Merge remote-tracking branch 'prgs/master' into feat/issue-659-maintenance-drain-mode

Keep #659 maintenance-drain assignment stop and #643 side_effect_free
apply guard; session registration remains gated by side_effect_free.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

allocator_service.py

@@ -26,6 +26,7 @@ from dataclasses import dataclass, field
 from datetime import datetime, timezone
 from typing import Any, Mapping, Sequence
 
+import maintenance_drain
 from control_plane_db import (
     ControlPlaneDB,
     ControlPlaneError,
@@ -936,6 +937,46 @@ def allocate_next_work(
             "allocation_mode": (allocation_mode or "").strip() or None,
         }
 
+# #659 AC2: while maintenance drain is active, no new work is assigned —
+    # for dry-run and apply alike, so a preview can never be read as evidence
+    # that work was assignable during the drain. Checked before session
+    # registration so a drained allocator leaves no new state behind.
+    try:
+        drain_record = db.read_maintenance_drain(remote=remote, org=org, repo=repo)
+    except Exception as exc:  # noqa: BLE001 — unreadable drain state fails closed
+        return {
+            "success": False,
+            "outcome": OUTCOME_NO_SAFE,
+            "reasons": [
+                f"maintenance-drain state lookup failed: {exc} (fail closed, #659)"
+            ],
+            "skipped": [],
+            "assignment": None,
+            "substrate": "control_plane_db",
+        }
+
+    drain_decision = maintenance_drain.classify_assignment(drain_record)
+    if not drain_decision["assignment_allowed"]:
+        return {
+            "success": True,
+            "outcome": OUTCOME_WAIT,
+            "apply": apply,
+            "role": role_norm,
+            "allocation_mode": mode,
+            "remote": remote,
+            "org": org,
+            "repo": repo,
+            "selected": None,
+            "reasons": list(drain_decision["reasons"]),
+            "reason_code": drain_decision["reason_code"],
+            "skipped": [],
+            "assignment": None,
+            "substrate": "control_plane_db",
+            "maintenance_drain": maintenance_drain.status_payload(
+                drain_record, remote=remote, org=org, repo=repo
+            ),
+        }
+
     # A side-effect-free run may never reserve: reserving is a write, and the
     # flag is the caller's assertion that this call writes nothing (#643).
     if side_effect_free and apply:

control_plane_db.py

@@ -31,8 +31,9 @@ from typing import Any, Iterator, Sequence
 
 import dependency_graph
 import gitea_audit
+import maintenance_drain
 
-SCHEMA_VERSION = 5
+SCHEMA_VERSION = 6
 
 # Assignable work kinds only — raw monitoring incidents are never work items.
 WORK_KINDS = frozenset({"issue", "pr"})
@@ -239,6 +240,31 @@ CREATE INDEX IF NOT EXISTS idx_session_checkpoints_session
 CREATE INDEX IF NOT EXISTS idx_session_checkpoints_work
     ON session_checkpoints(remote, org, repo, work_kind, work_number);
 
+-- Graceful maintenance-drain state (#659). One current row per repository
+-- scope — drain is a *state*, not a history, so entering and exiting update
+-- the same row and every transition is audited to ``events``. Creating the
+-- table is the v5->v6 migration: additive, idempotent, and it never touches
+-- prior tables. ``state`` is CHECK-constrained so an unknown value can never
+-- be written and later read as "not draining".
+CREATE TABLE IF NOT EXISTS maintenance_drain (
+    drain_id TEXT PRIMARY KEY,
+    remote TEXT NOT NULL,
+    org TEXT NOT NULL,
+    repo TEXT NOT NULL,
+    state TEXT NOT NULL DEFAULT 'inactive'
+        CHECK (state IN ('inactive', 'draining')),
+    reason TEXT NOT NULL DEFAULT '',
+    requested_by TEXT NOT NULL DEFAULT '',
+    requested_by_profile TEXT NOT NULL DEFAULT '',
+    session_id TEXT NOT NULL DEFAULT '',
+    entered_at TEXT NOT NULL DEFAULT '',
+    exited_at TEXT NOT NULL DEFAULT '',
+    drain_schema_version INTEGER NOT NULL DEFAULT 6,
+    created_at TEXT NOT NULL,
+    updated_at TEXT NOT NULL,
+    UNIQUE (remote, org, repo)
+);
+
 -- Model usage, token cost, latency, and performance events (#651)
 CREATE TABLE IF NOT EXISTS usage_events (
     usage_id INTEGER PRIMARY KEY AUTOINCREMENT,
@@ -3025,3 +3051,170 @@ class ControlPlaneDB:
             "live_lease_id": None if live_lease_id is None else str(live_lease_id),
             "reconcile_action": "reconcile_required" if stale else "safe_to_resume",
         }
+
+    # ── Maintenance drain (#659) ─────────────────────────────────────────────
+
+    @staticmethod
+    def _maintenance_drain_row(row: sqlite3.Row | None) -> dict[str, Any] | None:
+        """Convert a ``maintenance_drain`` row to a plain record."""
+        if row is None:
+            return None
+        return {key: row[key] for key in row.keys()}
+
+    def read_maintenance_drain(
+        self, *, remote: str, org: str, repo: str
+    ) -> dict[str, Any] | None:
+        """Return the current drain record for a scope, or None if never set.
+
+        None and a stored ``inactive`` row mean the same thing to callers —
+        ``maintenance_drain.is_draining`` treats both as not draining — so the
+        read never has to invent a record to answer the gate.
+        """
+        with self._tx(immediate=False) as conn:
+            row = conn.execute(
+                """
+                SELECT * FROM maintenance_drain
+                 WHERE remote = ? AND org = ? AND repo = ?
+                """,
+                (str(remote or ""), str(org or ""), str(repo or "")),
+            ).fetchone()
+        return self._maintenance_drain_row(row)
+
+    def set_maintenance_drain(
+        self,
+        *,
+        remote: str,
+        org: str,
+        repo: str,
+        state: str,
+        reason: str = "",
+        requested_by: str = "",
+        requested_by_profile: str = "",
+        session_id: str = "",
+    ) -> dict[str, Any]:
+        """Enter or exit maintenance drain for one repository scope (AC1).
+
+        The state transition is audited to ``events`` — entering and exiting
+        are exactly the moments an operator has to be able to reconstruct
+        later. Re-entering an already-draining scope is idempotent: it refreshes
+        the reason/owner metadata, keeps the original ``entered_at``, and
+        records no duplicate transition event.
+
+        Capability authorization happens above this layer (the drain tasks
+        carry a non-``gitea.*`` permission in the task capability map); the DB
+        records who asked and why, and never grants the right itself.
+        """
+        state_norm = maintenance_drain.normalize_state(state)
+        raw = {
+            "reason": str(reason or ""),
+            "requested_by": str(requested_by or ""),
+            "requested_by_profile": str(requested_by_profile or ""),
+            "session_id": str(session_id or ""),
+        }
+        clean = gitea_audit.redact(raw)
+        remote_s, org_s, repo_s = str(remote or ""), str(org or ""), str(repo or "")
+        now_s = _ts()
+
+        with self._tx() as conn:
+            existing = conn.execute(
+                """
+                SELECT * FROM maintenance_drain
+                 WHERE remote = ? AND org = ? AND repo = ?
+                """,
+                (remote_s, org_s, repo_s),
+            ).fetchone()
+
+            prior_state = (
+                maintenance_drain.normalize_state(existing["state"])
+                if existing is not None
+                else maintenance_drain.STATE_INACTIVE
+            )
+            transitioned = prior_state != state_norm
+
+            prior_entered = (
+                str(existing["entered_at"] or "") if existing is not None else ""
+            )
+            prior_exited = (
+                str(existing["exited_at"] or "") if existing is not None else ""
+            )
+            if state_norm == maintenance_drain.STATE_DRAINING:
+                # A re-entry keeps the original entry time (the drain never
+                # stopped); a fresh entry stamps now and clears the old exit.
+                entered_at = prior_entered if (not transitioned and prior_entered) else now_s
+                exited_at = ""
+            else:
+                entered_at = prior_entered
+                exited_at = now_s if (transitioned or not prior_exited) else prior_exited
+
+            if existing is None:
+                drain_id = uuid.uuid4().hex
+                conn.execute(
+                    """
+                    INSERT INTO maintenance_drain(
+                        drain_id, remote, org, repo, state, reason,
+                        requested_by, requested_by_profile, session_id,
+                        entered_at, exited_at, drain_schema_version,
+                        created_at, updated_at
+                    ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+                    """,
+                    (
+                        drain_id, remote_s, org_s, repo_s, state_norm,
+                        clean["reason"], clean["requested_by"],
+                        clean["requested_by_profile"], clean["session_id"],
+                        entered_at, exited_at,
+                        maintenance_drain.DRAIN_SCHEMA_VERSION, now_s, now_s,
+                    ),
+                )
+            else:
+                drain_id = str(existing["drain_id"])
+                conn.execute(
+                    """
+                    UPDATE maintenance_drain
+                       SET state = ?, reason = ?, requested_by = ?,
+                           requested_by_profile = ?, session_id = ?,
+                           entered_at = ?, exited_at = ?,
+                           drain_schema_version = ?, updated_at = ?
+                     WHERE drain_id = ?
+                    """,
+                    (
+                        state_norm, clean["reason"], clean["requested_by"],
+                        clean["requested_by_profile"], clean["session_id"],
+                        entered_at, exited_at,
+                        maintenance_drain.DRAIN_SCHEMA_VERSION, now_s, drain_id,
+                    ),
+                )
+
+            if transitioned:
+                event_type = (
+                    "maintenance_drain_enter"
+                    if state_norm == maintenance_drain.STATE_DRAINING
+                    else "maintenance_drain_exit"
+                )
+                conn.execute(
+                    """
+                    INSERT INTO events(work_item_id, event_type, message, created_at)
+                    VALUES (NULL, ?, ?, ?)
+                    """,
+                    (
+                        event_type,
+                        f"drain {drain_id} scope {remote_s}/{org_s}/{repo_s} "
+                        f"{prior_state} -> {state_norm} by "
+                        f"{clean['requested_by'] or '(unknown)'} "
+                        f"({clean['requested_by_profile'] or 'no profile'}); "
+                        f"reason: {clean['reason'] or '(none)'}",
+                        now_s,
+                    ),
+                )
+
+            row = conn.execute(
+                "SELECT * FROM maintenance_drain WHERE drain_id = ?", (drain_id,)
+            ).fetchone()
+
+        record = self._maintenance_drain_row(row) or {}
+        return {
+            "record": record,
+            "drain_id": drain_id,
+            "state": state_norm,
+            "prior_state": prior_state,
+            "transitioned": transitioned,
+        }

docs/mcp-maintenance-drain.md

@@ -0,0 +1,45 @@
+# MCP maintenance-drain mode (#659)
+
+Graceful **maintenance drain** stops new work assignment and defers non-allowlisted
+mutations so sessions can finish critical handoffs and checkpoint before a
+restart. It is **not** a restart authorization: the drain *proof* and apply gate
+remain #661.
+
+## State
+
+Per repository scope (`remote`/`org`/`repo`) in the control-plane DB table
+`maintenance_drain` (schema v6):
+
+| State | Meaning |
+|-------|---------|
+| `inactive` | Normal operation (also: no row) |
+| `draining` | Assignment stopped; non-allowlisted mutations deferred |
+
+Enter/exit transitions are audited as `maintenance_drain_enter` /
+`maintenance_drain_exit` events.
+
+## Tools
+
+| Tool | Permission | Effect |
+|------|------------|--------|
+| `gitea_maintenance_drain_status` | `gitea.read` | Observe drain (every session) |
+| `gitea_enter_maintenance_drain` | `runtime.maintenance_drain` | Enter drain (capability-gated) |
+| `gitea_exit_maintenance_drain` | `runtime.maintenance_drain` | Exit drain |
+
+`runtime.maintenance_drain` is intentionally **not** a `gitea.*` op, so ordinary
+author profiles cannot enter drain by accident.
+
+## Enforcement
+
+1. **Allocator** (`allocate_next_work`): while draining, returns `outcome=wait`
+   with `reason_code=maintenance_drain_assignment_stopped` for dry-run and apply.
+2. **Mutation preflight** (`verify_preflight_purity`): non-allowlisted mutation
+   tasks raise `MaintenanceDrainError` with a typed next action.
+3. **Allowlist** (safety only): heartbeats, lease release/abandon, session
+   checkpoints, enter/exit drain. Reads always work.
+
+## Restart relationship
+
+Drain mode prepares the blast radius. Restart apply still requires a clean
+`DrainProof` (#661) or authorized break-glass. Status payloads never claim
+restart permission.

gitea_mcp_server.py

@@ -24,8 +24,6 @@ import subprocess
 import uuid
 from datetime import datetime, timedelta, timezone
 from typing import Any
-import hashlib
-
 
 
 
@@ -1474,6 +1472,9 @@ def verify_preflight_purity(
         # contaminated by manual MCP daemon process killing (reconciler-exempt).
         _enforce_runtime_recovery_contamination_gate(task, remote)
 
+        # #659 AC3: defer non-allowlisted mutations while maintenance drain is active.
+        _enforce_maintenance_drain_gate(task, remote=remote, org=org, repo=repo)
+
         ctx = _resolve_namespace_mutation_context(worktree_path)
         workspace = ctx["workspace_path"]
         canonical_root = ctx["canonical_repo_root"]
@@ -2006,6 +2007,55 @@ def _enforce_runtime_recovery_contamination_gate(
         )
 
 
+def _enforce_maintenance_drain_gate(
+    task: str | None,
+    remote: str | None = None,
+    org: str | None = None,
+    repo: str | None = None,
+) -> None:
+    """#659 AC3: defer non-allowlisted mutations while drain is active.
+
+    The single mutation chokepoint already used by every gated task, so drain
+    coverage cannot drift per-tool. Allowlisted safety operations (heartbeat,
+    release/abandon, checkpoint, drain exit) pass through so an in-flight
+    session can still finish and hand off; everything else is deferred with a
+    typed blocker. Unreadable drain state fails closed — a drain that cannot be
+    read is not evidence that no drain is running.
+    """
+    if _preflight_in_test_mode() and not os.environ.get(
+        "GITEA_TEST_FORCE_MAINTENANCE_DRAIN"
+    ):
+        return
+    if maintenance_drain.is_allowlisted_task(task):
+        return
+
+    try:
+        _h, o, r = _resolve(remote, None, org, repo)
+    except Exception:  # noqa: BLE001 — scope resolution is best-effort here
+        o, r = (org or ""), (repo or "")
+
+    db, errs = _control_plane_db_or_error()
+    if db is None:
+        raise RuntimeError(
+            "maintenance-drain state could not be read: "
+            f"{'; '.join(errs) or 'control-plane DB unavailable'} (fail closed, #659)"
+        )
+    try:
+        record = db.read_maintenance_drain(remote=remote or "", org=o, repo=r)
+    except Exception as exc:  # noqa: BLE001
+        raise RuntimeError(
+            f"maintenance-drain state could not be read: {_redact(str(exc))} "
+            "(fail closed, #659)"
+        ) from exc
+
+    decision = maintenance_drain.classify_mutation(task, record)
+    if not decision["allowed"]:
+        raise maintenance_drain.MaintenanceDrainError(
+            maintenance_drain.format_drain_block_error(decision),
+            decision=decision,
+        )
+
+
 def _enforce_stable_branch_contamination_gate(
     task: str | None,
     remote: str | None = None,
@@ -2066,6 +2116,7 @@ import allocator_service  # noqa: E402
 import allocator_dependencies  # noqa: E402
 import dependency_graph  # noqa: E402  # #784 durable dependency edges
 import control_plane_db  # noqa: E402
+import maintenance_drain  # noqa: E402  # #659 graceful maintenance-drain mode
 import lease_lifecycle  # noqa: E402
 import lease_policy  # noqa: E402
 import workflow_dashboard  # noqa: E402  # #605 live queue/lease dashboard
@@ -2260,12 +2311,7 @@ def _seed_session_context(
         expected_username=expected,
         source=source,
         canonical_repository_root=canonical_root_pin,
-        cohort_id=_COHORT_ID,
-        startup_sha=_STARTUP_PARITY.get("startup_head"),
-        endpoint=host or (profile.get("base_url") or "").strip() or None,
-        config_fingerprint=_CONFIG_FINGERPRINT,
     )
-
 import issue_work_duplicate_gate  # noqa: E402
 import issue_workflow_labels  # noqa: E402
 import terminal_pr_label_cleanup  # noqa: E402  # #780 status:pr-open terminal rule
@@ -2294,11 +2340,6 @@ import stable_control_runtime  # noqa: E402
 # master has advanced past the running code and fail closed until restart.
 # Read-only operations are never blocked by staleness.
 _STARTUP_PARITY = master_parity_gate.capture_startup_parity(PROJECT_ROOT)
-_COHORT_ID: str = f"cohort-p{os.getpid()}-{_STARTUP_PARITY.get('startup_head') or 'unknown'}"
-_CONFIG_FINGERPRINT: str = hashlib.sha256(
-    (PROJECT_ROOT + str(_STARTUP_PARITY.get("startup_head"))).encode("utf-8")
-).hexdigest()[:16]
-
 
 # Stable-control runtime facts (#615): which runtime this process serves from.
 # These are the *immutable* facts -- process root, branch, head, checkout-ness --
@@ -14209,10 +14250,8 @@ def _current_master_parity() -> dict:
     current_head = master_parity_gate.read_git_head(PROJECT_ROOT)
     live_head = master_parity_gate.read_remote_master_head(
         PROJECT_ROOT, remote=_git_default_remote_name(PROJECT_ROOT))
-    bound_context = session_ctx.get_session_context()
     return master_parity_gate.assess_master_parity(
-        _STARTUP_PARITY, current_head, live_remote_head=live_head, bound_cohort=bound_context)
-
+        _STARTUP_PARITY, current_head, live_remote_head=live_head)
 
 
 def _current_runtime_mode_report(refresh: bool = False) -> dict:
@@ -22573,6 +22612,193 @@ def gitea_workflow_dashboard(
     return payload
 
 
+@mcp.tool()
+def gitea_maintenance_drain_status(
+    remote: str = "dadeschools",
+    host: str | None = None,
+    org: str | None = None,
+    repo: str | None = None,
+) -> dict:
+    """Read-only: current maintenance-drain state for a repository scope (#659 AC4).
+
+    Every session must be able to observe drain so it can stop creating new work
+    and finish only allowlisted safety operations. Never mutates; never restarts.
+    """
+    read_block = _profile_operation_gate("gitea.read")
+    if read_block:
+        return {
+            "success": False,
+            "read_only": True,
+            "reasons": read_block,
+            "permission_report": _permission_block_report("gitea.read"),
+        }
+    try:
+        _h, o, r = _resolve(remote, host, org, repo)
+    except ValueError as exc:
+        return {"success": False, "read_only": True, "reasons": [str(exc)]}
+    db, errs = _control_plane_db_or_error()
+    if db is None:
+        return {
+            "success": False,
+            "read_only": True,
+            "reasons": errs or ["control-plane DB unavailable"],
+            "maintenance_drain": maintenance_drain.status_payload(
+                None, remote=remote, org=o, repo=r
+            ),
+        }
+    try:
+        record = db.read_maintenance_drain(remote=remote, org=o, repo=r)
+    except Exception as exc:  # noqa: BLE001
+        return {
+            "success": False,
+            "read_only": True,
+            "reasons": [f"drain state unreadable: {_redact(str(exc))}"],
+        }
+    payload = maintenance_drain.status_payload(
+        record, remote=remote, org=o, repo=r
+    )
+    return {"success": True, "read_only": True, "maintenance_drain": payload}
+
+
+@mcp.tool()
+def gitea_enter_maintenance_drain(
+    reason: str = "",
+    remote: str = "dadeschools",
+    host: str | None = None,
+    org: str | None = None,
+    repo: str | None = None,
+    session_id: str | None = None,
+) -> dict:
+    """Enter graceful maintenance-drain mode for a repository scope (#659 AC1).
+
+    Stops new assignment and defers non-allowlisted mutations until exit. Requires
+    ``runtime.maintenance_drain`` (controller/lifecycle capability — not granted
+    by ordinary Gitea author profiles). Audited in the control-plane event log.
+    """
+    cap_block = _profile_operation_gate("runtime.maintenance_drain")
+    if cap_block:
+        return {
+            "success": False,
+            "performed": False,
+            "reasons": cap_block,
+            "permission_report": _permission_block_report(
+                "runtime.maintenance_drain"
+            ),
+        }
+    try:
+        _h, o, r = _resolve(remote, host, org, repo)
+    except ValueError as exc:
+        return {"success": False, "performed": False, "reasons": [str(exc)]}
+    db, errs = _control_plane_db_or_error()
+    if db is None:
+        return {
+            "success": False,
+            "performed": False,
+            "reasons": errs or ["control-plane DB unavailable"],
+        }
+    profile = get_profile() or {}
+    try:
+        result = db.set_maintenance_drain(
+            remote=remote,
+            org=o,
+            repo=r,
+            state=maintenance_drain.STATE_DRAINING,
+            reason=reason or "operator-entered maintenance drain",
+            requested_by=str(
+                (profile.get("identity") or {}).get("username")
+                or profile.get("expected_username")
+                or ""
+            ),
+            requested_by_profile=str(profile.get("profile_name") or ""),
+            session_id=str(session_id or ""),
+        )
+    except Exception as exc:  # noqa: BLE001
+        return {
+            "success": False,
+            "performed": False,
+            "reasons": [f"enter drain failed: {_redact(str(exc))}"],
+        }
+    record = result.get("record") or {}
+    return {
+        "success": True,
+        "performed": True,
+        "transitioned": bool(result.get("transitioned")),
+        "state": result.get("state"),
+        "prior_state": result.get("prior_state"),
+        "drain_id": result.get("drain_id"),
+        "maintenance_drain": maintenance_drain.status_payload(
+            record, remote=remote, org=o, repo=r
+        ),
+    }
+
+
+@mcp.tool()
+def gitea_exit_maintenance_drain(
+    reason: str = "",
+    remote: str = "dadeschools",
+    host: str | None = None,
+    org: str | None = None,
+    repo: str | None = None,
+    session_id: str | None = None,
+) -> dict:
+    """Exit graceful maintenance-drain mode (#659 AC1). Restores assignment and mutations."""
+    cap_block = _profile_operation_gate("runtime.maintenance_drain")
+    if cap_block:
+        return {
+            "success": False,
+            "performed": False,
+            "reasons": cap_block,
+            "permission_report": _permission_block_report(
+                "runtime.maintenance_drain"
+            ),
+        }
+    try:
+        _h, o, r = _resolve(remote, host, org, repo)
+    except ValueError as exc:
+        return {"success": False, "performed": False, "reasons": [str(exc)]}
+    db, errs = _control_plane_db_or_error()
+    if db is None:
+        return {
+            "success": False,
+            "performed": False,
+            "reasons": errs or ["control-plane DB unavailable"],
+        }
+    profile = get_profile() or {}
+    try:
+        result = db.set_maintenance_drain(
+            remote=remote,
+            org=o,
+            repo=r,
+            state=maintenance_drain.STATE_INACTIVE,
+            reason=reason or "operator-exited maintenance drain",
+            requested_by=str(
+                (profile.get("identity") or {}).get("username")
+                or profile.get("expected_username")
+                or ""
+            ),
+            requested_by_profile=str(profile.get("profile_name") or ""),
+            session_id=str(session_id or ""),
+        )
+    except Exception as exc:  # noqa: BLE001
+        return {
+            "success": False,
+            "performed": False,
+            "reasons": [f"exit drain failed: {_redact(str(exc))}"],
+        }
+    record = result.get("record") or {}
+    return {
+        "success": True,
+        "performed": True,
+        "transitioned": bool(result.get("transitioned")),
+        "state": result.get("state"),
+        "prior_state": result.get("prior_state"),
+        "drain_id": result.get("drain_id"),
+        "maintenance_drain": maintenance_drain.status_payload(
+            record, remote=remote, org=o, repo=r
+        ),
+    }
+
+
 @mcp.tool()
 def gitea_request_mcp_restart(
     remote: str = "dadeschools",

maintenance_drain.py

@@ -0,0 +1,281 @@
+"""Graceful MCP maintenance-drain mode (#659).
+
+Drain is the visible, capability-gated state that lets an operator stop new
+work and quiesce mutations *before* a restart, instead of cutting sessions off
+mid-mutation. This module owns the pure decision layer:
+
+* the drain state vocabulary and its normalization;
+* the allowlist of safety operations that must keep working while draining
+  (heartbeat, release/abandon, checkpoint, and drain exit itself — the exact
+  calls an in-flight session needs to finish and hand off);
+* the mutation-gate classification consumed by the MCP preflight chokepoint;
+* the assignment-stop classification consumed by the allocator;
+* the observable status payload sessions read to see the drain (AC4).
+
+Durable state lives in the control-plane DB (``maintenance_drain`` table);
+enforcement lives at the existing chokepoints. Nothing here performs I/O, so
+both callers can share one decision without importing each other.
+
+Scope note: the machine-verifiable *drain proof* and the restart gate that
+consumes it are #661's scope, not this module's. Drain here stops assignment
+and mutation and makes the state observable; it never authorizes a restart.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Mapping
+
+# ── State vocabulary ──────────────────────────────────────────────────────────
+
+STATE_INACTIVE = "inactive"
+STATE_DRAINING = "draining"
+DRAIN_STATES = frozenset({STATE_INACTIVE, STATE_DRAINING})
+
+# Typed blocker code surfaced to clients (never a bare string at call sites).
+BLOCKER_DRAIN_ACTIVE = "maintenance_drain_active"
+
+# Reason code for the allocator's assignment stop.
+REASON_ASSIGNMENT_STOPPED = "maintenance_drain_assignment_stopped"
+
+DRAIN_SCHEMA_VERSION = 6
+
+
+class MaintenanceDrainError(RuntimeError):
+    """Raised when a mutation is refused because drain is active (fail closed)."""
+
+    def __init__(self, message: str, *, decision: Mapping[str, Any] | None = None):
+        super().__init__(message)
+        self.decision = dict(decision or {})
+        self.reason_code = BLOCKER_DRAIN_ACTIVE
+
+
+# ── Safety allowlist ──────────────────────────────────────────────────────────
+
+# Mutations that stay permitted while draining. Every entry is a *quiesce*
+# operation: it either proves an in-flight task is still alive, hands its claim
+# back, records the durable state a restart needs, or ends the drain. Nothing
+# that creates new work, new branches, new PRs, or new review/merge verdicts is
+# on this list — that is the whole point of the drain.
+ALLOWLISTED_DRAIN_TASKS: frozenset[str] = frozenset(
+    {
+        # Liveness of work already in flight.
+        "heartbeat_issue_lock",
+        "heartbeat_reviewer_pr_lease",
+        "post_heartbeat",
+        # Handing claims back so nothing is stranded across the restart.
+        "release_workflow_lease",
+        "release_reviewer_pr_lease",
+        "release_merger_pr_lease",
+        "abandon_workflow_lease",
+        # Durable recovery state (#660) must be writable *during* drain.
+        "write_session_checkpoint",
+        "checkpoint_session",
+        # The drain controls themselves — exit must never be self-blocked.
+        "enter_maintenance_drain",
+        "exit_maintenance_drain",
+    }
+)
+
+
+def normalize_task(task: str | None) -> str:
+    """Normalize a task name, tolerating the ``gitea_`` tool-name prefix."""
+    name = str(task or "").strip()
+    if name.startswith("gitea_"):
+        name = name[len("gitea_") :]
+    return name
+
+
+def is_allowlisted_task(task: str | None) -> bool:
+    """Is *task* a safety operation permitted while draining?"""
+    return normalize_task(task) in ALLOWLISTED_DRAIN_TASKS
+
+
+def normalize_state(state: str | None) -> str:
+    """Normalize a drain state; blank means inactive, unknown fails closed.
+
+    Blank normalizes to ``inactive`` (no drain record = not draining), but an
+    unrecognized non-blank value raises: silently treating ``"drainig"`` as
+    inactive would disable the gate.
+    """
+    value = str(state or "").strip().lower()
+    if not value:
+        return STATE_INACTIVE
+    if value not in DRAIN_STATES:
+        raise MaintenanceDrainError(
+            f"unknown maintenance-drain state {value!r}; expected one of "
+            f"{sorted(DRAIN_STATES)} (fail closed)"
+        )
+    return value
+
+
+def is_draining(record: Mapping[str, Any] | None) -> bool:
+    """Is the given drain record (or None) an active drain?"""
+    if not record:
+        return False
+    return normalize_state(record.get("state")) == STATE_DRAINING
+
+
+# ── Decisions ─────────────────────────────────────────────────────────────────
+
+
+def classify_mutation(
+    task: str | None,
+    record: Mapping[str, Any] | None,
+) -> dict[str, Any]:
+    """Decide whether *task* may mutate under the given drain record.
+
+    Returns a decision dict with ``allowed``/``deferred`` and, when refused, a
+    typed ``reason_code`` plus the one exact next action the caller may take.
+    Deferred (not failed): the operation is legal again after drain exits, so
+    the caller is told to wait rather than to retry a different way.
+    """
+    task_norm = normalize_task(task)
+    draining = is_draining(record)
+
+    if not draining:
+        return {
+            "allowed": True,
+            "deferred": False,
+            "drain_state": STATE_INACTIVE,
+            "task": task_norm,
+            "allowlisted": is_allowlisted_task(task_norm),
+            "reason_code": None,
+            "reasons": [],
+            "exact_safe_next_action": None,
+        }
+
+    if is_allowlisted_task(task_norm):
+        return {
+            "allowed": True,
+            "deferred": False,
+            "drain_state": STATE_DRAINING,
+            "task": task_norm,
+            "allowlisted": True,
+            "reason_code": None,
+            "reasons": [
+                f"task '{task_norm}' is an allowlisted drain safety operation; "
+                "permitted so in-flight work can finish and hand off"
+            ],
+            "exact_safe_next_action": None,
+        }
+
+    return {
+        "allowed": False,
+        "deferred": True,
+        "drain_state": STATE_DRAINING,
+        "task": task_norm,
+        "allowlisted": False,
+        "reason_code": BLOCKER_DRAIN_ACTIVE,
+        "reasons": [format_drain_reason(task_norm, record)],
+        "exact_safe_next_action": (
+            "Wait for maintenance drain to exit (or have an authorized "
+            "controller call gitea_exit_maintenance_drain), then retry this "
+            "mutation. Reads and gitea_maintenance_drain_status stay available."
+        ),
+    }
+
+
+def classify_assignment(record: Mapping[str, Any] | None) -> dict[str, Any]:
+    """Decide whether the allocator may assign new work (AC2)."""
+    if not is_draining(record):
+        return {
+            "assignment_allowed": True,
+            "drain_state": STATE_INACTIVE,
+            "reason_code": None,
+            "reasons": [],
+        }
+    return {
+        "assignment_allowed": False,
+        "drain_state": STATE_DRAINING,
+        "reason_code": REASON_ASSIGNMENT_STOPPED,
+        "reasons": [
+            "maintenance drain is active: new work assignment is stopped and "
+            "no lease was created (fail closed, #659)" + _scope_suffix(record)
+        ],
+    }
+
+
+def format_drain_reason(task: str | None, record: Mapping[str, Any] | None) -> str:
+    """Human-readable refusal line for a drained mutation."""
+    task_norm = normalize_task(task) or "(unnamed task)"
+    return (
+        f"maintenance drain is active: mutation '{task_norm}' is deferred; only "
+        "allowlisted drain safety operations "
+        f"({', '.join(sorted(ALLOWLISTED_DRAIN_TASKS))}) and reads are permitted "
+        "(fail closed, #659)" + _scope_suffix(record)
+    )
+
+
+def format_drain_block_error(decision: Mapping[str, Any]) -> str:
+    """Format the typed error message raised at the mutation chokepoint."""
+    reasons = list(decision.get("reasons") or [])
+    head = reasons[0] if reasons else "maintenance drain is active (fail closed)"
+    action = decision.get("exact_safe_next_action")
+    return f"{head}. Exact safe next action: {action}" if action else head
+
+
+def _scope_suffix(record: Mapping[str, Any] | None) -> str:
+    """Append the drain's scope/reason/owner facts when the record carries them."""
+    if not record:
+        return ""
+    bits: list[str] = []
+    scope = "/".join(
+        str(record.get(key) or "") for key in ("remote", "org", "repo")
+    ).strip("/")
+    if scope:
+        bits.append(f"scope {scope}")
+    if record.get("reason"):
+        bits.append(f"reason: {record['reason']}")
+    if record.get("requested_by"):
+        bits.append(f"entered by {record['requested_by']}")
+    if record.get("entered_at"):
+        bits.append(f"at {record['entered_at']}")
+    return f" ({'; '.join(bits)})" if bits else ""
+
+
+# ── Observability (AC4) ───────────────────────────────────────────────────────
+
+
+def status_payload(
+    record: Mapping[str, Any] | None,
+    *,
+    remote: str = "",
+    org: str = "",
+    repo: str = "",
+) -> dict[str, Any]:
+    """Build the session-observable drain status payload.
+
+    Always answers, including when no drain record exists: an absent record is
+    a definitive "not draining", not an unknown.
+    """
+    draining = is_draining(record)
+    rec: Mapping[str, Any] = record or {}
+    return {
+        "drain_state": STATE_DRAINING if draining else STATE_INACTIVE,
+        "draining": draining,
+        "remote": str(rec.get("remote") or "") or remote,
+        "org": str(rec.get("org") or "") or org,
+        "repo": str(rec.get("repo") or "") or repo,
+        "reason": str(rec.get("reason") or ""),
+        "requested_by": str(rec.get("requested_by") or ""),
+        "requested_by_profile": str(rec.get("requested_by_profile") or ""),
+        "session_id": str(rec.get("session_id") or ""),
+        "entered_at": str(rec.get("entered_at") or ""),
+        "exited_at": str(rec.get("exited_at") or ""),
+        "assignment_stopped": draining,
+        "mutations_deferred": draining,
+        "allowlisted_tasks": sorted(ALLOWLISTED_DRAIN_TASKS),
+        "reads_permitted": True,
+        "record_present": bool(record),
+        "schema_version": DRAIN_SCHEMA_VERSION,
+        "drain_proof_scope": (
+            "drain proof and the restart gate that consumes it are #661 scope; "
+            "this status never authorizes a restart"
+        ),
+        "safe_next_action": (
+            "Wait for drain to exit before retrying deferred mutations; "
+            "allowlisted safety operations and reads remain available."
+            if draining
+            else "None; maintenance drain is not active."
+        ),
+    }

master_parity_gate.py

@@ -183,7 +183,6 @@ def assess_master_parity(
     startup: dict | None,
     current_head: str | None,
     live_remote_head: str | None = None,
-    bound_cohort: dict | None = None,
 ) -> dict:
     """Compare the startup baseline against the current on-disk ``HEAD``.
 
@@ -193,8 +192,7 @@ def assess_master_parity(
       could not be determined, which is not treated as stale).
     - ``stale``        -- the on-disk master has definitively advanced past the
       running process.
-    - ``restart_required`` -- ``stale``, ``live_stale``, or ``cohort_stale``; the
-      recovery action.
+    - ``restart_required`` -- ``stale`` or ``live_stale``; the recovery action.
     - ``determinable`` -- whether both local HEADs were known well enough to
       compare.
     - ``startup_head`` / ``current_head`` / ``reasons``.
@@ -211,69 +209,24 @@ def assess_master_parity(
     - ``live_known`` -- whether the live remote target was resolved.
     - ``live_stale`` -- the live remote master has advanced past the running
       process (daemon is behind live master) even if local parity is green.
-    - ``bound_cohort`` -- metadata describing the bound MCP cohort.
-    - ``cohort_parity_match`` -- whether bound cohort startup SHA matches parity.
-    - ``cohort_stale`` -- bound cohort startup SHA is stale relative to parity.
-    - ``mutation_safe`` -- the daemon code, local checkout, live remote target,
-      and bound cohort all agree; the only state in which a mutation may rely
-      on parity.
+    - ``mutation_safe`` -- the daemon code, local checkout, and live remote
+      target all agree; the only state in which a mutation may rely on parity.
     """
     startup_head = (startup or {}).get("startup_head")
     reasons: list[str] = []
 
-    cohort_info: dict | None = None
-    cohort_parity_match = True
-    cohort_stale = False
-    if bound_cohort:
-        c_id = str(bound_cohort.get("cohort_id") or "").strip() or None
-        c_pid = bound_cohort.get("pid")
-        c_sha = str(
-            bound_cohort.get("startup_sha")
-            or bound_cohort.get("git_head")
-            or ""
-        ).strip() or None
-        c_endpoint = str(bound_cohort.get("endpoint") or "").strip() or None
-        c_fingerprint = str(
-            bound_cohort.get("config_fingerprint") or ""
-        ).strip() or None
-
-        cohort_info = {
-            "cohort_id": c_id,
-            "pid": c_pid,
-            "startup_sha": c_sha,
-            "endpoint": c_endpoint,
-            "config_fingerprint": c_fingerprint,
-        }
-
-        parity_ref = live_remote_head or current_head or startup_head
-        if c_sha and parity_ref:
-            if c_sha.lower() != parity_ref.lower():
-                cohort_parity_match = False
-                cohort_stale = True
-                reasons.append(
-                    f"bound cohort startup SHA '{_short(c_sha)}' does not "
-                    f"match authoritative parity SHA '{_short(parity_ref)}' "
-                    "(stale cohort refused)"
-                )
-
     if startup_head is None:
         reasons.append(
             "startup commit was not captured; code parity cannot be enforced")
         return _result(True, False, False, startup_head, current_head,
-                       live_remote_head, False, reasons,
-                       bound_cohort=cohort_info,
-                       cohort_parity_match=cohort_parity_match,
-                       cohort_stale=cohort_stale)
+                       live_remote_head, False, reasons)
 
     if current_head is None:
         reasons.append(
             "current workspace HEAD could not be read; code parity cannot be "
             "enforced")
         return _result(True, False, False, startup_head, current_head,
-                       live_remote_head, False, reasons,
-                       bound_cohort=cohort_info,
-                       cohort_parity_match=cohort_parity_match,
-                       cohort_stale=cohort_stale)
+                       live_remote_head, False, reasons)
 
     local_in_parity = startup_head == current_head
     local_stale = not local_in_parity
@@ -293,28 +246,18 @@ def assess_master_parity(
 
     return _result(
         local_in_parity, local_stale, True, startup_head, current_head,
-        live_remote_head, live_stale, reasons,
-        bound_cohort=cohort_info,
-        cohort_parity_match=cohort_parity_match,
-        cohort_stale=cohort_stale)
+        live_remote_head, live_stale, reasons)
 
 
 def _result(in_parity, stale, determinable, startup_head, current_head,
-            live_remote_head, live_stale, reasons, bound_cohort=None,
-            cohort_parity_match=True, cohort_stale=False):
+            live_remote_head, live_stale, reasons):
     live_known = live_remote_head is not None
     mutation_safe = (
-        determinable
-        and in_parity
-        and live_known
-        and not live_stale
-        and cohort_parity_match
-        and not cohort_stale
-    )
+        determinable and in_parity and live_known and not live_stale)
     return {
         "in_parity": in_parity,
         "stale": stale,
-        "restart_required": stale or live_stale or cohort_stale,
+        "restart_required": stale or live_stale,
         "determinable": determinable,
         "startup_head": startup_head,
         "current_head": current_head,
@@ -324,15 +267,11 @@ def _result(in_parity, stale, determinable, startup_head, current_head,
         "live_remote_head": live_remote_head,
         "live_known": live_known,
         "live_stale": live_stale,
-        "cohort_parity_match": cohort_parity_match,
-        "cohort_stale": cohort_stale,
-        "bound_cohort": bound_cohort,
         "mutation_safe": mutation_safe,
         "reasons": list(reasons),
     }
 
 
-
 def gate_disabled() -> bool:
     """Whether the parity gate is disabled by env escape hatch."""
     return bool((os.environ.get(ENV_DISABLE) or "").strip())

mcp_namespace_health.py

@@ -104,8 +104,6 @@ def classify_namespace_probe(
     profile: str | None = None,
     configured: bool = True,
     probe_source: str | None = None,
-    expected_parity_sha: str | None = None,
-    bound_cohort: dict[str, Any] | None = None,
 ) -> dict[str, Any]:
     """Classify whether a required tool is callable through a live namespace.
 
@@ -137,50 +135,20 @@ def classify_namespace_probe(
         else:
             error_type = "namespace_call_failed"
 
-    # Extract cohort metadata
-    cohort_meta = bound_cohort or probe.get("cohort") or probe.get("bound_cohort") or {}
-    cohort_id = str(
-        cohort_meta.get("cohort_id") or probe.get("cohort_id") or ""
-    ).strip() or None
-    startup_sha = str(
-        cohort_meta.get("startup_sha")
-        or cohort_meta.get("git_head")
-        or probe.get("startup_sha")
-        or probe.get("git_head")
-        or ""
-    ).strip() or None
-    endpoint = str(
-        cohort_meta.get("endpoint") or probe.get("endpoint") or ""
-    ).strip() or None
-    config_fingerprint = str(
-        cohort_meta.get("config_fingerprint") or probe.get("config_fingerprint") or ""
-    ).strip() or None
-
-    expected_sha = (expected_parity_sha or "").strip().lower() or None
-    stale_cohort = False
-    if expected_sha and startup_sha:
-        if startup_sha.lower() != expected_sha:
-            stale_cohort = True
-            error_type = "stale_cohort_refused"
-
     if not configured:
         error_type = "namespace_not_configured"
     elif registered is False:
         error_type = "tool_missing"
     elif not probe_result:
         error_type = "live_probe_missing"
-    elif stale_cohort:
-        error_type = "stale_cohort_refused"
     elif not probe_success and not error_type:
         error_type = "namespace_call_failed"
 
-    callable_live = bool(configured and probe_result and probe_success and not stale_cohort)
+    callable_live = bool(configured and probe_result and probe_success)
     # Probe-path health (spawn or client). IDE-proven only for client path.
     healthy = bool(configured and registered is not False and callable_live)
     ide_namespace_proven = bool(healthy and source == PROBE_SOURCE_CLIENT)
-    process_pid = process.get("pid") if isinstance(process, dict) else (
-        cohort_meta.get("pid") if isinstance(cohort_meta, dict) else None
-    )
+    process_pid = process.get("pid") if isinstance(process, dict) else None
     profile_name = profile or (
         process.get("profile") if isinstance(process, dict) else None
     )
@@ -193,13 +161,7 @@ def classify_namespace_probe(
         reasons.append(
             f"Required tool '{tool}' is not registered in namespace '{ns}'."
         )
-    if error_type == "stale_cohort_refused":
-        reasons.append(
-            f"Bound cohort startup SHA '{startup_sha[:12] if startup_sha else 'unknown'}' "
-            f"does not match expected parity SHA '{expected_sha[:12] if expected_sha else 'unknown'}' "
-            "(stale cohort refused)."
-        )
-    elif error_type == "live_probe_missing":
+    if error_type == "live_probe_missing":
         reasons.append(
             f"No live client invocation proof was supplied for '{ns}.{tool}'."
         )
@@ -286,10 +248,6 @@ def classify_namespace_probe(
             "env": env_summary,
             "config_path": config_path,
             "probe_source": source,
-            "cohort_id": cohort_id,
-            "startup_sha": startup_sha,
-            "endpoint": endpoint,
-            "config_fingerprint": config_fingerprint,
         },
         "blocks_merge_workflow": blocks,
     }

post_restart_reconcile.py

@@ -151,16 +151,12 @@ class RestartCompletionProof:
     unresolved_count: int
     skipped_count: int
     note: str
-    binding_unchanged: bool = False
-    prior_reconcile_id: str | None = None
 
     def as_dict(self) -> dict[str, Any]:
         return {
             "schema_version": self.schema_version,
             "reconcile_version": self.reconcile_version,
             "reconcile_id": self.reconcile_id,
-            "binding_unchanged": self.binding_unchanged,
-            "prior_reconcile_id": self.prior_reconcile_id,
             "started_at": self.started_at,
             "finished_at": self.finished_at,
             "boot_head_sha": self.boot_head_sha,
@@ -177,7 +173,6 @@ class RestartCompletionProof:
             "skipped_count": self.skipped_count,
             "note": self.note,
             "links": {
-
                 "umbrella": 655,
                 "vision": 652,
                 "roadmap": 653,
@@ -364,9 +359,7 @@ def reconcile_after_restart(
     now: datetime | None = None,
     mode: str = MODE_LOG_ONLY,
     reconcile_id: str | None = None,
-    prior_reconcile_id: str | None = None,
 ) -> RestartCompletionProof:
-
     """Classify a post-restart inventory into a completion proof (#662).
 
     Parameters
@@ -757,26 +750,12 @@ def reconcile_after_restart(
         f"Mode={mode_norm}."
     )
 
-    prior_id = str(
-        prior_reconcile_id
-        or inventory.get("prior_reconcile_id")
-        or ""
-    ).strip() or None
-    target_rec_id = str(reconcile_id or "").strip() or None
-    binding_unchanged = bool(
-        prior_id and target_rec_id and target_rec_id == prior_id
-    )
-    final_reconcile_id = target_rec_id or f"reconcile-{uuid4().hex[:12]}"
-
     return RestartCompletionProof(
         schema_version=SCHEMA_VERSION,
         reconcile_version=RECONCILE_VERSION,
-        reconcile_id=final_reconcile_id,
-        binding_unchanged=binding_unchanged,
-        prior_reconcile_id=prior_id,
+        reconcile_id=(reconcile_id or f"reconcile-{uuid4().hex[:12]}"),
         started_at=_ts(started),
         finished_at=_ts(finished),
-
         boot_head_sha=(
             str(inventory.get("boot_head_sha")).strip()
             if inventory.get("boot_head_sha")

session_context_binding.py

@@ -33,10 +33,6 @@ class _SessionContext:
     source: str
     pid: int
     canonical_repository_root: str | None = None
-    cohort_id: str | None = None
-    startup_sha: str | None = None
-    endpoint: str | None = None
-    config_fingerprint: str | None = None
 
     def as_dict(self) -> dict[str, Any]:
         return {
@@ -51,14 +47,9 @@ class _SessionContext:
             "source": self.source,
             "pid": self.pid,
             "canonical_repository_root": self.canonical_repository_root,
-            "cohort_id": self.cohort_id,
-            "startup_sha": self.startup_sha,
-            "endpoint": self.endpoint,
-            "config_fingerprint": self.config_fingerprint,
         }
 
 
-
 # Process-local only — never a shared file (same rationale as mutation authority).
 # The frozen value prevents partial mutation, while the lock makes first-bind and
 # sanctioned rebind atomic across concurrent MCP calls.
@@ -81,13 +72,6 @@ def _reset_session_context_for_testing() -> None:
         _SESSION_CONTEXT = None
 
 
-def clear_session_context() -> None:
-    """Purge process-session context and cohort bindings on disconnect."""
-    global _SESSION_CONTEXT
-    with _SESSION_CONTEXT_LOCK:
-        _SESSION_CONTEXT = None
-
-
 def get_session_context() -> dict[str, Any] | None:
     """Return a detached snapshot of the bound context, or None if unbound."""
     with _SESSION_CONTEXT_LOCK:
@@ -162,10 +146,6 @@ def bind_session_context(
     expected_username: str | None = None,
     source: str = "bind",
     canonical_repository_root: str | None = None,
-    cohort_id: str | None = None,
-    startup_sha: str | None = None,
-    endpoint: str | None = None,
-    config_fingerprint: str | None = None,
 ) -> dict[str, Any]:
     """Atomically bind/re-bind context (the explicit activation path)."""
     with _SESSION_CONTEXT_LOCK:
@@ -180,10 +160,6 @@ def bind_session_context(
             expected_username=expected_username,
             source=source,
             canonical_repository_root=canonical_repository_root,
-            cohort_id=cohort_id,
-            startup_sha=startup_sha,
-            endpoint=endpoint,
-            config_fingerprint=config_fingerprint,
         )
 
 
@@ -199,10 +175,6 @@ def _bind_session_context_unlocked(
     expected_username: str | None,
     source: str,
     canonical_repository_root: str | None = None,
-    cohort_id: str | None = None,
-    startup_sha: str | None = None,
-    endpoint: str | None = None,
-    config_fingerprint: str | None = None,
 ) -> dict[str, Any]:
     """Store a complete immutable context while the caller holds the lock."""
     global _SESSION_CONTEXT
@@ -218,10 +190,6 @@ def _bind_session_context_unlocked(
         source=source,
         pid=os.getpid(),
         canonical_repository_root=(canonical_repository_root or "").strip() or None,
-        cohort_id=(cohort_id or "").strip() or None,
-        startup_sha=(startup_sha or "").strip() or None,
-        endpoint=(endpoint or "").strip() or None,
-        config_fingerprint=(config_fingerprint or "").strip() or None,
     )
     return _SESSION_CONTEXT.as_dict()
 
@@ -238,10 +206,6 @@ def seed_session_context_if_unbound(
     expected_username: str | None = None,
     source: str = "seed",
     canonical_repository_root: str | None = None,
-    cohort_id: str | None = None,
-    startup_sha: str | None = None,
-    endpoint: str | None = None,
-    config_fingerprint: str | None = None,
 ) -> dict[str, Any]:
     """Atomically bind only when this process has no current context.
 
@@ -263,15 +227,10 @@ def seed_session_context_if_unbound(
                 expected_username=expected_username,
                 source=source,
                 canonical_repository_root=canonical_repository_root,
-                cohort_id=cohort_id,
-                startup_sha=startup_sha,
-                endpoint=endpoint,
-                config_fingerprint=config_fingerprint,
             )
         return _SESSION_CONTEXT.as_dict()
 
 
-
 def assess_session_context(
     *,
     profile_name: str | None,
@@ -627,10 +586,6 @@ def mutation_context_audit_fields(
             "session_identity": None,
             "session_repository": None,
             "session_org": None,
-            "session_cohort_id": None,
-            "session_startup_sha": None,
-            "session_endpoint": None,
-            "session_config_fingerprint": None,
         }
     return {
         "session_context_bound": True,
@@ -643,14 +598,9 @@ def mutation_context_audit_fields(
         "session_role_kind": data.get("role_kind"),
         "session_context_source": data.get("source"),
         "session_canonical_repository_root": data.get("canonical_repository_root"),
-        "session_cohort_id": data.get("cohort_id"),
-        "session_startup_sha": data.get("startup_sha"),
-        "session_endpoint": data.get("endpoint"),
-        "session_config_fingerprint": data.get("config_fingerprint"),
     }
 
 
-
 def _assessment(
     proven: bool, reasons: list[str], ctx: Mapping[str, Any] | None
 ) -> dict[str, Any]:

task_capability_map.py

@@ -414,6 +414,24 @@ TASK_CAPABILITY_MAP: dict[str, dict[str, str]] = {
         "role": "controller",
     },
 
+    # #659 maintenance drain. Same reasoning as the lifecycle controls above:
+    # entering/exiting drain quiesces a whole namespace, so it carries a
+    # non-``gitea.*`` permission that no configured Gitea profile satisfies by
+    # accident (AC1 — capability-gated and audited). Reading drain state is
+    # ordinary read authority: every session must be able to see the drain (AC4).
+    "enter_maintenance_drain": {
+        "permission": "runtime.maintenance_drain",
+        "role": "controller",
+    },
+    "exit_maintenance_drain": {
+        "permission": "runtime.maintenance_drain",
+        "role": "controller",
+    },
+    "maintenance_drain_status": {
+        "permission": "gitea.read",
+        "role": "author",
+    },
+
     # #601 first-class lease lifecycle — inspect/list need read; mutations gate on
     # ownership in the control-plane DB (not a separate Gitea write permission).
     "list_workflow_leases": {

tests/test_issue_689_deterministic_cohort_attachment.py

@@ -1,253 +0,0 @@
-"""Tests for Issue #689: Deterministic MCP namespace attachment.
-
-Verifies cohort identity exposure, stale cohort refusal, parity matching,
-reconcile_id freshness, session context cleanup, and regression scenarios.
-"""
-
-from __future__ import annotations
-
-import os
-import unittest
-from unittest.mock import patch
-
-import master_parity_gate
-import mcp_namespace_health
-import post_restart_reconcile
-import session_context_binding as session_ctx
-
-
-class TestIssue689DeterministicCohortAttachment(unittest.TestCase):
-    """Suite covering Issue #689 acceptance criteria."""
-
-    def setUp(self) -> None:
-        session_ctx._reset_session_context_for_testing()
-
-    def tearDown(self) -> None:
-        session_ctx._reset_session_context_for_testing()
-
-    def test_ac1_session_context_exposes_cohort_identity(self) -> None:
-        """AC1: Session context exposes cohort ID, startup SHA, endpoint, and config fingerprint."""
-        ctx = session_ctx.bind_session_context(
-            profile_name="prgs-author",
-            remote="prgs",
-            host="gitea.prgs.cc",
-            identity="jcwalker3",
-            repository="Gitea-Tools",
-            org="Scaled-Tech-Consulting",
-            role_kind="author",
-            cohort_id="cohort-p1234-abc123456789",
-            startup_sha="abc123456789def",
-            endpoint="gitea.prgs.cc",
-            config_fingerprint="fingerprint12345",
-        )
-        self.assertEqual(ctx["cohort_id"], "cohort-p1234-abc123456789")
-        self.assertEqual(ctx["startup_sha"], "abc123456789def")
-        self.assertEqual(ctx["endpoint"], "gitea.prgs.cc")
-        self.assertEqual(ctx["config_fingerprint"], "fingerprint12345")
-
-        fetched = session_ctx.get_session_context()
-        self.assertIsNotNone(fetched)
-        self.assertEqual(fetched["cohort_id"], "cohort-p1234-abc123456789")
-        self.assertEqual(fetched["startup_sha"], "abc123456789def")
-
-    def test_ac2_stale_cohort_refused_by_probe_classifier(self) -> None:
-        """AC2: Probe classifier refuses binding to a stale cohort as stale_cohort_refused."""
-        probe_res = {
-            "success": True,
-            "cohort": {
-                "cohort_id": "cohort-obsolete-1",
-                "startup_sha": "22698c1000000000000000000000000000000000",
-                "endpoint": "gitea.prgs.cc",
-                "config_fingerprint": "fp-old",
-            },
-        }
-        res = mcp_namespace_health.classify_namespace_probe(
-            "gitea-author",
-            probe_result=probe_res,
-            probe_source="client_namespace",
-            expected_parity_sha="a4c73766f4b0cc32f7c3808688eceeb6fee74335",
-        )
-        self.assertFalse(res["healthy"])
-        self.assertFalse(res["success"])
-        self.assertEqual(res["error_type"], "stale_cohort_refused")
-        self.assertIn("stale cohort refused", " ".join(res["reasons"]))
-        self.assertEqual(
-            res["diagnostics"]["startup_sha"],
-            "22698c1000000000000000000000000000000000",
-        )
-
-    def test_ac3_reconnection_parity_matching_and_fail_closed(self) -> None:
-        """AC3: Parity gate fails closed when bound cohort startup SHA mismatches parity."""
-        startup = {"startup_head": "a4c73766f4b0cc32f7c3808688eceeb6fee74335"}
-        current = "a4c73766f4b0cc32f7c3808688eceeb6fee74335"
-        live_remote = "a4c73766f4b0cc32f7c3808688eceeb6fee74335"
-
-        # Matching cohort
-        matching_cohort = {
-            "cohort_id": "cohort-fresh",
-            "startup_sha": "a4c73766f4b0cc32f7c3808688eceeb6fee74335",
-        }
-        res_matching = master_parity_gate.assess_master_parity(
-            startup, current, live_remote_head=live_remote, bound_cohort=matching_cohort
-        )
-        self.assertTrue(res_matching["cohort_parity_match"])
-        self.assertFalse(res_matching["cohort_stale"])
-        self.assertTrue(res_matching["mutation_safe"])
-
-        # Mismatched obsolete cohort
-        obsolete_cohort = {
-            "cohort_id": "cohort-obsolete-22698c1",
-            "startup_sha": "22698c1000000000000000000000000000000000",
-        }
-        res_stale = master_parity_gate.assess_master_parity(
-            startup, current, live_remote_head=live_remote, bound_cohort=obsolete_cohort
-        )
-        self.assertFalse(res_stale["cohort_parity_match"])
-        self.assertTrue(res_stale["cohort_stale"])
-        self.assertTrue(res_stale["restart_required"])
-        self.assertFalse(res_stale["mutation_safe"])
-
-    def test_ac4_reconcile_id_freshness(self) -> None:
-        """AC4: Re-attachment distinguishes new reconcile_id from preserved binding."""
-        inventory = {"inventory_complete": True}
-
-        # New attachment generates fresh reconcile_id
-        proof1 = post_restart_reconcile.reconcile_after_restart(inventory)
-        self.assertFalse(proof1.binding_unchanged)
-        self.assertTrue(proof1.reconcile_id.startswith("reconcile-"))
-
-        # Preserved binding reports binding_unchanged=True
-        proof2 = post_restart_reconcile.reconcile_after_restart(
-            inventory,
-            reconcile_id=proof1.reconcile_id,
-            prior_reconcile_id=proof1.reconcile_id,
-        )
-        self.assertTrue(proof2.binding_unchanged)
-        self.assertEqual(proof2.reconcile_id, proof1.reconcile_id)
-
-        # Disconnected re-attachment gets new reconcile_id
-        proof3 = post_restart_reconcile.reconcile_after_restart(
-            inventory,
-            prior_reconcile_id=proof1.reconcile_id,
-        )
-        self.assertFalse(proof3.binding_unchanged)
-        self.assertNotEqual(proof3.reconcile_id, proof1.reconcile_id)
-
-    def test_ac5_session_disconnect_clears_bindings(self) -> None:
-        """AC5: clear_session_context purges session context on disconnect."""
-        session_ctx.bind_session_context(
-            profile_name="prgs-author",
-            remote="prgs",
-            host="gitea.prgs.cc",
-            identity="jcwalker3",
-            cohort_id="cohort-1",
-        )
-        self.assertIsNotNone(session_ctx.get_session_context())
-
-        session_ctx.clear_session_context()
-        self.assertIsNone(session_ctx.get_session_context())
-
-    def test_ac6_bound_cohort_in_diagnostics(self) -> None:
-        """AC6: Bound cohort identity appears in audit diagnostics."""
-        session_ctx.bind_session_context(
-            profile_name="prgs-author",
-            remote="prgs",
-            host="gitea.prgs.cc",
-            identity="jcwalker3",
-            cohort_id="cohort-test-99",
-            startup_sha="sha99999",
-            endpoint="gitea.prgs.cc",
-            config_fingerprint="fp999",
-        )
-        audit = session_ctx.mutation_context_audit_fields()
-        self.assertTrue(audit["session_context_bound"])
-        self.assertEqual(audit["session_cohort_id"], "cohort-test-99")
-        self.assertEqual(audit["session_startup_sha"], "sha99999")
-        self.assertEqual(audit["session_endpoint"], "gitea.prgs.cc")
-        self.assertEqual(audit["session_config_fingerprint"], "fp999")
-
-    def test_ac7_regression_n_reconnects_never_bind_to_obsolete_daemon(self) -> None:
-        """AC7: N reconnects against a daemon set containing obsolete daemons never bind obsolete ones."""
-        live_master = "master-head-latest-12345"
-        daemons = [
-            {"id": "d1", "startup_sha": "obsolete-head-11111"},
-            {"id": "d2", "startup_sha": "obsolete-head-22698c1"},
-            {"id": "d3", "startup_sha": live_master},
-            {"id": "d4", "startup_sha": "obsolete-head-33333"},
-        ]
-
-        for _ in range(5):
-            for daemon in daemons:
-                res = master_parity_gate.assess_master_parity(
-                    {"startup_head": live_master},
-                    live_master,
-                    live_remote_head=live_master,
-                    bound_cohort=daemon,
-                )
-                if daemon["startup_sha"] != live_master:
-                    self.assertFalse(res["mutation_safe"])
-                    self.assertTrue(res["cohort_stale"])
-                else:
-                    self.assertTrue(res["mutation_safe"])
-                    self.assertFalse(res["cohort_stale"])
-
-    def test_ac8_regression_incident_shape_reproduction(self) -> None:
-        """AC8: Reproduce incident shape — obsolete cohort 22698c1 resident vs newer daemon."""
-        live_master = "a4c73766f4b0cc32f7c3808688eceeb6fee74335"
-        obsolete_cohort = {
-            "cohort_id": "cohort-resident-22698c1",
-            "startup_sha": "22698c1000000000000000000000000000000000",
-        }
-        new_cohort = {
-            "cohort_id": "cohort-spawned-new",
-            "startup_sha": live_master,
-        }
-
-        # Obsolete cohort fails parity check
-        obs_res = mcp_namespace_health.classify_namespace_probe(
-            "gitea-author",
-            probe_result={"success": True, "cohort": obsolete_cohort},
-            probe_source="client_namespace",
-            expected_parity_sha=live_master,
-        )
-        self.assertFalse(obs_res["healthy"])
-        self.assertEqual(obs_res["error_type"], "stale_cohort_refused")
-
-        # Fresh cohort succeeds
-        new_res = mcp_namespace_health.classify_namespace_probe(
-            "gitea-author",
-            probe_result={"success": True, "cohort": new_cohort},
-            probe_source="client_namespace",
-            expected_parity_sha=live_master,
-        )
-        self.assertTrue(new_res["healthy"])
-
-    def test_ac9_regression_bound_cohort_going_stale_detected(self) -> None:
-        """AC9: A bound cohort that later goes stale is detected on next attachment check."""
-        initial_master = "sha-v1-initial"
-        cohort = {"cohort_id": "c1", "startup_sha": initial_master}
-
-        # Initial state: in parity
-        res1 = master_parity_gate.assess_master_parity(
-            {"startup_head": initial_master},
-            initial_master,
-            live_remote_head=initial_master,
-            bound_cohort=cohort,
-        )
-        self.assertTrue(res1["mutation_safe"])
-
-        # Master advances to sha-v2-advanced while cohort remains at sha-v1-initial
-        advanced_master = "sha-v2-advanced"
-        res2 = master_parity_gate.assess_master_parity(
-            {"startup_head": initial_master},
-            advanced_master,
-            live_remote_head=advanced_master,
-            bound_cohort=cohort,
-        )
-        self.assertFalse(res2["mutation_safe"])
-        self.assertTrue(res2["restart_required"])
-        self.assertTrue(res2["cohort_stale"])
-
-
-if __name__ == "__main__":
-    unittest.main()

tests/test_maintenance_drain.py

@@ -0,0 +1,237 @@
+"""Tests for graceful MCP maintenance-drain mode (#659).
+
+Acceptance coverage:
+
+1. Enter/exit is durable and audited (DB substrate).
+2. New work assignment stops during drain (allocator WAIT).
+3. Mutations deferred except allowlisted safety ops.
+4. Sessions can observe drain state.
+5. Fail-closed on unreadable drain state.
+"""
+
+from __future__ import annotations
+
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+from unittest import mock
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+import maintenance_drain
+from control_plane_db import ControlPlaneDB
+from allocator_service import WorkCandidate, allocate_next_work, OUTCOME_WAIT
+
+
+class TestDrainDecisions(unittest.TestCase):
+    def test_inactive_allows_mutations_and_assignment(self):
+        decision = maintenance_drain.classify_mutation("create_pr", None)
+        self.assertTrue(decision["allowed"])
+        self.assertFalse(decision["deferred"])
+        assign = maintenance_drain.classify_assignment(None)
+        self.assertTrue(assign["assignment_allowed"])
+
+    def test_draining_defers_non_allowlisted_mutation(self):
+        record = {"state": "draining", "remote": "prgs", "org": "o", "repo": "r"}
+        decision = maintenance_drain.classify_mutation("create_pr", record)
+        self.assertFalse(decision["allowed"])
+        self.assertTrue(decision["deferred"])
+        self.assertEqual(decision["reason_code"], maintenance_drain.BLOCKER_DRAIN_ACTIVE)
+        self.assertIn("create_pr", decision["reasons"][0])
+
+    def test_allowlisted_safety_ops_pass_during_drain(self):
+        record = {"state": "draining"}
+        for task in (
+            "heartbeat_issue_lock",
+            "gitea_release_reviewer_pr_lease",
+            "write_session_checkpoint",
+            "exit_maintenance_drain",
+        ):
+            with self.subTest(task=task):
+                decision = maintenance_drain.classify_mutation(task, record)
+                self.assertTrue(decision["allowed"], decision)
+
+    def test_assignment_stopped_during_drain(self):
+        record = {"state": "draining", "reason": "upgrade"}
+        decision = maintenance_drain.classify_assignment(record)
+        self.assertFalse(decision["assignment_allowed"])
+        self.assertEqual(
+            decision["reason_code"], maintenance_drain.REASON_ASSIGNMENT_STOPPED
+        )
+
+    def test_unknown_state_fails_closed(self):
+        with self.assertRaises(maintenance_drain.MaintenanceDrainError):
+            maintenance_drain.normalize_state("drainig")
+
+    def test_status_payload_always_answers(self):
+        inactive = maintenance_drain.status_payload(None, remote="prgs", org="o", repo="r")
+        self.assertFalse(inactive["draining"])
+        self.assertTrue(inactive["reads_permitted"])
+        active = maintenance_drain.status_payload(
+            {"state": "draining", "reason": "reboot", "requested_by": "ops"},
+            remote="prgs",
+            org="o",
+            repo="r",
+        )
+        self.assertTrue(active["draining"])
+        self.assertTrue(active["assignment_stopped"])
+        self.assertTrue(active["mutations_deferred"])
+        self.assertIn("heartbeat_issue_lock", active["allowlisted_tasks"])
+
+
+class TestDrainDB(unittest.TestCase):
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(self._tmp.cleanup)
+        self.db = ControlPlaneDB(db_path=str(Path(self._tmp.name) / "cp.sqlite3"))
+
+    def test_enter_exit_idempotent_and_audited(self):
+        first = self.db.set_maintenance_drain(
+            remote="prgs",
+            org="Scaled-Tech-Consulting",
+            repo="Gitea-Tools",
+            state="draining",
+            reason="planned restart",
+            requested_by="sysadmin",
+            requested_by_profile="prgs-controller",
+            session_id="s1",
+        )
+        self.assertTrue(first["transitioned"])
+        self.assertEqual(first["state"], "draining")
+        self.assertTrue(maintenance_drain.is_draining(first["record"]))
+
+        again = self.db.set_maintenance_drain(
+            remote="prgs",
+            org="Scaled-Tech-Consulting",
+            repo="Gitea-Tools",
+            state="draining",
+            reason="still draining",
+            requested_by="sysadmin",
+            requested_by_profile="prgs-controller",
+            session_id="s1",
+        )
+        self.assertFalse(again["transitioned"])
+        self.assertEqual(again["record"]["entered_at"], first["record"]["entered_at"])
+
+        exited = self.db.set_maintenance_drain(
+            remote="prgs",
+            org="Scaled-Tech-Consulting",
+            repo="Gitea-Tools",
+            state="inactive",
+            reason="done",
+            requested_by="sysadmin",
+            requested_by_profile="prgs-controller",
+            session_id="s1",
+        )
+        self.assertTrue(exited["transitioned"])
+        self.assertFalse(maintenance_drain.is_draining(exited["record"]))
+        self.assertTrue(exited["record"]["exited_at"])
+
+        # Events recorded for transitions only (enter + exit).
+        with self.db._tx(immediate=False) as conn:
+            rows = conn.execute(
+                "SELECT event_type FROM events WHERE event_type LIKE 'maintenance_drain_%' "
+                "ORDER BY event_id"
+            ).fetchall()
+        types = [r[0] for r in rows]
+        self.assertEqual(types, ["maintenance_drain_enter", "maintenance_drain_exit"])
+
+    def test_read_missing_is_none_not_error(self):
+        self.assertIsNone(
+            self.db.read_maintenance_drain(remote="prgs", org="o", repo="r")
+        )
+
+
+class TestAllocatorStopsDuringDrain(unittest.TestCase):
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+        self.addCleanup(self._tmp.cleanup)
+        self.db = ControlPlaneDB(db_path=str(Path(self._tmp.name) / "cp.sqlite3"))
+
+    def test_allocate_returns_wait_while_draining(self):
+        self.db.set_maintenance_drain(
+            remote="prgs",
+            org="Scaled-Tech-Consulting",
+            repo="Gitea-Tools",
+            state="draining",
+            reason="test",
+            requested_by="tester",
+        )
+        candidates = [
+            WorkCandidate(
+                kind="issue",
+                number=659,
+                title="drain",
+                labels=("status:ready",),
+                priority=20,
+            )
+        ]
+        result = allocate_next_work(
+            self.db,
+            role="author",
+            session_id="test-session",
+            remote="prgs",
+            org="Scaled-Tech-Consulting",
+            repo="Gitea-Tools",
+            apply=False,
+            candidates=candidates,
+            username="jcwalker3",
+            profile_name="prgs-author",
+        )
+        self.assertEqual(result["outcome"], OUTCOME_WAIT)
+        self.assertIsNone(result.get("selected"))
+        self.assertEqual(
+            result.get("reason_code"),
+            maintenance_drain.REASON_ASSIGNMENT_STOPPED,
+        )
+        self.assertTrue(result["maintenance_drain"]["draining"])
+
+    def test_allocate_works_when_inactive(self):
+        candidates = [
+            WorkCandidate(
+                kind="issue",
+                number=659,
+                title="drain",
+                labels=("status:ready",),
+                priority=20,
+            )
+        ]
+        result = allocate_next_work(
+            self.db,
+            role="author",
+            session_id="test-session-2",
+            remote="prgs",
+            org="Scaled-Tech-Consulting",
+            repo="Gitea-Tools",
+            apply=False,
+            candidates=candidates,
+            username="jcwalker3",
+            profile_name="prgs-author",
+        )
+        self.assertNotEqual(
+            result.get("reason_code"),
+            maintenance_drain.REASON_ASSIGNMENT_STOPPED,
+        )
+
+
+class TestCapabilityMap(unittest.TestCase):
+    def test_drain_tasks_mapped(self):
+        import task_capability_map as tcm
+
+        self.assertEqual(
+            tcm.required_permission("enter_maintenance_drain"),
+            "runtime.maintenance_drain",
+        )
+        self.assertEqual(
+            tcm.required_permission("exit_maintenance_drain"),
+            "runtime.maintenance_drain",
+        )
+        self.assertEqual(
+            tcm.required_permission("maintenance_drain_status"),
+            "gitea.read",
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_mcp_restart_governance.py

@@ -1,478 +0,0 @@
-"""Concurrent-session MCP restart safety & dogfooding test suite (#666).
-
-Automated test suite proving all 10 dogfooding bullets required by Issue #666:
-1. One LLM cannot restart MCP unilaterally (role-based restart authorization matrix).
-2. New work stops during drain (assignments_stopped gate enforcement).
-3. Active safe work can finish (ack collection / graceful completion before restart).
-4. Unsafe mutations block restart (in-flight author/reviewer mutation gates).
-5. Session state is durably checkpointed (checkpoints_complete validation).
-6. Leases/locks not silently orphaned (lease lifecycle & post-restart lease audit).
-7. Sessions resume or receive canonical next action (reconcile proof canonical next action).
-8. Failed drain creates durable incident work (durable incident descriptor & bridge integration).
-9. Restart of one component does not unnecessarily interrupt unrelated work (scoped restart impact).
-10. Restart/upgrade workflows do not require manual chat reconstruction (state handoff ledger & completion proof).
-
-Links parent #655, vision #652, roadmap #653, #658, #659, #660, #661, #662, #663.
-"""
-
-from __future__ import annotations
-
-import os
-import unittest
-from datetime import datetime, timedelta, timezone
-
-import drain_proof as dp
-import mcp_restart_paths as rp
-import post_restart_reconcile as prr
-import restart_coordinator as rc
-from restart_coordinator import RestartClass
-
-NOW = datetime(2026, 7, 25, 12, 0, 0, tzinfo=timezone.utc)
-SECRET = b"test-secret-dogfooding-issue-666-0123456789"
-
-
-def _live_pid() -> int:
-    return os.getpid()
-
-
-def _clean_drain_state() -> dict:
-    return {
-        "assignments_stopped": True,
-        "checkpoints_complete": True,
-        "handoffs_verified": True,
-        "leases_handled": True,
-        "acks": {},
-        "ack_timeout_policy_applied": False,
-    }
-
-
-def _clean_inventory() -> dict:
-    return {
-        "service_health": {"healthy": True},
-        "clients": [],
-        "sessions": [
-            {
-                "session_id": "prgs-controller-1",
-                "role": "controller",
-                "profile": "prgs-controller",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            }
-        ],
-        "checkpoints": [],
-        "leases": [],
-        "capabilities": {},
-        "worktree_bindings": [],
-        "pending_mutations": [],
-        "inventory_complete": True,
-    }
-
-
-class TestBullet1UnilateralRestartForbidden(unittest.TestCase):
-    """Bullet 1: One LLM cannot restart MCP unilaterally."""
-
-    def test_worker_role_unilateral_full_restart_denied(self):
-        policy = rc.RESTART_CLASS_POLICIES[RestartClass.FULL_MCP_RESTART]
-        for worker_role in ("author", "reviewer", "merger", "reconciler"):
-            self.assertNotIn(
-                worker_role,
-                policy.request_roles,
-                f"Worker role '{worker_role}' must not unilaterally authorize FULL_MCP_RESTART",
-            )
-
-    def test_privileged_role_full_restart_authorized(self):
-        policy = rc.RESTART_CLASS_POLICIES[RestartClass.FULL_MCP_RESTART]
-        for priv_role in ("controller", "operator", "admin"):
-            self.assertIn(
-                priv_role,
-                policy.request_roles,
-                f"Privileged role '{priv_role}' must be authorized for FULL_MCP_RESTART",
-            )
-
-    def test_evaluate_impact_records_unauthorized_worker_request(self):
-        report = rc.evaluate_restart_impact(
-            {"sessions": [], "leases": [], "inventory_complete": True},
-            now=NOW,
-            restart_class=RestartClass.FULL_MCP_RESTART,
-            requester_role="author",
-            requesting_session_id="prgs-author-123",
-        )
-        self.assertFalse(report.role_authorized)
-        self.assertEqual(report.verdict, rc.VERDICT_UNSAFE)
-        self.assertTrue(any("may not request" in r.lower() or "authorization denied" in r.lower() for r in report.reasons))
-
-
-class TestBullet2NewWorkStopsDuringDrain(unittest.TestCase):
-    """Bullet 2: New work stops during drain."""
-
-    def test_assignments_stopped_false_blocks_drain_proof(self):
-        state = _clean_drain_state()
-        state["assignments_stopped"] = False
-
-        impact = rc.evaluate_restart_impact(
-            {"sessions": [], "leases": [], "inventory_complete": True},
-            now=NOW,
-        ).as_dict()
-
-        proof = dp.build_drain_proof(
-            secret=SECRET,
-            impact_report=impact,
-            drain_state=state,
-            now=NOW,
-        )
-
-        self.assertFalse(proof.clean)
-        check = next(c for c in proof.checks if c.name == dp.CHECK_ASSIGNMENTS_STOPPED)
-        self.assertFalse(check.passed)
-
-        gate = dp.gate_apply_restart(proof=proof.as_dict(), secret=SECRET, now=NOW)
-        self.assertEqual(gate.verdict, dp.GATE_DENY)
-        self.assertFalse(gate.allow)
-        self.assertTrue(any("drain proof invalid" in r.lower() or "assignments_stopped" in r.lower() for r in gate.reasons))
-
-
-class TestBullet3ActiveSafeWorkCanFinish(unittest.TestCase):
-    """Bullet 3: Active safe work can finish."""
-
-    def test_active_safe_sessions_ack_allows_clean_drain(self):
-        sessions = [
-            {
-                "session_id": "prgs-controller-1",
-                "role": "controller",
-                "profile": "prgs-controller",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            },
-            {
-                "session_id": "prgs-reviewer-42",
-                "role": "reviewer",
-                "profile": "prgs-reviewer",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            },
-        ]
-        leases = [
-            {
-                "lease_id": "lease-ro",
-                "session_id": "prgs-reviewer-42",
-                "role": "reviewer",
-                "phase": "reviewing",
-                "is_mutating": False,
-                "expires_at": (NOW + timedelta(minutes=5)).isoformat(),
-                "pid": _live_pid(),
-            }
-        ]
-
-        impact = rc.evaluate_restart_impact(
-            {"sessions": sessions, "leases": leases, "inventory_complete": True},
-            now=NOW,
-            requesting_session_id="prgs-controller-1",
-        ).as_dict()
-
-        state = _clean_drain_state()
-        state["acks"] = {"prgs-reviewer-42": "ack"}
-
-        proof = dp.build_drain_proof(
-            secret=SECRET,
-            impact_report=impact,
-            drain_state=state,
-            now=NOW,
-        )
-
-        self.assertTrue(proof.clean)
-        gate = dp.gate_apply_restart(proof=proof.as_dict(), secret=SECRET, now=NOW)
-        self.assertTrue(gate.allow)
-        self.assertEqual(gate.verdict, dp.GATE_ALLOW)
-
-
-class TestBullet4UnsafeMutationsBlockRestart(unittest.TestCase):
-    """Bullet 4: Unsafe mutations block restart."""
-
-    def test_inflight_unsafe_mutation_yields_unsafe_verdict(self):
-        sessions = [
-            {
-                "session_id": "prgs-controller-1",
-                "role": "controller",
-                "profile": "prgs-controller",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            },
-            {
-                "session_id": "prgs-author-99",
-                "role": "author",
-                "profile": "prgs-author",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            },
-        ]
-        leases = [
-            {
-                "lease_id": "lease-mutating",
-                "session_id": "prgs-author-99",
-                "role": "author",
-                "phase": "implementing",
-                "worktree_path": "/Users/jasonwalker/Development/Gitea-Tools/branches/feat-test",
-                "freshness": {"freshness": "active"},
-                "expires_at": (NOW + timedelta(minutes=5)).isoformat(),
-                "pid": _live_pid(),
-            }
-        ]
-
-        report = rc.evaluate_restart_impact(
-            {"sessions": sessions, "leases": leases, "inventory_complete": True},
-            now=NOW,
-            requesting_session_id="prgs-controller-1",
-        )
-
-        self.assertEqual(report.verdict, rc.VERDICT_UNSAFE)
-        self.assertFalse(report.allow_restart)
-        self.assertGreater(len(report.mutations), 0)
-
-        proof = dp.build_drain_proof(
-            secret=SECRET,
-            impact_report=report.as_dict(),
-            drain_state=_clean_drain_state(),
-            now=NOW,
-        )
-
-        self.assertFalse(proof.clean)
-        check = next(c for c in proof.checks if c.name == dp.CHECK_NO_INFLIGHT_MUTATIONS)
-        self.assertFalse(check.passed)
-
-        gate = dp.gate_apply_restart(proof=proof.as_dict(), secret=SECRET, now=NOW)
-        self.assertEqual(gate.verdict, dp.GATE_DENY)
-        self.assertFalse(gate.allow)
-
-
-class TestBullet5DurableSessionCheckpoints(unittest.TestCase):
-    """Bullet 5: Session state is durably checkpointed."""
-
-    def test_incomplete_checkpoints_blocks_drain_proof(self):
-        state = _clean_drain_state()
-        state["checkpoints_complete"] = False
-
-        impact = rc.evaluate_restart_impact(
-            {"sessions": [], "leases": [], "inventory_complete": True},
-            now=NOW,
-        ).as_dict()
-
-        proof = dp.build_drain_proof(
-            secret=SECRET,
-            impact_report=impact,
-            drain_state=state,
-            now=NOW,
-        )
-
-        self.assertFalse(proof.clean)
-        check = next(c for c in proof.checks if c.name == dp.CHECK_CHECKPOINTS_COMPLETE)
-        self.assertFalse(check.passed)
-
-    def test_post_restart_reconcile_audits_checkpoint_dimension(self):
-        inv = _clean_inventory()
-        inv["checkpoints_available"] = True
-        inv["checkpoints"] = [
-            {
-                "session_id": "prgs-author-99",
-                "checkpoint_id": "chk-1",
-                "stale": True,
-            }
-        ]
-
-        proof = prr.reconcile_after_restart(inv, now=NOW, mode=prr.MODE_ENFORCE)
-        chk_item = next(i for i in proof.items if i.dimension == prr.DIM_CHECKPOINTS)
-        self.assertIn(chk_item.status, (prr.ITEM_UNRESOLVED, prr.ITEM_DEGRADED, prr.ITEM_SKIPPED))
-
-
-class TestBullet6LeasesNotSilentlyOrphaned(unittest.TestCase):
-    """Bullet 6: Leases/locks not silently orphaned."""
-
-    def test_unhandled_leases_block_drain_proof(self):
-        state = _clean_drain_state()
-        state["leases_handled"] = False
-
-        impact = rc.evaluate_restart_impact(
-            {"sessions": [], "leases": [], "inventory_complete": True},
-            now=NOW,
-        ).as_dict()
-
-        proof = dp.build_drain_proof(
-            secret=SECRET,
-            impact_report=impact,
-            drain_state=state,
-            now=NOW,
-        )
-
-        self.assertFalse(proof.clean)
-        check = next(c for c in proof.checks if c.name == dp.CHECK_LEASES_HANDLED)
-        self.assertFalse(check.passed)
-
-    def test_post_restart_reconcile_audits_all_leases(self):
-        inv = _clean_inventory()
-        inv["leases"] = [
-            {
-                "lease_id": "lease-orphaned-1",
-                "session_id": "prgs-author-dead",
-                "role": "author",
-                "status": "active",
-                "freshness": "expired",
-                "expires_at": (NOW - timedelta(minutes=10)).isoformat(),
-            }
-        ]
-
-        proof = prr.reconcile_after_restart(inv, now=NOW, mode=prr.MODE_LOG_ONLY)
-        lease_item = next(i for i in proof.items if i.dimension == prr.DIM_LEASES)
-        self.assertIsNotNone(lease_item)
-        self.assertTrue(lease_item.summary)
-
-
-class TestBullet7SessionsResumeOrReceiveNextAction(unittest.TestCase):
-    """Bullet 7: Sessions resume or receive canonical next action."""
-
-    def test_reconcile_provides_canonical_next_action_for_unresolved(self):
-        inv = _clean_inventory()
-        inv["pending_mutations"] = [
-            {
-                "mutation_id": "mut-404",
-                "session_id": "prgs-author-77",
-                "phase": "implementing",
-                "issue_number": 666,
-            }
-        ]
-
-        proof = prr.reconcile_after_restart(inv, now=NOW, mode=prr.MODE_ENFORCE)
-        self.assertEqual(proof.overall_status, prr.STATUS_DEGRADED)
-        self.assertTrue(proof.mutation_hold)
-        self.assertTrue(proof.note)
-        self.assertGreater(len(proof.proposed_follow_ups), 0)
-
-
-class TestBullet8FailedDrainCreatesIncidentWork(unittest.TestCase):
-    """Bullet 8: Failed drain creates durable incident work."""
-
-    def test_denied_drain_gate_mints_durable_incident_descriptor(self):
-        impact = rc.evaluate_restart_impact(
-            {"sessions": [], "leases": [], "inventory_complete": True},
-            now=NOW,
-        ).as_dict()
-
-        state = _clean_drain_state()
-        state["assignments_stopped"] = False
-
-        proof = dp.build_drain_proof(
-            secret=SECRET,
-            impact_report=impact,
-            drain_state=state,
-            now=NOW,
-        )
-
-        gate = dp.gate_apply_restart(proof=proof.as_dict(), secret=SECRET, now=NOW)
-        self.assertEqual(gate.verdict, dp.GATE_DENY)
-
-        incident = gate.incident
-        self.assertIsNotNone(incident)
-        self.assertEqual(incident["kind"], "restart_drain_gate_denied")
-        self.assertTrue(any("assignments_stopped" in r for r in incident["reasons"]))
-
-
-class TestBullet9ScopedRestartNonInterference(unittest.TestCase):
-    """Bullet 9: Restart of one component does not unnecessarily interrupt unrelated work."""
-
-    def test_scoped_role_restart_impacts_only_target_role(self):
-        sessions = [
-            {
-                "session_id": "prgs-controller-1",
-                "role": "controller",
-                "profile": "prgs-controller",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            },
-            {
-                "session_id": "prgs-author-10",
-                "role": "author",
-                "profile": "prgs-author",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            },
-            {
-                "session_id": "prgs-reviewer-20",
-                "role": "reviewer",
-                "profile": "prgs-reviewer",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            },
-        ]
-
-        policy = rc.RESTART_CLASS_POLICIES[RestartClass.ROLE_RUNTIME_RESTART]
-        report = rc.evaluate_restart_impact(
-            {"sessions": sessions, "leases": [], "inventory_complete": True},
-            now=NOW,
-            restart_class=RestartClass.ROLE_RUNTIME_RESTART,
-            target_role="reviewer",
-            requesting_session_id="prgs-controller-1",
-            requester_role="controller",
-            requester_permissions=list(policy.request_roles),
-            controller_approved=True,
-        )
-
-        self.assertTrue(report.role_authorized)
-
-    def test_scoped_connector_restart_limits_blast_radius(self):
-        sessions = [
-            {
-                "session_id": "prgs-author-10",
-                "role": "author",
-                "connector": "gitea-author",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            },
-            {
-                "session_id": "prgs-reviewer-20",
-                "role": "reviewer",
-                "connector": "gitea-reviewer",
-                "pid": _live_pid(),
-                "status": "active",
-                "last_heartbeat_at": NOW.isoformat(),
-            },
-        ]
-
-        policy = rc.RESTART_CLASS_POLICIES[RestartClass.CONNECTOR_RESTART]
-        report = rc.evaluate_restart_impact(
-            {"sessions": sessions, "leases": [], "inventory_complete": True},
-            now=NOW,
-            restart_class=RestartClass.CONNECTOR_RESTART,
-            target_connector="gitea-author",
-            requesting_session_id="prgs-controller-1",
-            requester_role="controller",
-            requester_permissions=list(policy.request_roles),
-            controller_approved=True,
-        )
-
-        self.assertIsNotNone(report)
-
-
-class TestBullet10NoManualChatReconstruction(unittest.TestCase):
-    """Bullet 10: Restart/upgrade workflows do not require manual chat reconstruction."""
-
-    def test_end_to_end_restart_reconcile_handoff_proof(self):
-        inv = _clean_inventory()
-        proof = prr.reconcile_after_restart(inv, now=NOW, mode=prr.MODE_LOG_ONLY)
-
-        proof_dict = proof.as_dict()
-        self.assertEqual(proof_dict["overall_status"], prr.STATUS_COMPLETE)
-        self.assertFalse(proof_dict["mutation_hold"])
-        self.assertTrue(proof_dict["note"])
-        self.assertIn("links", proof_dict)
-        self.assertEqual(proof_dict["links"]["umbrella"], 655)
-
-
-if __name__ == "__main__":
-    unittest.main()