /dev/null
+++ b/docs/evidence/GUIDE_LINEAGE_RENDERER.md
@@ -0,0 +1,22 @@
+# Artifact Lineage Markdown Renderer
+
+How artifact lineage is documented in markdown.
+
+## Format
+```markdown
+## Artifact Lineage
+
+```mermaid
+flowchart LR
+  A[Input CSV] --> B[score_candidates]
+  B --> C[Ranked JSONL]
+  B --> D[Evidence certs]
+```
+```
+
+## Rules
+- Lineage diagrams should use Mermaid flowchart format.
+- Show inputs → process → outputs.
+- Label each node with the artifact name and format.
+- Keep lineage diagrams focused on one artifact type at a time.
+- Include the lineage in the artifact's documentation.
diff --git a/docs/evidence/GUIDE_STALE_ARTIFACT_DETECTOR.md b/docs/evidence/GUIDE_STALE_ARTIFACT_DETECTOR.md
new file mode 100644
index 00000000..7d9f025a
--- /dev/null
+++ b/docs/evidence/GUIDE_STALE_ARTIFACT_DETECTOR.md
@@ -0,0 +1,21 @@
+# Stale Artifact Detector
+
+Detects artifacts that may be stale (out of sync with current code).
+
+## What It Checks
+1. Schema files — compare schema version against code.
+2. Benchmark outputs — check if they match current pipeline.
+3. Generated docs — check if they reference current features.
+4. Config files — check if all keys are still valid.
+
+## Usage
+```bash
+python scripts/check_doc_links.py  # Check docs for broken references
+make bench-gate                     # Check benchmark regression gates
+python scripts/check_pyproject.py   # Check pyproject.toml consistency
+```
+
+## Manual Checks
+- Evidence certificates should be regenerated after scoring changes.
+- Benchmarks should be re-run after config changes.
+- Docs should be reviewed after major feature changes.
diff --git a/docs/evidence/MATRIX_CLAIM_EVIDENCE.md b/docs/evidence/MATRIX_CLAIM_EVIDENCE.md
new file mode 100644
index 00000000..1e32e6f1
--- /dev/null
+++ b/docs/evidence/MATRIX_CLAIM_EVIDENCE.md
@@ -0,0 +1,16 @@
+# Claim-to-Evidence Matrix Artifact
+
+Maps claims to their supporting evidence artifacts.
+
+| Claim | Evidence Level | Supporting Artifact | Status |
+|-------|:--------------:|--------------------|:------:|
+| Pipeline ranks AMPs | 2 (baseline) | METRICS_CURRENT.md | ✅ |
+| Pipeline is reproducible | 2 (baseline) | RUN_MANIFEST_STANDARD.md | ✅ |
+| Candidates are novel | 2 (baseline) | NOVELTY_CHECKLIST.md | ✅ |
+| Candidates are safe | 2 (baseline) | LIMITATIONS_OVERVIEW.md | ⚠️ Limited |
+
+## Rules
+- Every claim should be paired with its evidence.
+- The evidence level should match the claim strength.
+- If evidence is missing, mark the claim as unsupported.
+- This matrix should be reviewed before any public release.
diff --git a/docs/evidence/POLICY_ARTIFACT_RETENTION.md b/docs/evidence/POLICY_ARTIFACT_RETENTION.md
new file mode 100644
index 00000000..00b98e9a
--- /dev/null
+++ b/docs/evidence/POLICY_ARTIFACT_RETENTION.md
@@ -0,0 +1,18 @@
+# Artifact Retention Policy
+
+How long different artifact types are retained.
+
+| Artifact Type | Retention | Location | Cleanup |
+|:-------------|:---------:|:---------|:--------|
+| Evidence certificates | Permanent | outputs/evidence/ | Manual |
+| Run manifests | 2 years | outputs/ | Automated |
+| Benchmark results | Permanent | outputs/ | Manual |
+| Lab results | Permanent | outputs/lab_results/ | Manual |
+| Temporary outputs | 90 days | outputs/ | Automated |
+| Logs | 30 days | outputs/logs/ | Automated |
+
+## Rules
+- Permanent artifacts must never be deleted without a decision log entry.
+- Temporary outputs can be cleaned up with `make clean`.
+- Lab results must be preserved for the duration of the study.
+- Retention periods start from the date of creation.
diff --git a/docs/evidence/POLICY_FLAKY_TEST.md b/docs/evidence/POLICY_FLAKY_TEST.md
new file mode 100644
index 00000000..6dbc9dc6
--- /dev/null
+++ b/docs/evidence/POLICY_FLAKY_TEST.md
@@ -0,0 +1,19 @@
+# Flaky Test Quarantine Policy
+
+How flaky tests are handled.
+
+## Definition
+A flaky test is one that passes and fails without code changes.
+
+## Process
+1. Identify the flaky test and document the failure pattern in an issue.
+2. If the test is low-value, remove it.
+3. If the test is high-value, quarantine it by marking with `pytest.mark.flaky`.
+4. Investigate the root cause within one sprint.
+5. Fix or permanently remove the test.
+
+## Quarantine Rules
+- Flaky tests must not block CI.
+- Quarantined tests should be tracked in a GitHub issue.
+- Quarantined tests should be reviewed weekly.
+- If a test remains flaky for more than 2 sprints, it should be removed.
diff --git a/docs/evidence/SCHEMA_LIMITATIONS_SECTION.md b/docs/evidence/SCHEMA_LIMITATIONS_SECTION.md
new file mode 100644
index 00000000..16c03687
--- /dev/null
+++ b/docs/evidence/SCHEMA_LIMITATIONS_SECTION.md
@@ -0,0 +1,26 @@
+# Limitations Section Schema
+
+Standard structure for limitations sections in documentation.
+
+## Required Sections
+- What the limitation is
+- Why it exists
+- What it does NOT affect
+- Planned mitigation or status
+
+## Example
+```markdown
+### Charge-Dominated Scoring
+The pipeline's primary discriminative signal is charge density,
+not sophisticated multi-feature scoring. The ensemble AUROC of
+0.7792 is charge-inflated (collapses to 0.5103 under exact charge
+control). This does NOT affect the safety or novelty scoring.
+**Mitigation:** `charge_bias` score available per candidate;
+charge-matched benchmarks track this gap.
+```
+
+## Rules
+- Every limitations section should follow this schema.
+- Limitations should be specific, not general.
+- Include both impact and mitigation.
+- Link to related docs where available.
diff --git a/tests/test_artifact_bundle_validator.py b/tests/test_artifact_bundle_validator.py
new file mode 100644
index 00000000..ca8e5a59
--- /dev/null
+++ b/tests/test_artifact_bundle_validator.py
@@ -0,0 +1,20 @@
+"""Test artifact bundle validation."""
+from pathlib import Path
+
+
+def test_batch_pack_script_exists():
+    assert Path("scripts/build_lab_batch_pack.py").exists()
+
+
+def test_evidence_directory_exists():
+    ev = Path("outputs/evidence")
+    if ev.exists():
+        certs = list(ev.glob("*.json"))
+        assert len(certs) > 0, "Evidence directory exists but is empty"
+
+
+def test_phase3_evidence_exists():
+    ev = Path("outputs/evidence_wave0_5")
+    if ev.exists():
+        certs = list(ev.glob("*.json"))
+        assert len(certs) > 0