diff --git a/docs/REVIEW-STATUS-GUIDE.md b/docs/REVIEW-STATUS-GUIDE.md
new file mode 100644
index 0000000..c222374
--- /dev/null
+++ b/docs/REVIEW-STATUS-GUIDE.md
@@ -0,0 +1,90 @@
+# Review Status Guide
+
+Open Problem Lab needs a clear vocabulary for claims after review. Merge is not the final state of knowledge. A claim can become narrower, disputed, replaced, or ready for field review as new evidence arrives.
+
+This guide gives reviewers and agents a small set of status words to keep confidence visible.
+
+## Status words
+
+| Status | Meaning |
+|---|---|
+| `proposed` | Structured, but not accepted as repository knowledge. |
+| `accepted` | Passed review for the scope stated in the artifact. |
+| `bounded` | Useful only under a narrower scope than first proposed. |
+| `disputed` | Serious unresolved counter-evidence or reviewer concern exists. |
+| `superseded` | Replaced by a newer or stronger artifact. |
+| `archived` | Preserved for history, but should not support new claims. |
+| `field-pending` | Desk evidence exists, but operational validity is not established. |
+| `replication-pending` | Initial evidence exists, but independent replication is still required. |
+
+The labels are not prestige labels. They are routing labels. They tell the next contributor how much weight a claim can carry.
+
+## When status should change
+
+A status should change when a serious reviewer would use the claim differently.
+
+Examples:
+
+- `accepted` to `bounded`: the result survives only for a narrower population, geography, dataset, season, or method.
+- `accepted` to `disputed`: a credible replication problem or field contradiction appears.
+- `disputed` to `bounded`: the objection is real but limited to a known scope.
+- `accepted` to `superseded`: a newer artifact is clearly stronger and should be the citation target.
+- `field-pending` to `accepted`: operational review supports the claim inside its stated use case.
+
+Do not change status for rhetorical disagreement. The trigger should be evidence, method, scope, or review substance.
+
+## Minimum status-change note
+
+Use this structure in the relevant PR, issue, or problem-pack note:
+
+```markdown
+## Review status change
+
+**Artifact:** Link to claim, evidence record, or section.
+
+**Previous status:** proposed | accepted | bounded | disputed | superseded | archived | field-pending | replication-pending
+
+**New status:** proposed | accepted | bounded | disputed | superseded | archived | field-pending | replication-pending
+
+**Reason:** One paragraph.
+
+**Evidence or review event:** Link to PR, issue, source update, replication artifact, or reviewer comment.
+
+**Practical effect:** What should future contributors do differently?
+
+**Reversal condition:** What would justify changing the status again?
+```
+
+## Reviewer rule
+
+A reviewer can approve a PR while requiring a weaker status.
+
+That distinction matters. A contribution may be structured, sourced, and useful while still not strong enough to be treated as accepted operational knowledge.
+
+Reviewers should prefer explicit status over vague concern when a claim is useful but over-scoped.
+
+## Agent rule
+
+Before using a claim as support, an agent should inspect its status.
+
+- `accepted`: may be used inside the stated scope.
+- `bounded`: must carry the boundary forward.
+- `disputed`: may be discussed, but should not be used as a load-bearing premise.
+- `superseded`: cite the newer artifact instead.
+- `archived`: do not use as support except in historical context.
+- `field-pending`: do not treat as deployment-ready.
+- `replication-pending`: do not treat as independently established.
+
+If an agent cannot determine status, it should say so rather than silently upgrading confidence.
+
+## Acceptance test
+
+A status guide is working when a new contributor can answer five questions quickly:
+
+1. Can I rely on this claim?
+2. Inside what boundary?
+3. What is the strongest known objection?
+4. What would change the repository's mind?
+5. Which artifact should I cite now?
+
+If the repository cannot answer those questions, it is accumulating content faster than trust.