diff --git a/.github/CODEOWNERS b/.github/CODEOWNERS
index d4ecfa5..e937724 100644
--- a/.github/CODEOWNERS
+++ b/.github/CODEOWNERS
@@ -8,9 +8,9 @@
 # CI fails if this file drifts from its source, and rejects PRs that
 # edit this file directly without also editing the yml.
 
-*             @ragnorc
+*             @ragnorc @aaltshuler
 
-crates/**     @ragnorc
+crates/**     @ragnorc @aaltshuler
 docs/**       @ragnorc
 README.md     @ragnorc
 AGENTS.md     @ragnorc
diff --git a/.github/DISCUSSION_TEMPLATE/rfc.yml b/.github/DISCUSSION_TEMPLATE/rfc.yml
new file mode 100644
index 0000000..2a63525
--- /dev/null
+++ b/.github/DISCUSSION_TEMPLATE/rfc.yml
@@ -0,0 +1,34 @@
+labels: ["rfc"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Use this to **incubate an RFC** — socialize a design and reach rough
+        consensus before writing the formal document. When it's ready, graduate
+        it into a pull request that adds `docs/rfcs/NNNN-title.md`
+        (see [docs/rfcs/README.md](../blob/main/docs/rfcs/README.md)); a
+        maintainer merging that PR is acceptance.
+
+        For a plain feature request or open-ended idea, use the **Ideas**
+        category instead. For bugs, open an [Issue](../../issues/new/choose).
+  - type: textarea
+    id: problem
+    attributes:
+      label: Problem / motivation
+      description: What needs solving, and why is it worth the long-run cost?
+    validations:
+      required: true
+  - type: textarea
+    id: sketch
+    attributes:
+      label: Proposed direction (sketch)
+      description: A rough shape of the design. Detail comes later in the RFC document.
+    validations:
+      required: true
+  - type: textarea
+    id: invariants
+    attributes:
+      label: Invariants touched
+      description: Which items in docs/dev/invariants.md does this affect or risk? Any deny-list brush?
+    validations:
+      required: false
diff --git a/.github/ISSUE_TEMPLATE/bug_report.yml b/.github/ISSUE_TEMPLATE/bug_report.yml
new file mode 100644
index 0000000..8e19465
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/bug_report.yml
@@ -0,0 +1,55 @@
+name: Bug report
+description: Report a reproducible problem or wrong behavior in OmniGraph.
+title: "bug: <short summary>"
+labels: ["bug", "needs-triage"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Issues are for **reporting problems** — concrete, reproducible bugs.
+        For ideas, feature requests, or questions, please use
+        [Discussions](../../discussions) instead.
+        For a security vulnerability, follow [SECURITY.md](../../blob/main/SECURITY.md) — do **not** file it here.
+
+        A maintainer will triage this; once labelled **`accepted`** it's open for a pull request
+        (see [GOVERNANCE.md](../../blob/main/GOVERNANCE.md)).
+  - type: textarea
+    id: what-happened
+    attributes:
+      label: What happened
+      description: What went wrong, and what you expected instead.
+    validations:
+      required: true
+  - type: textarea
+    id: repro
+    attributes:
+      label: Steps to reproduce
+      description: Minimal steps, commands, schema/query, or a failing snippet.
+      placeholder: |
+        1. omnigraph init ...
+        2. omnigraph ...
+        3. observed: ...  / expected: ...
+    validations:
+      required: true
+  - type: input
+    id: version
+    attributes:
+      label: Version
+      description: Output of `omnigraph --version` (or the engine/crate version) and how you installed it.
+    validations:
+      required: true
+  - type: input
+    id: environment
+    attributes:
+      label: Environment
+      description: OS, architecture, and storage backend (local FS / S3 / RustFS / MinIO).
+    validations:
+      required: false
+  - type: textarea
+    id: logs
+    attributes:
+      label: Logs / output
+      description: Relevant error text or logs. Will be rendered as code.
+      render: shell
+    validations:
+      required: false
diff --git a/.github/ISSUE_TEMPLATE/config.yml b/.github/ISSUE_TEMPLATE/config.yml
new file mode 100644
index 0000000..50720b8
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/config.yml
@@ -0,0 +1,13 @@
+# Issues are for problem reports only. Disable blank issues so everything is
+# routed: bugs through the form, everything else to Discussions / SECURITY.md.
+blank_issues_enabled: false
+contact_links:
+  - name: 💡 Idea, feature request, or RFC
+    url: https://github.com/ModernRelay/omnigraph/discussions
+    about: Propose features and designs in Discussions. RFCs graduate from there into a docs/rfcs/ pull request.
+  - name: ❓ Question or help
+    url: https://github.com/ModernRelay/omnigraph/discussions
+    about: Ask in Discussions — questions are not tracked as Issues.
+  - name: 🔒 Security vulnerability
+    url: https://github.com/ModernRelay/omnigraph/blob/main/SECURITY.md
+    about: Report security issues privately per SECURITY.md — never as a public Issue.
diff --git a/.github/PULL_REQUEST_TEMPLATE.md b/.github/PULL_REQUEST_TEMPLATE.md
new file mode 100644
index 0000000..2a548c7
--- /dev/null
+++ b/.github/PULL_REQUEST_TEMPLATE.md
@@ -0,0 +1,29 @@
+<!--
+  Thanks for contributing! See CONTRIBUTING.md and GOVERNANCE.md.
+  A substantive PR needs a backing accepted issue or accepted RFC.
+  Maintainers: your internal process applies; the link requirement below
+  is for external contributions.
+-->
+
+## What & why
+
+<!-- One or two sentences: what this changes and why. -->
+
+## Backing issue / RFC
+
+<!-- Pick one. A substantive change needs (1) or (2). -->
+
+- [ ] Fixes an **accepted** issue: Closes #
+- [ ] Implements / is an **accepted** RFC: <link to docs/rfcs/NNNN-*.md>
+- [ ] **Trivial fast-lane** (typo / docs / dependency bump / comment / one-line CI) — no issue/RFC required
+
+## Checklist
+
+- [ ] Change is focused (one logical change)
+- [ ] Tests added/updated for behavior changes (or N/A)
+- [ ] Public docs updated if user-facing surface changed (or N/A)
+- [ ] Reviewed against [docs/dev/invariants.md](../blob/main/docs/dev/invariants.md) — no Hard Invariant weakened, no deny-list item hit (or justified)
+
+## Notes for reviewers
+
+<!-- Anything that helps review: tradeoffs, follow-ups, areas of risk. -->
diff --git a/.github/codeowners-roles.yml b/.github/codeowners-roles.yml
index c5e36a9..ce4014d 100644
--- a/.github/codeowners-roles.yml
+++ b/.github/codeowners-roles.yml
@@ -22,6 +22,7 @@ roles:
       compiler.
     members:
       - ragnorc
+      - aaltshuler
 
   docs:
     description: >
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 8d9c687..2d77ef0 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -1,10 +1,29 @@
 # Contributing
 
-Small bug fixes and documentation improvements are welcome directly through pull
-requests.
+Thanks for your interest in OmniGraph. This page is the practical how-to; the
+rules and decision authority behind it live in [GOVERNANCE.md](GOVERNANCE.md).
 
-For larger changes, please open an issue or design discussion first so the
-proposed direction is clear before implementation starts.
+## Start in the right place
+
+| I want to… | Go to | Notes |
+|---|---|---|
+| **Report a bug** or wrong behavior | **[Open an Issue](../../issues/new/choose)** | Concrete and reproducible. A maintainer triages it; once labelled **`accepted`** it's open for a PR. |
+| **Suggest a feature / share an idea / ask** | **[Start a Discussion](../../discussions)** | Ideas and questions live here, not in Issues. |
+| **Propose a design / RFC** | **An RFC pull request** | Anyone can author one — see [docs/rfcs/README.md](docs/rfcs/README.md). A maintainer merging it is acceptance. |
+| **Fix something / implement a change** | **A pull request** | Must link an `accepted` issue or an accepted RFC — unless it's trivial (below). |
+| **Report a security vulnerability** | **[SECURITY.md](SECURITY.md)** | Do **not** open a public Issue. |
+
+### When can I just open a PR?
+The **trivial fast-lane** — open directly, no prior issue/RFC needed: typo and
+wording fixes, doc corrections, dependency bumps, comment fixes, obvious
+one-line CI tweaks. Anything more substantial needs a backing `accepted` issue
+or accepted RFC first, so the *why* is agreed before the *how* is reviewed. A PR
+that turns out to be non-trivial will be redirected — that's about process, not
+the merit of the change.
+
+> **Maintainers (ModernRelay team)** follow a separate internal process and are
+> not bound by the intake rules above. Everyone is bound by review, CODEOWNERS,
+> branch protection, and CI.
 
 ## Development
 
@@ -49,6 +68,11 @@ CI runs both.
 
 ## Pull Requests
 
-- keep changes focused
-- include tests for behavior changes when practical
-- update public docs when the user-facing surface changes
+- **Link the backing issue or RFC** (`Closes #123`, or reference the RFC) — or
+  mark the PR as trivial per the fast-lane.
+- Keep changes focused; one logical change per PR.
+- Include tests for behavior changes when practical.
+- Update public docs when the user-facing surface changes.
+
+New to the codebase? Read [AGENTS.md](AGENTS.md) — the architecture map and the
+always-on invariants every change is reviewed against.
diff --git a/GOVERNANCE.md b/GOVERNANCE.md
new file mode 100644
index 0000000..5878f1f
--- /dev/null
+++ b/GOVERNANCE.md
@@ -0,0 +1,106 @@
+# Governance
+
+This document describes how **external contributions** to OmniGraph are
+proposed, accepted, and merged. It exists so an outside contributor can answer,
+without asking: *where does my report/idea/change go, who decides, and what has
+to happen before code lands?*
+
+> **Scope.** This governs the public contribution surface — Issues,
+> Discussions, RFCs, and pull requests from people outside the ModernRelay
+> team. **Maintainers operate under a separate internal process** and are not
+> bound by the intake gates below. Everyone, maintainer or not, is still bound
+> by the universal gates: branch protection on `main` and CODEOWNERS review
+> (see [docs/dev/branch-protection.md](docs/dev/branch-protection.md) and
+> [docs/dev/codeowners.md](docs/dev/codeowners.md)).
+
+## Roles
+
+| Role | Who | Authority |
+|---|---|---|
+| **Maintainer** | The code owners in [`.github/CODEOWNERS`](.github/CODEOWNERS) (generated from [`.github/codeowners-roles.yml`](.github/codeowners-roles.yml)) | Validate issues, accept/reject RFCs, review and merge PRs, set direction. Final decision authority. |
+| **Contributor** | Anyone else | Report problems (Issues), propose ideas (Discussions), author RFCs, and open pull requests. |
+
+Decision authority rests with the maintainers. CODEOWNERS is the single source
+of truth for who that is; this document does not duplicate the list.
+
+## The three channels
+
+Each channel has one job. Using the right one is the first thing we ask of a
+contribution.
+
+| Channel | Purpose | Not for |
+|---|---|---|
+| **[Issues](../../issues)** | **Report a problem** — a bug, a regression, a documented behavior that's wrong. Something concrete and reproducible. | Feature requests, ideas, questions, or design proposals (→ Discussions). |
+| **[Discussions](../../discussions)** | **Propose and explore** — new ideas, feature requests, questions, and the incubation of RFCs. | Bug reports (→ Issues). |
+| **Pull requests** | **Land a sanctioned change** — a fix for a *validated* issue, an *accepted* RFC, or a trivial change (see fast-lane). | Substantive change with no backing issue/RFC — it will be redirected. |
+
+## How a change becomes mergeable
+
+```
+            ┌─────────── bug ───────────┐        ┌──────── idea / feature ────────┐
+            ▼                            │        ▼                                │
+        Issue (problem report)           │   Discussion (idea / RFC incubation)    │
+            │                            │        │                                │
+   maintainer triage                     │   rough consensus                        │
+            │                            │        │ graduate                         │
+            ▼                            │        ▼                                  │
+  label: accepted  ──────────┐          │   RFC PR  (docs/rfcs/NNNN-*.md)           │
+            │                 │          │        │                                  │
+            │                 │          │   maintainer review                       │
+            ▼                 ▼          │        ▼                                  │
+     Pull request  ◀──────────┴──────────│──  merged == accepted                     │
+   (links the issue or the accepted RFC) ◀───────┘ (implementation PRs reference it) │
+            │
+   review + CODEOWNERS + branch protection
+            ▼
+         merged
+```
+
+### Issues → validated
+A new issue starts unlabeled. A maintainer triages it and, if it's a real,
+in-scope problem, applies the **`accepted`** label. **Only `accepted` issues are
+open for a contributor PR.** This prevents the "I fixed an issue you hadn't
+agreed was a problem" rejection. Want to fix something? Get the issue accepted
+first, or pick one already labelled `accepted` / `help wanted`.
+
+### Discussions → RFCs → accepted
+Ideas and feature requests start in **Discussions**. Anyone — including external
+contributors — may then **author an RFC** by opening a pull request that adds
+`docs/rfcs/NNNN-title.md` (see [docs/rfcs/README.md](docs/rfcs/README.md)). The
+RFC is reviewed as code; **a maintainer merging it is the act of acceptance**
+(it becomes the durable decision record). Implementation PRs then reference the
+accepted RFC.
+
+Authoring an RFC is open to everyone; **accepting one is a maintainer
+decision.** Maintainers may also decline an RFC, with rationale, by closing it.
+
+### Pull requests → sanctioned
+A contributor PR must do one of:
+1. link a maintainer-**`accepted`** issue it fixes, or
+2. be (or reference) an **accepted RFC**, or
+3. qualify for the **trivial fast-lane**.
+
+**Trivial fast-lane** — these may be opened directly, no prior issue/RFC:
+typo and wording fixes, documentation corrections, dependency bumps, comment
+fixes, and obviously-correct one-line CI tweaks. When in doubt, open an Issue or
+Discussion first; a PR that turns out to be non-trivial will be asked to.
+
+A substantive PR with no backing issue/RFC will be closed with a pointer to the
+right channel — not as a judgment of the idea, but to keep design discussion
+where it's reviewable.
+
+## What maintainers do *not* gate
+Maintainers' own changes do not pass through the intake gates above — the team
+runs a separate internal process. The universal gates (review, CODEOWNERS,
+branch protection, CI) apply to everyone. Enforcement of the intake rules is, to
+start, **by convention and review** (PR template + labels); an automated check
+keyed to author association may be added later if volume warrants.
+
+## Code of conduct & security
+- Conduct: [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md).
+- Security issues are **not** public Issues — see [SECURITY.md](SECURITY.md).
+
+## Changing this document
+Governance changes the same way code does: a pull request, reviewed by
+maintainers. This file describes the external surface; the internal maintainer
+process is intentionally out of scope here.
diff --git a/crates/omnigraph/src/db/manifest.rs b/crates/omnigraph/src/db/manifest.rs
index 7fcf7de..3b2886f 100644
--- a/crates/omnigraph/src/db/manifest.rs
+++ b/crates/omnigraph/src/db/manifest.rs
@@ -48,6 +48,22 @@ const OBJECT_TYPE_TABLE_VERSION: &str = "table_version";
 const OBJECT_TYPE_TABLE_TOMBSTONE: &str = "table_tombstone";
 const TABLE_VERSION_MANAGEMENT_KEY: &str = "table_version_management";
 
+/// Apply pending internal-schema migrations against `__manifest` on the
+/// open-for-write path, independent of a publish.
+///
+/// `Omnigraph::open(ReadWrite)` calls this before the coordinator reads branch
+/// state, so branch-observing code (`branch_list`, the schema-apply
+/// blocking-branch checks) sees the post-migration graph. In particular the
+/// v2→v3 step sweeps legacy `__run__*` staging branches off `__manifest`
+/// (MR-770); running it here closes the window where those branches would
+/// otherwise block schema apply before the first publish runs the migration.
+///
+/// Idempotent: a no-op stamp read when the on-disk version already matches.
+pub(crate) async fn migrate_on_open(root_uri: &str) -> Result<()> {
+    let mut dataset = open_manifest_dataset(root_uri, None).await?;
+    migrations::migrate_internal_schema(&mut dataset).await
+}
+
 /// Immutable point-in-time view of the database.
 ///
 /// Cheap to create (no storage I/O). All reads within a query go through one
diff --git a/crates/omnigraph/src/db/manifest/migrations.rs b/crates/omnigraph/src/db/manifest/migrations.rs
index bbb7995..e2801fe 100644
--- a/crates/omnigraph/src/db/manifest/migrations.rs
+++ b/crates/omnigraph/src/db/manifest/migrations.rs
@@ -46,7 +46,11 @@ use crate::error::{OmniError, Result};
 /// - v2 — `__manifest.object_id` carries the unenforced-PK annotation,
 ///   engaging Lance's bloom-filter conflict resolver at commit time. Added
 ///   alongside `expected_table_versions` OCC on `ManifestBatchPublisher::publish`.
-pub(super) const INTERNAL_MANIFEST_SCHEMA_VERSION: u32 = 2;
+/// - v3 — one-time sweep of legacy `__run__<id>` staging branches left on the
+///   `__manifest` dataset by the pre-v0.4.0 Run state machine (removed in
+///   MR-771). Once swept, the `is_internal_run_branch` defense-in-depth guard
+///   is no longer needed (MR-770).
+pub(super) const INTERNAL_MANIFEST_SCHEMA_VERSION: u32 = 3;
 
 const INTERNAL_SCHEMA_VERSION_KEY: &str = "omnigraph:internal_schema_version";
 const OBJECT_ID_PK_KEY: &str = "lance-schema:unenforced-primary-key";
@@ -89,6 +93,10 @@ pub(super) async fn migrate_internal_schema(dataset: &mut Dataset) -> Result<()>
                 migrate_v1_to_v2(dataset).await?;
                 current = 2;
             }
+            2 => {
+                migrate_v2_to_v3(dataset).await?;
+                current = 3;
+            }
             other => {
                 return Err(OmniError::manifest_internal(format!(
                     "no internal-schema migration registered for v{} → v{}",
@@ -122,6 +130,51 @@ async fn migrate_v1_to_v2(dataset: &mut Dataset) -> Result<()> {
     set_stamp(dataset, 2).await
 }
 
+/// v2 → v3: sweep legacy `__run__<id>` staging branches off the `__manifest`
+/// dataset, then bump the stamp.
+///
+/// The pre-v0.4.0 Run state machine (removed in MR-771) created graph-level
+/// staging branches named `__run__<ulid>` on `__manifest`. MR-771 stopped
+/// creating them but left any pre-existing ones in place; Lance's
+/// `list_branches` still enumerates them, so they leak into `branch_list()`
+/// and count as blocking branches at schema-apply time. This one-time sweep
+/// removes them so the `is_internal_run_branch` guard can retire (MR-770).
+///
+/// The `"__run__"` prefix is inlined here on purpose: this migration must keep
+/// working after the `run_registry` module (the guard) is deleted, so it does
+/// not depend on it.
+///
+/// Idempotent under both sequential retry and concurrent runners: each run
+/// re-enumerates `list_branches` fresh, and `force_delete_branch` tolerates a
+/// branch that is already gone — so a crash before the stamp bump, or a second
+/// process opening the same legacy graph at the same time, never errors out.
+async fn migrate_v2_to_v3(dataset: &mut Dataset) -> Result<()> {
+    const LEGACY_RUN_BRANCH_PREFIX: &str = "__run__";
+    let branches = dataset
+        .list_branches()
+        .await
+        .map_err(|e| OmniError::Lance(e.to_string()))?;
+    let run_branches: Vec<String> = branches
+        .into_keys()
+        .filter(|name| {
+            name.trim_start_matches('/')
+                .starts_with(LEGACY_RUN_BRANCH_PREFIX)
+        })
+        .collect();
+    for name in run_branches {
+        // `force_delete_branch` deletes even when the `BranchContents` is
+        // already gone. Plain `delete_branch` errors "BranchContents not
+        // found", which would fail a second concurrent open (or a retry that
+        // raced another runner) after the first one swept the branch. Force is
+        // exactly Lance's documented path for cleaning up zombie branches.
+        dataset
+            .force_delete_branch(&name)
+            .await
+            .map_err(|e| OmniError::Lance(e.to_string()))?;
+    }
+    set_stamp(dataset, 3).await
+}
+
 async fn set_stamp(dataset: &mut Dataset, version: u32) -> Result<()> {
     dataset
         .update_schema_metadata([(INTERNAL_SCHEMA_VERSION_KEY.to_string(), version.to_string())])
diff --git a/crates/omnigraph/src/db/manifest/tests.rs b/crates/omnigraph/src/db/manifest/tests.rs
index effa0b5..885a2a8 100644
--- a/crates/omnigraph/src/db/manifest/tests.rs
+++ b/crates/omnigraph/src/db/manifest/tests.rs
@@ -1461,6 +1461,80 @@ async fn test_publish_migrates_pre_stamp_manifest_to_current_version() {
     assert!(reopened.snapshot().entry("node:Person").is_some());
 }
 
+#[tokio::test]
+async fn test_v2_to_v3_sweeps_legacy_run_branches_on_write_open() {
+    let dir = tempfile::tempdir().unwrap();
+    let uri = dir.path().to_str().unwrap();
+    let catalog = build_test_catalog();
+    let mut mc = ManifestCoordinator::init(uri, &catalog).await.unwrap();
+
+    // Synthesize a pre-MR-770 graph: several stale `__run__` staging branches
+    // left on `__manifest` (a real legacy graph accumulates one per run), plus
+    // a real user branch that must survive the sweep. Multiple run branches
+    // exercise the migration's delete loop on a single reused dataset handle.
+    mc.create_branch("__run__01J9LEGACY").await.unwrap();
+    mc.create_branch("__run__01J9SECOND").await.unwrap();
+    mc.create_branch("__run__01J9THIRD").await.unwrap();
+    mc.create_branch("feature").await.unwrap();
+    let before = mc.list_branches().await.unwrap();
+    assert_eq!(
+        before.iter().filter(|b| b.starts_with("__run__")).count(),
+        3,
+        "precondition: three legacy run branches exist on __manifest; got {before:?}",
+    );
+
+    // Rewind the internal-schema stamp to v2 so the next write-open runs the
+    // v2 → v3 sweep arm (init stamps at the current version, which is past it).
+    {
+        let mut ds = open_manifest_dataset(uri, None).await.unwrap();
+        ds.update_schema_metadata([(
+            "omnigraph:internal_schema_version".to_string(),
+            Some("2".to_string()),
+        )])
+        .await
+        .unwrap();
+        let post = open_manifest_dataset(uri, None).await.unwrap();
+        assert_eq!(super::migrations::read_stamp(&post), 2, "stamp rewound to v2");
+    }
+
+    // A no-op publish forces the open-for-write path, which runs the migration.
+    let mut expected = HashMap::new();
+    expected.insert("node:Person".to_string(), 1);
+    GraphNamespacePublisher::new(uri, None)
+        .publish(&[], &expected)
+        .await
+        .unwrap();
+
+    // Stamp advanced to current; the legacy run branch is physically gone from
+    // `__manifest` (checked via the raw, unfiltered manifest list — not the
+    // guard-filtered `branch_list`), and the real branch + `main` survive.
+    let post = open_manifest_dataset(uri, None).await.unwrap();
+    assert_eq!(
+        super::migrations::read_stamp(&post),
+        super::migrations::INTERNAL_MANIFEST_SCHEMA_VERSION,
+    );
+    let reopened = ManifestCoordinator::open(uri).await.unwrap();
+    let after = reopened.list_branches().await.unwrap();
+    assert!(
+        !after.iter().any(|b| b.starts_with("__run__")),
+        "legacy run branch must be swept; got {after:?}",
+    );
+    assert!(after.iter().any(|b| b == "feature"), "user branch must survive");
+    assert!(after.iter().any(|b| b == "main"), "main must survive");
+
+    // Idempotent: a second write-open finds the stamp at current and does not
+    // re-run the sweep or error.
+    GraphNamespacePublisher::new(uri, None)
+        .publish(&[], &expected)
+        .await
+        .unwrap();
+    let final_ds = open_manifest_dataset(uri, None).await.unwrap();
+    assert_eq!(
+        super::migrations::read_stamp(&final_ds),
+        super::migrations::INTERNAL_MANIFEST_SCHEMA_VERSION,
+    );
+}
+
 #[tokio::test]
 async fn test_publish_rejects_manifest_stamped_at_future_version() {
     let dir = tempfile::tempdir().unwrap();
diff --git a/crates/omnigraph/src/db/mod.rs b/crates/omnigraph/src/db/mod.rs
index 8702f88..13e1c74 100644
--- a/crates/omnigraph/src/db/mod.rs
+++ b/crates/omnigraph/src/db/mod.rs
@@ -3,7 +3,6 @@ pub mod graph_coordinator;
 pub mod manifest;
 mod omnigraph;
 mod recovery_audit;
-mod run_registry;
 mod schema_state;
 pub(crate) mod write_queue;
 
@@ -15,7 +14,6 @@ pub use omnigraph::{
     CleanupPolicyOptions, InitOptions, MergeOutcome, Omnigraph, OpenMode, SchemaApplyOptions,
     SchemaApplyResult, SkipReason, TableCleanupStats, TableOptimizeStats,
 };
-pub(crate) use run_registry::is_internal_run_branch;
 
 pub(crate) const SCHEMA_APPLY_LOCK_BRANCH: &str = "__schema_apply_lock__";
 
@@ -69,5 +67,8 @@ pub(crate) fn is_schema_apply_lock_branch(name: &str) -> bool {
 }
 
 pub(crate) fn is_internal_system_branch(name: &str) -> bool {
-    is_internal_run_branch(name) || is_schema_apply_lock_branch(name)
+    // Legacy `__run__*` staging branches (Run state machine, removed MR-771)
+    // are swept off `__manifest` by the v2→v3 internal-schema migration, so the
+    // only internal branch the engine still creates is the schema-apply lock.
+    is_schema_apply_lock_branch(name)
 }
diff --git a/crates/omnigraph/src/db/omnigraph.rs b/crates/omnigraph/src/db/omnigraph.rs
index f10739e..8d23ebb 100644
--- a/crates/omnigraph/src/db/omnigraph.rs
+++ b/crates/omnigraph/src/db/omnigraph.rs
@@ -347,6 +347,16 @@ impl Omnigraph {
         mode: OpenMode,
     ) -> Result<Self> {
         let root = normalize_root_uri(uri)?;
+        // Apply pending internal-schema migrations before the coordinator reads
+        // branch state, so `branch_list` and the schema-apply blocking-branch
+        // checks observe the post-migration graph — notably the v2→v3 sweep of
+        // legacy `__run__*` staging branches (MR-770). ReadWrite only: a
+        // read-only open must not trigger object-store writes, so a read-only
+        // open of an unmigrated legacy graph still lists `__run__*` until its
+        // first read-write open (an accepted, documented limitation).
+        if matches!(mode, OpenMode::ReadWrite) {
+            crate::db::manifest::migrate_on_open(&root).await?;
+        }
         // Open the coordinator first so the schema-staging recovery sweep can
         // compare its snapshot against any leftover staging files.
         let mut coordinator = GraphCoordinator::open(&root, Arc::clone(&storage)).await?;
@@ -1494,12 +1504,6 @@ pub(crate) fn normalize_branch_name(branch: &str) -> Result<Option<String>> {
 }
 
 pub(crate) fn ensure_public_branch_ref(branch: &str, operation: &str) -> Result<()> {
-    if super::is_internal_run_branch(branch) {
-        return Err(OmniError::manifest(format!(
-            "{} does not allow internal run ref '{}'",
-            operation, branch
-        )));
-    }
     if is_internal_system_branch(branch) {
         return Err(OmniError::manifest(format!(
             "{} does not allow internal system ref '{}'",
@@ -1903,7 +1907,6 @@ fn json_value_from_array(array: &dyn Array, row: usize) -> Result<serde_json::Va
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::db::is_internal_run_branch;
     use crate::db::manifest::ManifestCoordinator;
     use async_trait::async_trait;
     use serde_json::Value;
@@ -2245,11 +2248,11 @@ edge WorksAt: Person -> Company
     #[tokio::test]
     async fn test_apply_schema_succeeds_after_load() {
         // Historical: schema apply used to be blocked by leftover
-        // `__run__` branches. A defense-in-depth filter now skips
-        // internal system branches, and run branches were made
-        // ephemeral on every terminal state — so in practice no
-        // `__run__` branch survives publish. The filter still guards
-        // the invariant.
+        // `__run__` branches. The Run state machine was removed in
+        // MR-771, so a fresh graph never creates a `__run__` branch;
+        // legacy ones are swept by the v2→v3 manifest migration. This
+        // asserts the invariant a current graph upholds: publish leaves
+        // no `__run__` branch behind, so schema apply proceeds.
         let dir = tempfile::tempdir().unwrap();
         let uri = dir.path().to_str().unwrap();
         let mut db = Omnigraph::init(uri, TEST_SCHEMA).await.unwrap();
@@ -2264,8 +2267,8 @@ edge WorksAt: Person -> Company
 
         let all_branches = db.coordinator.read().await.all_branches().await.unwrap();
         assert!(
-            !all_branches.iter().any(|b| is_internal_run_branch(b)),
-            "run branch should be deleted after publish, got: {:?}",
+            !all_branches.iter().any(|b| b.starts_with("__run__")),
+            "no __run__ branch should exist after publish, got: {:?}",
             all_branches
         );
 
@@ -2277,6 +2280,56 @@ edge WorksAt: Person -> Company
         assert!(result.applied, "schema apply should have applied");
     }
 
+    /// Regression (MR-770): a pre-v0.4.0 graph that still carries a stale
+    /// `__run__*` branch on `__manifest` must not block schema apply. The
+    /// v2→v3 sweep runs in `Omnigraph::open(ReadWrite)` — before the
+    /// schema-apply blocking-branch check — so apply succeeds with no
+    /// intervening publish.
+    ///
+    /// Confirmed to fail before the open-time migration landed: the reopened
+    /// graph still listed `__run__legacy`, and `apply_schema` returned
+    /// "found non-main branches: __run__legacy".
+    #[tokio::test]
+    async fn legacy_run_branch_is_swept_on_open_and_does_not_block_schema_apply() {
+        let dir = tempfile::tempdir().unwrap();
+        let uri = dir.path().to_str().unwrap();
+        let mut db = Omnigraph::init(uri, TEST_SCHEMA).await.unwrap();
+
+        // Synthesize a legacy graph: a stale `__run__` branch on `__manifest`
+        // plus the manifest stamp rewound to v2 (pre-sweep).
+        db.branch_create("__run__legacy").await.unwrap();
+        drop(db);
+        {
+            let mut ds = lance::Dataset::open(&format!("{}/__manifest", uri))
+                .await
+                .unwrap();
+            ds.update_schema_metadata([(
+                "omnigraph:internal_schema_version".to_string(),
+                Some("2".to_string()),
+            )])
+            .await
+            .unwrap();
+        }
+
+        // Reopen (ReadWrite): the open-time migration must sweep `__run__legacy`
+        // before any branch-observing code runs.
+        let db = Omnigraph::open(uri).await.unwrap();
+        let branches = db.branch_list().await.unwrap();
+        assert!(
+            !branches.iter().any(|b| b.starts_with("__run__")),
+            "open-time migration must sweep legacy __run__ branches; got {branches:?}",
+        );
+
+        // Schema apply must proceed with no intervening publish — the
+        // blocking-branch check no longer sees `__run__legacy`.
+        let desired = TEST_SCHEMA.replace(
+            "    age: I32?\n}",
+            "    age: I32?\n    nickname: String?\n}",
+        );
+        let result = db.apply_schema(&desired).await.unwrap();
+        assert!(result.applied, "schema apply should have applied");
+    }
+
     #[tokio::test]
     async fn test_apply_schema_adds_index_for_existing_property() {
         let dir = tempfile::tempdir().unwrap();
diff --git a/crates/omnigraph/src/db/omnigraph/schema_apply.rs b/crates/omnigraph/src/db/omnigraph/schema_apply.rs
index 76fa58f..0a01163 100644
--- a/crates/omnigraph/src/db/omnigraph/schema_apply.rs
+++ b/crates/omnigraph/src/db/omnigraph/schema_apply.rs
@@ -61,11 +61,11 @@ async fn plan_schema_for_apply(
 ) -> Result<PlannedSchemaApply> {
     db.ensure_schema_state_valid().await?;
     let branches = db.coordinator.read().await.all_branches().await?;
-    // Skip `main` and internal system branches. The schema-apply lock branch
-    // is excluded because it is the cluster-wide schema-apply serializer.
-    // `__run__*` branches are no longer created; the filter remains as
-    // defense-in-depth for legacy graphs with leftover staging branches.
-    // A future production sweep will let this guard go.
+    // Skip `main` and internal system branches (the schema-apply lock branch,
+    // the cluster-wide schema-apply serializer). Legacy `__run__*` staging
+    // branches were swept off `__manifest` by the v2→v3 migration that runs in
+    // `Omnigraph::open(ReadWrite)` before this check (MR-770), so they no
+    // longer appear here.
     let blocking_branches = branches
         .into_iter()
         .filter(|branch| branch != "main" && !is_internal_system_branch(branch))
diff --git a/crates/omnigraph/src/db/run_registry.rs b/crates/omnigraph/src/db/run_registry.rs
deleted file mode 100644
index ee3d336..0000000
--- a/crates/omnigraph/src/db/run_registry.rs
+++ /dev/null
@@ -1,16 +0,0 @@
-// The Run state machine has been removed. Mutations now write directly
-// to target tables and use the publisher's `expected_table_versions`
-// CAS for cross-table OCC; `__run__<id>` staging branches and the
-// `_graph_runs.lance` state machine no longer exist.
-//
-// What remains is the branch-name predicate, kept as a defense-in-depth
-// guard against users naming a public branch `__run__*`. A future
-// production sweep of legacy `_graph_runs.lance` rows and stale
-// `__run__*` branches will let this predicate (and this file) go too.
-
-pub(crate) const INTERNAL_RUN_BRANCH_PREFIX: &str = "__run__";
-
-pub(crate) fn is_internal_run_branch(name: &str) -> bool {
-    name.trim_start_matches('/')
-        .starts_with(INTERNAL_RUN_BRANCH_PREFIX)
-}
diff --git a/crates/omnigraph/src/exec/merge.rs b/crates/omnigraph/src/exec/merge.rs
index a161e42..88ec3cb 100644
--- a/crates/omnigraph/src/exec/merge.rs
+++ b/crates/omnigraph/src/exec/merge.rs
@@ -1092,9 +1092,9 @@ impl Omnigraph {
         target: &str,
         actor_id: Option<&str>,
     ) -> Result<MergeOutcome> {
-        if is_internal_run_branch(source) || is_internal_run_branch(target) {
+        if is_internal_system_branch(source) || is_internal_system_branch(target) {
             return Err(OmniError::manifest(format!(
-                "branch_merge does not allow internal run refs ('{}' -> '{}')",
+                "branch_merge does not allow internal system refs ('{}' -> '{}')",
                 source, target
             )));
         }
diff --git a/crates/omnigraph/src/exec/mod.rs b/crates/omnigraph/src/exec/mod.rs
index a114b78..4076414 100644
--- a/crates/omnigraph/src/exec/mod.rs
+++ b/crates/omnigraph/src/exec/mod.rs
@@ -35,7 +35,7 @@ use time::format_description::well_known::Rfc3339;
 
 use crate::db::commit_graph::CommitGraph;
 use crate::db::manifest::ManifestCoordinator;
-use crate::db::{MergeOutcome, Omnigraph, is_internal_run_branch};
+use crate::db::{MergeOutcome, Omnigraph, is_internal_system_branch};
 use crate::db::{ReadTarget, Snapshot};
 use crate::embedding::EmbeddingClient;
 use crate::error::{MergeConflict, MergeConflictKind, OmniError, Result};
diff --git a/crates/omnigraph/src/loader/mod.rs b/crates/omnigraph/src/loader/mod.rs
index 9876309..9d7329d 100644
--- a/crates/omnigraph/src/loader/mod.rs
+++ b/crates/omnigraph/src/loader/mod.rs
@@ -288,21 +288,24 @@ async fn load_jsonl_reader<R: BufRead>(
     let mut node_rows: HashMap<String, Vec<JsonValue>> = HashMap::new();
     let mut edge_rows: HashMap<String, Vec<(String, String, JsonValue)>> = HashMap::new();
 
-    for (line_num, line) in reader.lines().enumerate() {
-        let line = line?;
-        let line = line.trim();
-        if line.is_empty() {
-            continue;
-        }
-        let value: JsonValue = serde_json::from_str(line).map_err(|e| {
-            OmniError::manifest(format!("invalid JSON on line {}: {}", line_num + 1, e))
+    // Parse a stream of JSON values. Accepts both compact JSONL (one object
+    // per line) and pretty-printed JSON where a single object spans multiple
+    // lines — serde's streaming deserializer treats any whitespace (including
+    // newlines) between top-level values as a separator.
+    for (idx, parsed) in serde_json::Deserializer::from_reader(reader)
+        .into_iter::<JsonValue>()
+        .enumerate()
+    {
+        let record_num = idx + 1;
+        let value: JsonValue = parsed.map_err(|e| {
+            OmniError::manifest(format!("invalid JSON at record {}: {}", record_num, e))
         })?;
 
         if let Some(type_name) = value.get("type").and_then(|v| v.as_str()) {
             if !catalog.node_types.contains_key(type_name) {
                 return Err(OmniError::manifest(format!(
-                    "line {}: unknown node type '{}'",
-                    line_num + 1,
+                    "record {}: unknown node type '{}'",
+                    record_num,
                     type_name
                 )));
             }
@@ -317,8 +320,8 @@ async fn load_jsonl_reader<R: BufRead>(
         } else if let Some(edge_name) = value.get("edge").and_then(|v| v.as_str()) {
             if catalog.lookup_edge_by_name(edge_name).is_none() {
                 return Err(OmniError::manifest(format!(
-                    "line {}: unknown edge type '{}'",
-                    line_num + 1,
+                    "record {}: unknown edge type '{}'",
+                    record_num,
                     edge_name
                 )));
             }
@@ -326,14 +329,14 @@ async fn load_jsonl_reader<R: BufRead>(
                 .get("from")
                 .and_then(|v| v.as_str())
                 .ok_or_else(|| {
-                    OmniError::manifest(format!("line {}: edge missing 'from'", line_num + 1))
+                    OmniError::manifest(format!("record {}: edge missing 'from'", record_num))
                 })?
                 .to_string();
             let to = value
                 .get("to")
                 .and_then(|v| v.as_str())
                 .ok_or_else(|| {
-                    OmniError::manifest(format!("line {}: edge missing 'to'", line_num + 1))
+                    OmniError::manifest(format!("record {}: edge missing 'to'", record_num))
                 })?
                 .to_string();
             let data = value
@@ -347,8 +350,8 @@ async fn load_jsonl_reader<R: BufRead>(
                 .push((from, to, data));
         } else {
             return Err(OmniError::manifest(format!(
-                "line {}: expected 'type' or 'edge' field",
-                line_num + 1
+                "record {}: expected 'type' or 'edge' field",
+                record_num
             )));
         }
     }
diff --git a/crates/omnigraph/tests/writes.rs b/crates/omnigraph/tests/writes.rs
index 13cb10f..0a309c9 100644
--- a/crates/omnigraph/tests/writes.rs
+++ b/crates/omnigraph/tests/writes.rs
@@ -371,11 +371,10 @@ async fn cancelled_mutation_future_leaves_no_state() {
 
     // Cancel-safety property: no graph-level run/staging state remains.
     //
-    // Note: `branch_list()` already filters `__run__*` via
-    // `is_internal_system_branch`, so a runtime "no `__run__` branches" check
-    // would be vacuous. The structural property that no `__run__` branches
-    // can ever be created is enforced by deletion of `begin_run` etc. in
-    // (verified by the build itself — those symbols no longer exist).
+    // No `__run__` branches can ever be created: the Run state machine
+    // (`begin_run` etc.) was deleted in MR-771 — verified by the build itself,
+    // those symbols no longer exist. Any legacy `__run__*` branch on an
+    // upgraded graph is swept by the v2→v3 manifest migration.
     //
     // (1) The branch list is unchanged: cancellation/completion cannot
     //     synthesize new public branches.
@@ -442,34 +441,40 @@ async fn repeated_loads_do_not_accumulate_branches() {
     assert_eq!(db.branch_list().await.unwrap(), vec!["main".to_string()]);
 }
 
-/// User code must not be able to write to internal `__run__*` names.
-/// The branch-name guard predicate is kept as defense-in-depth; it
-/// will be removed once a future production sweep retires the legacy
-/// branches.
+/// After MR-770, `__run__*` is an ordinary branch name — the Run state machine
+/// and its `is_internal_run_branch` guard are gone. The surviving internal-ref
+/// guard still rejects the active `__schema_apply_lock__` branch on the public
+/// create/merge APIs.
 #[tokio::test]
-async fn public_branch_apis_reject_internal_run_refs() {
+async fn public_branch_apis_reject_internal_system_refs() {
     let dir = tempfile::tempdir().unwrap();
     let mut db = init_and_load(&dir).await;
 
-    let create_err = db.branch_create("__run__synthetic").await.unwrap_err();
+    // `__run__*` is no longer reserved — creating it now succeeds.
+    db.branch_create("__run__formerly_reserved")
+        .await
+        .expect("__run__ prefix is a normal branch name post-MR-770");
+
+    // The schema-apply lock branch is still rejected on public branch APIs.
+    let create_err = db.branch_create("__schema_apply_lock__").await.unwrap_err();
     let OmniError::Manifest(err) = create_err else {
         panic!("expected Manifest error");
     };
     assert!(
-        err.message.contains("internal run ref"),
+        err.message.contains("internal system ref"),
         "unexpected error: {}",
         err.message
     );
 
     let merge_err = db
-        .branch_merge("__run__synthetic", "main")
+        .branch_merge("__schema_apply_lock__", "main")
         .await
         .unwrap_err();
     let OmniError::Manifest(err) = merge_err else {
         panic!("expected Manifest error");
     };
     assert!(
-        err.message.contains("internal run refs"),
+        err.message.contains("internal system refs"),
         "unexpected error: {}",
         err.message
     );
diff --git a/docs/dev/codeowners.md b/docs/dev/codeowners.md
index 14bba0b..50c4dc7 100644
--- a/docs/dev/codeowners.md
+++ b/docs/dev/codeowners.md
@@ -14,8 +14,8 @@ The tables below are **generated** from `.github/codeowners-roles.yml` by `.gith
 
 | Path | Owners | Role(s) |
 |---|---|---|
-| `*` | @ragnorc | engineering |
-| `crates/**` | @ragnorc | engineering |
+| `*` | @ragnorc @aaltshuler | engineering |
+| `crates/**` | @ragnorc @aaltshuler | engineering |
 | `docs/**` | @ragnorc | docs |
 | `README.md` | @ragnorc | docs |
 | `AGENTS.md` | @ragnorc | docs |
@@ -26,7 +26,7 @@ The tables below are **generated** from `.github/codeowners-roles.yml` by `.gith
 
 | Role | Members | Description |
 |---|---|---|
-| `engineering` | @ragnorc | All production code under crates/**. Engine, CLI, server, compiler. |
+| `engineering` | @ragnorc @aaltshuler | All production code under crates/**. Engine, CLI, server, compiler. |
 | `docs` | @ragnorc | Documentation under docs/**, plus repo-level docs (README.md, AGENTS.md, CLAUDE.md symlink, SECURITY.md). |
 
 <!-- END GENERATED OWNERSHIP -->
diff --git a/docs/dev/index.md b/docs/dev/index.md
index 600c969..1e41342 100644
--- a/docs/dev/index.md
+++ b/docs/dev/index.md
@@ -51,6 +51,18 @@ constraints. User-facing behavior should still be documented through
 | Install and deployment packaging | [install.md](../user/install.md), [deployment.md](../user/deployment.md) |
 | Release history | [releases/](../releases/) |
 
+## Contribution & Governance
+
+| Area | Read |
+|---|---|
+| How to contribute (external) | [CONTRIBUTING.md](../../CONTRIBUTING.md) |
+| Governance model, roles, decision authority | [GOVERNANCE.md](../../GOVERNANCE.md) |
+| Public contribution RFC track | [rfcs/](../rfcs/) |
+
+The `docs/rfcs/` track is the **public, externally-authorable** RFC process. The
+maintainer/internal RFCs below (`rfc-00N-*.md`) are a separate, team-owned
+track; don't conflate the two.
+
 ## Active Implementation Plans
 
 Working documents for in-flight feature work. Removed when the work lands.
diff --git a/docs/dev/writes.md b/docs/dev/writes.md
index 4f03f2a..c346858 100644
--- a/docs/dev/writes.md
+++ b/docs/dev/writes.md
@@ -14,8 +14,11 @@ publisher's row-level CAS on `__manifest` is the single fence.
 
 - No `RunRecord`, no `_graph_runs.lance`, no `_graph_run_actors.lance`.
 - No `omnigraph run *` CLI subcommands and no `/runs/*` HTTP endpoints.
-- No `__run__<id>` staging branches. (Legacy on-disk artifacts from
-  pre-MR-771 repos are inert; MR-770 sweeps them in production.)
+- No `__run__<id>` staging branches; `__run__*` is no longer a reserved
+  name. The branch-name guard was removed in MR-770, and any stale
+  `__run__*` branch on an upgraded graph is swept off `__manifest` by the
+  v2→v3 internal-schema migration on first read-write open. (The inert
+  `_graph_runs.lance` bytes remain until a `delete_prefix` primitive lands.)
 - Cancelled mutation futures leave **no graph-level state** — only orphaned
   Lance fragments, which the existing `omnigraph cleanup` pipe reclaims.
 
@@ -241,9 +244,14 @@ list`.
 
 ## Migration code
 
-`db/manifest/migrations.rs` does not change. Active deletion of
-`_graph_runs.lance` belongs in MR-770 (the production sweep) — this PR
-stops *creating* run state but does not destroy legacy bytes on disk.
+`db/manifest/migrations.rs` carries the v2→v3 internal-schema step (MR-770):
+a one-time sweep that deletes legacy `__run__*` staging branches off
+`__manifest`. It runs in `Omnigraph::open(ReadWrite)` (via
+`manifest::migrate_on_open`, before the coordinator reads branch state) and
+again on the publisher's write path; both are idempotent once the stamp is at
+v3. Deleting the inert `_graph_runs.lance` / `_graph_run_actors.lance` dataset
+*bytes* is still deferred — it needs a `StorageAdapter::delete_prefix`
+primitive — but those bytes are invisible to graph-level state.
 
 ## Mid-query partial failure: closed by MR-794
 
diff --git a/docs/releases/v0.6.1.md b/docs/releases/v0.6.1.md
index aafe1af..0acc34b 100644
--- a/docs/releases/v0.6.1.md
+++ b/docs/releases/v0.6.1.md
@@ -7,6 +7,7 @@ v0.6.1 focuses on operational polish after v0.6.0: stored-query registries, safe
 - **Stored-query registries.** `omnigraph.yaml` can declare curated `queries:` blocks per graph. Servers load and type-check them at startup, `omnigraph queries validate` checks them offline, `omnigraph queries list` shows exposed queries and typed params, `GET /queries` exposes a typed catalog, and `POST /queries/{name}` invokes a stored query without accepting ad hoc `.gq` source from the client.
 - **Stored-query policy gate.** New Cedar action `invoke_query` gates the stored-query invocation surface. Stored mutations are double-gated: `invoke_query` to reach the stored query and `change` for the actual write.
 - **Safer branch deletion.** `branch_delete` now treats the manifest as the authority, flips branch visibility atomically, and reclaims per-table/commit-graph forks as derived state. If best-effort reclaim is interrupted, `cleanup` reconciles orphaned forks; reusing a branch name before cleanup reports an actionable error.
+- **Legacy `__run__` cleanup (MR-770).** Removed the last functional remnant of the Run state machine (retired in v0.4.0): the `__run__` branch-name guard. A new v2→v3 `__manifest` internal-schema migration sweeps any stale `__run__*` staging branches on the first read-write open, so `__run__*` is no longer a reserved branch name. This closes the "unpromoted `__run__` branches block reads" condition behind the zombie-run cascade incident; the inert `_graph_runs.lance` row cleanup is tracked separately (it needs a `delete_prefix` primitive).
 - **Blob-safe optimize.** `omnigraph optimize` skips tables with `Blob` properties instead of failing the whole sweep on Lance's blob-v2 compaction decode bug. Skips are visible in human output, `--json` as `skipped`, `TableOptimizeStats.skipped`, and logs; non-blob tables still compact normally.
 - **Deployment improvements.** The container entrypoint now composes `OMNIGRAPH_TARGET_URI` with `OMNIGRAPH_CONFIG`, so operators can keep the graph URI in env while loading policy/query config from a mounted file. The local RustFS bootstrap pins RustFS beta.3 and allows the current insecure local-dev default credentials.
 - **Windows release support.** Tagged and edge releases now publish Windows x86_64 archives containing `omnigraph.exe` and `omnigraph-server.exe`, with a PowerShell installer and Windows install docs.
@@ -17,6 +18,7 @@ v0.6.1 focuses on operational polish after v0.6.0: stored-query registries, safe
 - A graph selected by name (`--target` or `server.graph`) now uses `graphs.<name>.policy` and `graphs.<name>.queries`. Top-level `policy` / `queries` blocks are only for anonymous bare-URI single-graph mode; using them with a named graph now fails loudly with migration guidance.
 - `mcp.expose` defaults to `true` for stored-query registry entries. Set `mcp: { expose: false }` for service-only queries that should not appear in the catalog.
 - `invoke_query` is graph-scoped, not branch-scoped. Branch/snapshot access remains enforced by the inner `read` / `change` gate.
+- **Legacy `__run__` migration.** Graphs created before v0.4.0 are migrated automatically on the first **read-write** open by a v0.6.1 binary (one-time `__manifest` stamp v2→v3 sweep of stale `__run__*` branches). No action required. Two caveats: (1) a graph opened **read-only** still lists any stale `__run__*` branch until its first read-write open, since the migration is write-path-only like all manifest migrations — long-lived read-only deployments should be opened read-write once after upgrading; (2) the inert `_graph_runs.lance` / `_graph_run_actors.lance` dataset bytes are left in place until a future `delete_prefix` primitive (they are invisible to graph-level state).
 - Blob tables are not compacted until the upstream Lance fix lands, so fragment count and deleted-row space on blob tables are not reclaimed by `optimize`. Reads, writes, and query results are unaffected; no on-disk migration is required.
 - `TableOptimizeStats` is now `#[non_exhaustive]` and gains a `skipped: Option<SkipReason>` field (so does the new `SkipReason` enum). This is a source-level change only for downstream code that built this returned result struct by literal — rare, since it is produced by `optimize` and consumed by reading its fields; field access is unaffected, and `#[non_exhaustive]` keeps future additions non-breaking.
 
diff --git a/docs/rfcs/0000-template.md b/docs/rfcs/0000-template.md
new file mode 100644
index 0000000..48f4bda
--- /dev/null
+++ b/docs/rfcs/0000-template.md
@@ -0,0 +1,54 @@
+# RFC NNNN: <title>
+
+| | |
+|---|---|
+| **Status** | Proposed |
+| **Author(s)** | <your name / handle> |
+| **Discussion** | <link to the originating Discussion, if any> |
+| **Implementation** | <issue/PR links, filled in as work lands> |
+
+> Status is maintained by maintainers: `Proposed` while the PR is open,
+> `Accepted` on merge, `Declined` on close, `Superseded by NNNN` later.
+
+## Summary
+
+One paragraph: what this changes, in plain terms.
+
+## Motivation
+
+What problem does this solve, and why is it worth the ongoing cost? Tie it to a
+concrete need (a Discussion, a recurring issue, a user request). Per the
+project's first principle, argue the *long-run liability*, not just the
+short-term convenience.
+
+## Guide-level explanation
+
+Explain the change as you'd teach it to a user or contributor: new commands,
+syntax, API shapes, behavior. Examples first.
+
+## Reference-level design
+
+The precise design: data structures, IR/AST/planner changes, storage/format
+impact, migration path, error behavior. Enough that a reviewer can find the
+holes.
+
+## Invariants & deny-list check
+
+Which Hard Invariants in [../dev/invariants.md](../dev/invariants.md) does this
+touch? Does it brush against any deny-list item — and if so, why is this the
+justified exception? State explicitly that no invariant is weakened, or which
+Known Gap moves.
+
+## Drawbacks & alternatives
+
+What does this cost, what did you reject, and why. "Do nothing" is a valid
+alternative to weigh.
+
+## Reversibility
+
+Is this reversible? On-disk/wire/format and substrate choices are near-permanent
+and demand more evidence; a CLI flag or doc is cheap to undo. Say which this is.
+
+## Unresolved questions
+
+What's deliberately left open for review to settle.
diff --git a/docs/rfcs/README.md b/docs/rfcs/README.md
new file mode 100644
index 0000000..99cdd76
--- /dev/null
+++ b/docs/rfcs/README.md
@@ -0,0 +1,66 @@
+# RFCs
+
+Substantial changes to OmniGraph — new user-facing surface, format or protocol
+changes, anything irreversible or cross-cutting — go through a lightweight RFC
+so the design is agreed *as reviewable code* before implementation starts. This
+is the public RFC track, open to **anyone, including external contributors**.
+
+This complements the always-on review bar in
+[../dev/invariants.md](../dev/invariants.md): the invariants say *what every
+change must respect*; an RFC says *why this particular change is worth making and
+how*.
+
+> **Two tracks, don't conflate them.** This `docs/rfcs/` directory is the
+> **public contribution** track (anyone authors; maintainers accept). The
+> maintainer-internal RFCs under `docs/dev/rfc-00N-*.md` are a separate,
+> team-owned track for in-flight internal work. If you're an outside
+> contributor, you're in the right place here.
+
+## When you need one
+
+- **RFC required:** new query/schema/CLI/HTTP surface; on-disk or wire-format
+  changes; a new substrate dependency; anything the deny-list in
+  [../dev/invariants.md](../dev/invariants.md) flags; anything irreversible
+  ("reversibility shapes evidence demand").
+- **RFC not required:** bug fixes for an `accepted` issue, and the trivial
+  fast-lane (typos, docs, deps) — see [../../CONTRIBUTING.md](../../CONTRIBUTING.md).
+
+If you're unsure, start a [Discussion](../../../discussions); a maintainer will
+tell you whether it needs an RFC.
+
+## Lifecycle
+
+```
+Discussion (incubate, get rough consensus)
+      │ graduate
+      ▼
+RFC pull request  →  adds docs/rfcs/NNNN-title.md  (Status: Proposed)
+      │
+maintainer review  ──▶  changes requested / declined (PR closed, with rationale)
+      │
+      ▼
+merged  ==  Accepted   (the merged file is the durable decision record)
+      │
+      ▼
+Implementation PR(s)  reference the accepted RFC
+```
+
+- **Author:** anyone. **Acceptance:** a maintainer decision, performed by
+  merging the RFC PR. Declining is closing it with rationale.
+- The merged RFC *is* the accepted record — there is no separate sign-off step.
+- Later reversals don't edit history: supersede with a new RFC that links back
+  and flip the old one's `Status` to `Superseded`.
+
+## Numbering & naming
+
+- File: `docs/rfcs/NNNN-kebab-title.md`, where `NNNN` is the next free
+  zero-padded integer (`0001`, `0002`, …). `0000-template.md` is reserved.
+- Pick the number when you open the PR; if it collides with another in-flight
+  RFC, the second to merge bumps theirs.
+
+## Status values
+
+`Proposed` (open PR) · `Accepted` (merged) · `Declined` (closed) ·
+`Superseded by NNNN` · `Implemented` (set once the work lands, optional).
+
+Copy [0000-template.md](0000-template.md) to start.
diff --git a/docs/user/audit.md b/docs/user/audit.md
index e8abe5b..ab028ac 100644
--- a/docs/user/audit.md
+++ b/docs/user/audit.md
@@ -4,4 +4,4 @@
 - `_as` variants of every write API let callers override the actor: `mutate_as`, `ingest_as`, `branch_merge_as`, `apply_schema_as`, etc.
 - Actor IDs are persisted on `GraphCommit.actor_id` with split storage in `_graph_commit_actors.lance` (the commit graph is split into `_graph_commits.lance` for the linkage and `_graph_commit_actors.lance` for the actor map).
 - HTTP server uses the bearer-token actor automatically; CLI uses the local user / explicit env (no implicit actor).
-- Pre-v0.4.0 graphs also stored actor IDs on `RunRecord.actor_id` in `_graph_runs.lance` / `_graph_run_actors.lance`. The Run state machine was removed in MR-771; those files are inert post-v0.4.0 and reclaimed by MR-770's production sweep.
+- Pre-v0.4.0 graphs also stored actor IDs on `RunRecord.actor_id` in `_graph_runs.lance` / `_graph_run_actors.lance`. The Run state machine was removed in MR-771; those files are inert post-v0.4.0. The v2→v3 manifest migration sweeps any stale `__run__*` branches on first write-open (MR-770); the inert dataset bytes remain until a `delete_prefix` primitive lands.
diff --git a/docs/user/branches-commits.md b/docs/user/branches-commits.md
index c1894f9..0565186 100644
--- a/docs/user/branches-commits.md
+++ b/docs/user/branches-commits.md
@@ -9,8 +9,8 @@ Lance supports branching at the dataset level: a branch is a named lineage of ve
 OmniGraph builds *graph branches* on top by branching every sub-table coherently:
 
 - `branch_create(name)` / `branch_create_from(target, name)` — disallowed name `main`; fails if branch exists; ensures the schema-apply lock is idle. Atomic and authority-first like `branch_delete`: it flips the `__manifest` branch (authority), then creates the derived commit-graph branch, force-dropping any orphaned commit-graph ref left by an incomplete prior delete (the manifest branch is fresh, so a same-named commit-graph branch is provably a zombie). If commit-graph creation fails, the manifest branch is rolled back so the name never half-exists.
-- `branch_list()` — returns public branches, **filters internal** `__run__…` and `__schema_apply_lock__` prefixes.
-- `branch_delete(name)` — refuses if there are descendants or active runs on the branch. The manifest is the single authority for branch existence: deletion flips the `__manifest` branch ref first (one atomic op), after which the branch is gone from every snapshot. The owned per-table forks and the commit-graph branch are derived state, reclaimed best-effort with `force_delete_branch` after the flip. A failure during that reclaim (transient object-store error) does not fail the call or block the authority flip; the leftover forks are unreachable orphans that the [`cleanup`](maintenance.md) reconciler converges. One consequence: if a delete's best-effort reclaim fails, reusing that branch name before the next `cleanup` surfaces a clear error pointing at `cleanup` (the stale fork would otherwise collide on first write).
+- `branch_list()` — returns public branches, **filters the internal** `__schema_apply_lock__` branch.
+- `branch_delete(name)` — refuses if there are descendants on the branch, or if it is the current branch. The manifest is the single authority for branch existence: deletion flips the `__manifest` branch ref first (one atomic op), after which the branch is gone from every snapshot. The owned per-table forks and the commit-graph branch are derived state, reclaimed best-effort with `force_delete_branch` after the flip. A failure during that reclaim (transient object-store error) does not fail the call or block the authority flip; the leftover forks are unreachable orphans that the [`cleanup`](maintenance.md) reconciler converges. One consequence: if a delete's best-effort reclaim fails, reusing that branch name before the next `cleanup` surfaces a clear error pointing at `cleanup` (the stale fork would otherwise collide on first write).
 - **Lazy forking**: a branch only forks a sub-table when that sub-table is first mutated on it. Pure-read branches share fragments with their source. A fork collision is classified by the manifest authority, not by Lance branch versions: if the live manifest already records the fork on the active branch, a concurrent first-write won and the caller gets a retryable "refresh and retry"; if the manifest does not, a physical branch there is an orphan and the caller is pointed at `cleanup`.
 - `sync_branch(branch)` — re-binds the in-memory handle to the latest head of the branch.
 
@@ -51,10 +51,10 @@ Notes:
 
 ## L2 — Internal system branches
 
-Filtered from `branch_list()` but visible to internals:
+Internal or legacy branch refs:
 
-- `__schema_apply_lock__` — serializes schema migrations.
-- `__run__<run-id>` — legacy from the pre-v0.4.0 Run state machine (removed in MR-771). The branch-name guard predicate `is_internal_run_branch` is kept as defense-in-depth so users cannot create a branch matching the legacy prefix; the filter will be removed once production legacy branches are swept (MR-770).
+- `__schema_apply_lock__` — serializes schema migrations; filtered from `branch_list()` but visible to internals.
+- `__run__<run-id>` — legacy from the pre-v0.4.0 Run state machine (removed in MR-771). These are swept off `__manifest` on the first read-write open by the v2→v3 internal-schema migration (MR-770), and `__run__*` is no longer a reserved name. Known limitation: a pre-v0.4.0 graph opened **read-only** still surfaces any stale `__run__*` branch in `branch_list()` until its first read-write open (the migration is write-path-only, like all manifest migrations).
 
 ## L2 — Recovery audit trail
 
diff --git a/docs/user/constants.md b/docs/user/constants.md
index 8f13555..210155e 100644
--- a/docs/user/constants.md
+++ b/docs/user/constants.md
@@ -4,11 +4,11 @@
 |---|---|---|
 | `MANIFEST_DIR` | `__manifest` | `db/manifest/layout.rs` |
 | Commit graph dir | `_graph_commits.lance` | `db/commit_graph.rs` |
-| Run registry dir (legacy, removed MR-771) | `_graph_runs.lance` | inert post-v0.4.0; reclaimed by MR-770 |
-| Run branch prefix (legacy, removed MR-771) | `__run__` | filtered by `is_internal_run_branch` defense-in-depth |
+| Run registry dir (legacy, removed MR-771) | `_graph_runs.lance` | inert post-v0.4.0; bytes remain until a `delete_prefix` primitive lands |
+| Run branch prefix (legacy, removed MR-771/MR-770) | `__run__` | swept off `__manifest` by the v2→v3 migration; no longer a reserved name |
 | Schema apply lock | `__schema_apply_lock__` | `db/mod.rs` |
 | Manifest publisher retry budget | `PUBLISHER_RETRY_BUDGET = 5` | `db/manifest/publisher.rs` |
-| Internal manifest schema version | `INTERNAL_MANIFEST_SCHEMA_VERSION = 2` | `db/manifest/migrations.rs` |
+| Internal manifest schema version | `INTERNAL_MANIFEST_SCHEMA_VERSION = 3` | `db/manifest/migrations.rs` |
 | Merge stage batch | `MERGE_STAGE_BATCH_ROWS = 8192` | `exec/merge.rs` |
 | Maintenance concurrency | `OMNIGRAPH_MAINTENANCE_CONCURRENCY=8` | `db/omnigraph/optimize.rs` |
 | Lance blob compaction support | `LANCE_SUPPORTS_BLOB_COMPACTION = false` | `db/omnigraph/optimize.rs` |
diff --git a/docs/user/storage.md b/docs/user/storage.md
index c22d4d6..d1c52b5 100644
--- a/docs/user/storage.md
+++ b/docs/user/storage.md
@@ -22,7 +22,7 @@ OmniGraph is **not** a single Lance dataset; it is a *graph* of datasets coordin
   - `edges/{fnv1a64-hex(edge_type_name)}` — one Lance dataset per edge type
   - `__manifest/` — the catalog of all sub-tables and their published versions
   - `_graph_commits.lance` / `_graph_commit_actors.lance` — the commit graph and its actor map
-  - (legacy `_graph_runs.lance` / `_graph_run_actors.lance` from pre-v0.4.0 graphs are inert; the run state machine was removed in MR-771 and these files are cleaned up via MR-770's production sweep)
+  - (legacy `_graph_runs.lance` / `_graph_run_actors.lance` from pre-v0.4.0 graphs are inert; the run state machine was removed in MR-771. The v2→v3 manifest migration sweeps stale `__run__*` branches on first write-open; the inert dataset bytes themselves remain until a `delete_prefix` storage primitive lands)
 - **Manifest row schema** (`object_id, object_type, location, metadata, base_objects, table_key, table_version, table_branch, row_count`):
   - `object_type` ∈ `table | table_version | table_tombstone`
   - `table_key` ∈ `node:<TypeName> | edge:<EdgeName>`
@@ -47,6 +47,7 @@ Adding a new on-disk shape change is one constant bump (`INTERNAL_MANIFEST_SCHEM
 |---|---|
 | v1 (implicit, pre-stamp) | `__manifest.object_id` had no PK annotation; publisher had no row-level CAS protection. |
 | v2 | `__manifest.object_id` carries `lance-schema:unenforced-primary-key=true`; row-level CAS engaged. Stamped as `omnigraph:internal_schema_version=2`. |
+| v3 | One-time sweep of legacy `__run__*` staging branches (pre-v0.4.0 Run state machine, removed MR-771) off `__manifest`. Runs at `Omnigraph::open(ReadWrite)` and on publish. Stamped as `omnigraph:internal_schema_version=3`. |
 
 ## On-disk layout
 
@@ -91,7 +92,7 @@ flowchart TB
 - **Graph root** is one directory (or S3 prefix). Everything below is part of one OmniGraph graph.
 - **`__manifest/`** is a Lance dataset whose rows describe which sub-table version is published at which graph-branch. Reading a snapshot starts here.
 - **`nodes/`** and **`edges/`** are sibling directories holding one Lance dataset per declared type. Names are `fnv1a64-hex` of the type name to keep paths fixed-length and case-safe.
-- **`_graph_commits.lance`** is an L2 dataset that records the graph-level commit DAG, with a paired `_graph_commit_actors.lance` for the actor map. (Pre-v0.4.0 graphs also have inert `_graph_runs.lance` / `_graph_run_actors.lance` from the removed Run state machine; MR-770 sweeps these in production.)
+- **`_graph_commits.lance`** is an L2 dataset that records the graph-level commit DAG, with a paired `_graph_commit_actors.lance` for the actor map. (Pre-v0.4.0 graphs also have inert `_graph_runs.lance` / `_graph_run_actors.lance` from the removed Run state machine; the v2→v3 migration sweeps their stale `__run__*` branches, and the dataset bytes are reclaimed once `delete_prefix` lands.)
 - **`_graph_commit_recoveries.lance`** — one row per recovery sweep action. Joined to `_graph_commits.lance` by `graph_commit_id`; the linked commit row carries `actor_id=omnigraph:recovery`. Operators correlate recoveries with the original mutations they rolled forward / back via this join. See `crates/omnigraph/src/db/recovery_audit.rs`.
 - **`__recovery/{ulid}.json`** — transient sidecar files written by the four migrated writers (`MutationStaging::finalize`, `schema_apply`, `branch_merge`, `ensure_indices`) before Phase B begins, deleted after Phase C succeeds. A sidecar persisting after process exit means the writer crashed in the Phase B → Phase C window; the next `Omnigraph::open` recovery sweep processes it. Steady-state directory is empty. See `crates/omnigraph/src/db/manifest/recovery.rs`.
 - **`_refs/branches/{name}.json`** is graph-level branch metadata — pointers from a branch name to the manifest version it heads.
diff --git a/scripts/check-agents-md.sh b/scripts/check-agents-md.sh
index abc6469..02a177a 100755
--- a/scripts/check-agents-md.sh
+++ b/scripts/check-agents-md.sh
@@ -34,10 +34,15 @@ PY
 canonical=()
 while IFS= read -r line; do
   canonical+=("$line")
-done < <(find docs -type f -name '*.md' ! -path 'docs/releases/*' ! -path 'docs/internal/*' | sort)
+done < <(find docs -type f -name '*.md' ! -path 'docs/releases/*' ! -path 'docs/internal/*' ! -path 'docs/rfcs/*' | sort)
 if [[ -d docs/releases ]]; then
   canonical+=("docs/releases/")
 fi
+# RFCs are a growing collection (like releases): represent the directory, not
+# every per-RFC file. The dir must be linked from an audience index.
+if [[ -d docs/rfcs ]]; then
+  canonical+=("docs/rfcs/")
+fi
 
 linked=()
 for index_file in "${index_files[@]}"; do