diff --git a/docs/REVIEW-ROUTING-MATRIX.md b/docs/REVIEW-ROUTING-MATRIX.md
new file mode 100644
index 0000000..4cb9ebd
--- /dev/null
+++ b/docs/REVIEW-ROUTING-MATRIX.md
@@ -0,0 +1,78 @@
+# Review Routing Matrix
+
+Open Problem Lab should route review by claim risk, not by who happens to see the pull request first.
+
+This document defines a lightweight routing matrix for deciding when a submission needs general review, domain review, field review, replication review, or maintainer review.
+
+## Why this matters
+
+A protocol can have strong standards on paper and still fail if the wrong reviewer approves the wrong claim.
+
+The question is not only whether a PR has been reviewed. The question is whether the review matched the claim.
+
+## Routing levels
+
+| Route | Use when | Minimum reviewer |
+|---|---|---|
+| `general` | Docs structure, typo fixes, navigation, low-risk explanation. | Any maintainer or trusted contributor. |
+| `evidence` | Adds or changes evidence records, source interpretation, claim boundaries. | Reviewer comfortable with evidence quality and citation standards. |
+| `domain` | Makes a substantive claim inside a technical or field domain. | Domain-knowledgeable reviewer. |
+| `quantitative` | Adds metrics, rankings, model results, estimates, or statistical claims. | Reviewer who can inspect method and uncertainty. |
+| `field` | Implies real-world action, program design, implementation, or policy use. | Field-reality reviewer or practitioner-informed reviewer. |
+| `replication` | Claim is safety-sensitive, quantitative, or load-bearing. | Independent reviewer or reproduction artifact. |
+| `maintainer` | Changes schemas, workflows, governance, generated indexes, or acceptance gates. | Maintainer. |
+
+## Minimum routing note
+
+For non-trivial PRs, add this note to the PR body or review comment:
+
+```markdown
+## Review routing
+
+**Claim type:** docs | evidence | domain | quantitative | field | governance | generated artifact
+
+**Suggested route:** general | evidence | domain | quantitative | field | replication | maintainer
+
+**Why this route:** One sentence.
+
+**Reviewer gap:** What expertise is missing if this is merged now?
+```
+
+## Escalation triggers
+
+Escalate review when a PR:
+
+- changes a claim's status;
+- changes a metric, threshold, or ranking;
+- introduces a new source that carries a load-bearing claim;
+- touches generated canonical indexes;
+- changes schemas or validation rules;
+- turns desk evidence into operational advice;
+- makes a claim where a false positive would waste money, time, trust, or field capacity.
+
+## Reviewer guidance
+
+A reviewer should be able to say either:
+
+> I am the right reviewer for this route.
+
+or:
+
+> I can review structure, but this still needs domain, field, quantitative, replication, or maintainer review.
+
+Partial review is useful when it is labeled. Unlabeled partial review is dangerous.
+
+## Agent guidance
+
+Agents should propose the review route before asking for review.
+
+An agent should not mark work as ready simply because it passes validation. Validation checks structure. Review routing checks whether the right human or artifact has attacked the right risk.
+
+## Acceptance test
+
+Review routing is working when a contributor can look at a PR and know:
+
+1. what kind of claim it changes;
+2. which review route is required;
+3. which reviewer gap remains;
+4. whether merge would create an unsupported confidence upgrade.